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         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  45         struct inet_sock *inet = inet_sk(sk);
  46         struct dccp_sock *dp = dccp_sk(sk);
  47         __be16 orig_sport, orig_dport;
  48         __be32 daddr, nexthop;
  49         struct flowi4 *fl4;
  50         struct rtable *rt;
  51         int err;
  52         struct ip_options_rcu *inet_opt;
  53
  54         dp->dccps_role = DCCP_ROLE_CLIENT;
  55
  56         if (addr_len < sizeof(struct sockaddr_in))
  57                 return -EINVAL;
  58
  59         if (usin->sin_family != AF_INET)
  60                 return -EAFNOSUPPORT;
  61
  62         nexthop = daddr = usin->sin_addr.s_addr;
  63
  64         inet_opt = rcu_dereference_protected(inet->inet_opt,
  65                                              sock_owned_by_user(sk));
  66         if (inet_opt != NULL && inet_opt->opt.srr) {
  67                 if (daddr == 0)
  68                         return -EINVAL;
  69                 nexthop = inet_opt->opt.faddr;
  70         }
  71
  72         orig_sport = inet->inet_sport;
  73         orig_dport = usin->sin_port;
  74         fl4 = &inet->cork.fl.u.ip4;
  75         rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
  76                               RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  77                               IPPROTO_DCCP,
  78                               orig_sport, orig_dport, sk);
  79         if (IS_ERR(rt))
  80                 return PTR_ERR(rt);
  81
  82         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  83                 ip_rt_put(rt);
  84                 return -ENETUNREACH;
  85         }
  86
  87         if (inet_opt == NULL || !inet_opt->opt.srr)
  88                 daddr = fl4->daddr;
  89
  90         if (inet->inet_saddr == 0)
  91                 inet->inet_saddr = fl4->saddr;
  92         sk_rcv_saddr_set(sk, inet->inet_saddr);
  93         inet->inet_dport = usin->sin_port;
  94         sk_daddr_set(sk, daddr);
  95
  96         inet_csk(sk)->icsk_ext_hdr_len = 0;
  97         if (inet_opt)
  98                 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
  99         /*
 100          * Socket identity is still unknown (sport may be zero).
 101          * However we set state to DCCP_REQUESTING and not releasing socket
 102          * lock select source port, enter ourselves into the hash tables and
 103          * complete initialization after this.
 104          */
 105         dccp_set_state(sk, DCCP_REQUESTING);
 106         err = inet_hash_connect(&dccp_death_row, sk);
 107         if (err != 0)
 108                 goto failure;
 109
 110         rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 111                                inet->inet_sport, inet->inet_dport, sk);
 112         if (IS_ERR(rt)) {
 113                 err = PTR_ERR(rt);
 114                 rt = NULL;
 115                 goto failure;
 116         }
 117         /* OK, now commit destination to socket.  */
 118         sk_setup_caps(sk, &rt->dst);
 119
 120         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
 121                                                     inet->inet_daddr,
 122                                                     inet->inet_sport,
 123                                                     inet->inet_dport);
 124         inet->inet_id = dp->dccps_iss ^ jiffies;
 125
 126         err = dccp_connect(sk);
 127         rt = NULL;
 128         if (err != 0)
 129                 goto failure;
 130 out:
 131         return err;
 132 failure:
 133         /*
 134          * This unhashes the socket and releases the local port, if necessary.
 135          */
 136         dccp_set_state(sk, DCCP_CLOSED);
 137         ip_rt_put(rt);
 138         sk->sk_route_caps = 0;
 139         inet->inet_dport = 0;
 140         goto out;
 141 }
 142 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 143
 144 /*
 145  * This routine does path mtu discovery as defined in RFC1191.
 146  */
 147 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 148                                           const struct iphdr *iph,
 149                                           u32 mtu)
 150 {
 151         struct dst_entry *dst;
 152         const struct inet_sock *inet = inet_sk(sk);
 153         const struct dccp_sock *dp = dccp_sk(sk);
 154
 155         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 156          * send out by Linux are always < 576bytes so they should go through
 157          * unfragmented).
 158          */
 159         if (sk->sk_state == DCCP_LISTEN)
 160                 return;
 161
 162         dst = inet_csk_update_pmtu(sk, mtu);
 163         if (!dst)
 164                 return;
 165
 166         /* Something is about to be wrong... Remember soft error
 167          * for the case, if this connection will not able to recover.
 168          */
 169         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 170                 sk->sk_err_soft = EMSGSIZE;
 171
 172         mtu = dst_mtu(dst);
 173
 174         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 175             ip_sk_accept_pmtu(sk) &&
 176             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 177                 dccp_sync_mss(sk, mtu);
 178
 179                 /*
 180                  * From RFC 4340, sec. 14.1:
 181                  *
 182                  *      DCCP-Sync packets are the best choice for upward
 183                  *      probing, since DCCP-Sync probes do not risk application
 184                  *      data loss.
 185                  */
 186                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 187         } /* else let the usual retransmit timer handle it */
 188 }
 189
 190 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
 191 {
 192         struct dst_entry *dst = __sk_dst_check(sk, 0);
 193
 194         if (dst)
 195                 dst->ops->redirect(dst, sk, skb);
 196 }
 197
 198 void dccp_req_err(struct sock *sk, u64 seq)
 199         {
 200         struct request_sock *req = inet_reqsk(sk);
 201         struct net *net = sock_net(sk);
 202
 203         /*
 204          * ICMPs are not backlogged, hence we cannot get an established
 205          * socket here.
 206          */
 207         if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
 208                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 209         } else {
 210                 /*
 211                  * Still in RESPOND, just remove it silently.
 212                  * There is no good way to pass the error to the newly
 213                  * created socket, and POSIX does not want network
 214                  * errors returned from accept().
 215                  */
 216                 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
 217         }
 218         reqsk_put(req);
 219 }
 220 EXPORT_SYMBOL(dccp_req_err);
 221
 222 /*
 223  * This routine is called by the ICMP module when it gets some sort of error
 224  * condition. If err < 0 then the socket should be closed and the error
 225  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 226  * After adjustment header points to the first 8 bytes of the tcp header. We
 227  * need to find the appropriate port.
 228  *
 229  * The locking strategy used here is very "optimistic". When someone else
 230  * accesses the socket the ICMP is just dropped and for some paths there is no
 231  * check at all. A more general error queue to queue errors for later handling
 232  * is probably better.
 233  */
 234 static void dccp_v4_err(struct sk_buff *skb, u32 info)
 235 {
 236         const struct iphdr *iph = (struct iphdr *)skb->data;
 237         const u8 offset = iph->ihl << 2;
 238         const struct dccp_hdr *dh;
 239         struct dccp_sock *dp;
 240         struct inet_sock *inet;
 241         const int type = icmp_hdr(skb)->type;
 242         const int code = icmp_hdr(skb)->code;
 243         struct sock *sk;
 244         __u64 seq;
 245         int err;
 246         struct net *net = dev_net(skb->dev);
 247
 248         /* Only need dccph_dport & dccph_sport which are the first
 249          * 4 bytes in dccp header.
 250          * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us.
 251          */
 252         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_sport) > 8);
 253         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_dport) > 8);
 254         dh = (struct dccp_hdr *)(skb->data + offset);
 255
 256         sk = __inet_lookup_established(net, &dccp_hashinfo,
 257                                        iph->daddr, dh->dccph_dport,
 258                                        iph->saddr, ntohs(dh->dccph_sport),
 259                                        inet_iif(skb));
 260         if (!sk) {
 261                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 262                 return;
 263         }
 264
 265         if (sk->sk_state == DCCP_TIME_WAIT) {
 266                 inet_twsk_put(inet_twsk(sk));
 267                 return;
 268         }
 269         seq = dccp_hdr_seq(dh);
 270         if (sk->sk_state == DCCP_NEW_SYN_RECV)
 271                 return dccp_req_err(sk, seq);
 272
 273         bh_lock_sock(sk);
 274         /* If too many ICMPs get dropped on busy
 275          * servers this needs to be solved differently.
 276          */
 277         if (sock_owned_by_user(sk))
 278                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 279
 280         if (sk->sk_state == DCCP_CLOSED)
 281                 goto out;
 282
 283         dp = dccp_sk(sk);
 284         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 285             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 286                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 287                 goto out;
 288         }
 289
 290         switch (type) {
 291         case ICMP_REDIRECT:
 292                 if (!sock_owned_by_user(sk))
 293                         dccp_do_redirect(skb, sk);
 294                 goto out;
 295         case ICMP_SOURCE_QUENCH:
 296                 /* Just silently ignore these. */
 297                 goto out;
 298         case ICMP_PARAMETERPROB:
 299                 err = EPROTO;
 300                 break;
 301         case ICMP_DEST_UNREACH:
 302                 if (code > NR_ICMP_UNREACH)
 303                         goto out;
 304
 305                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 306                         if (!sock_owned_by_user(sk))
 307                                 dccp_do_pmtu_discovery(sk, iph, info);
 308                         goto out;
 309                 }
 310
 311                 err = icmp_err_convert[code].errno;
 312                 break;
 313         case ICMP_TIME_EXCEEDED:
 314                 err = EHOSTUNREACH;
 315                 break;
 316         default:
 317                 goto out;
 318         }
 319
 320         switch (sk->sk_state) {
 321         case DCCP_REQUESTING:
 322         case DCCP_RESPOND:
 323                 if (!sock_owned_by_user(sk)) {
 324                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 325                         sk->sk_err = err;
 326
 327                         sk->sk_error_report(sk);
 328
 329                         dccp_done(sk);
 330                 } else
 331                         sk->sk_err_soft = err;
 332                 goto out;
 333         }
 334
 335         /* If we've already connected we will keep trying
 336          * until we time out, or the user gives up.
 337          *
 338          * rfc1122 4.2.3.9 allows to consider as hard errors
 339          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 340          * but it is obsoleted by pmtu discovery).
 341          *
 342          * Note, that in modern internet, where routing is unreliable
 343          * and in each dark corner broken firewalls sit, sending random
 344          * errors ordered by their masters even this two messages finally lose
 345          * their original sense (even Linux sends invalid PORT_UNREACHs)
 346          *
 347          * Now we are in compliance with RFCs.
 348          *                                                      --ANK (980905)
 349          */
 350
 351         inet = inet_sk(sk);
 352         if (!sock_owned_by_user(sk) && inet->recverr) {
 353                 sk->sk_err = err;
 354                 sk->sk_error_report(sk);
 355         } else /* Only an error on timeout */
 356                 sk->sk_err_soft = err;
 357 out:
 358         bh_unlock_sock(sk);
 359         sock_put(sk);
 360 }
 361
 362 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 363                                       __be32 src, __be32 dst)
 364 {
 365         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 366 }
 367
 368 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 369 {
 370         const struct inet_sock *inet = inet_sk(sk);
 371         struct dccp_hdr *dh = dccp_hdr(skb);
 372
 373         dccp_csum_outgoing(skb);
 374         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 375                                                  inet->inet_saddr,
 376                                                  inet->inet_daddr);
 377 }
 378 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 379
 380 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 381 {
 382         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 383                                            ip_hdr(skb)->saddr,
 384                                            dccp_hdr(skb)->dccph_dport,
 385                                            dccp_hdr(skb)->dccph_sport);
 386 }
 387
 388 /*
 389  * The three way handshake has completed - we got a valid ACK or DATAACK -
 390  * now create the new socket.
 391  *
 392  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 393  */
 394 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
 395                                        struct sk_buff *skb,
 396                                        struct request_sock *req,
 397                                        struct dst_entry *dst,
 398                                        struct request_sock *req_unhash,
 399                                        bool *own_req)
 400 {
 401         struct inet_request_sock *ireq;
 402         struct inet_sock *newinet;
 403         struct sock *newsk;
 404
 405         if (sk_acceptq_is_full(sk))
 406                 goto exit_overflow;
 407
 408         newsk = dccp_create_openreq_child(sk, req, skb);
 409         if (newsk == NULL)
 410                 goto exit_nonewsk;
 411
 412         newinet            = inet_sk(newsk);
 413         ireq               = inet_rsk(req);
 414         sk_daddr_set(newsk, ireq->ir_rmt_addr);
 415         sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
 416         newinet->inet_saddr     = ireq->ir_loc_addr;
 417         newinet->inet_opt       = ireq->opt;
 418         ireq->opt          = NULL;
 419         newinet->mc_index  = inet_iif(skb);
 420         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 421         newinet->inet_id   = jiffies;
 422
 423         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
 424                 goto put_and_exit;
 425
 426         sk_setup_caps(newsk, dst);
 427
 428         dccp_sync_mss(newsk, dst_mtu(dst));
 429
 430         if (__inet_inherit_port(sk, newsk) < 0)
 431                 goto put_and_exit;
 432         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
 433
 434         return newsk;
 435
 436 exit_overflow:
 437         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 438 exit_nonewsk:
 439         dst_release(dst);
 440 exit:
 441         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 442         return NULL;
 443 put_and_exit:
 444         inet_csk_prepare_forced_close(newsk);
 445         dccp_done(newsk);
 446         goto exit;
 447 }
 448 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 449
 450 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 451                                            struct sk_buff *skb)
 452 {
 453         struct rtable *rt;
 454         const struct iphdr *iph = ip_hdr(skb);
 455         struct flowi4 fl4 = {
 456                 .flowi4_oif = inet_iif(skb),
 457                 .daddr = iph->saddr,
 458                 .saddr = iph->daddr,
 459                 .flowi4_tos = RT_CONN_FLAGS(sk),
 460                 .flowi4_proto = sk->sk_protocol,
 461                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
 462                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
 463         };
 464
 465         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 466         rt = ip_route_output_flow(net, &fl4, sk);
 467         if (IS_ERR(rt)) {
 468                 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 469                 return NULL;
 470         }
 471
 472         return &rt->dst;
 473 }
 474
 475 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
 476 {
 477         int err = -1;
 478         struct sk_buff *skb;
 479         struct dst_entry *dst;
 480         struct flowi4 fl4;
 481
 482         dst = inet_csk_route_req(sk, &fl4, req);
 483         if (dst == NULL)
 484                 goto out;
 485
 486         skb = dccp_make_response(sk, dst, req);
 487         if (skb != NULL) {
 488                 const struct inet_request_sock *ireq = inet_rsk(req);
 489                 struct dccp_hdr *dh = dccp_hdr(skb);
 490
 491                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
 492                                                               ireq->ir_rmt_addr);
 493                 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 494                                             ireq->ir_rmt_addr,
 495                                             ireq->opt);
 496                 err = net_xmit_eval(err);
 497         }
 498
 499 out:
 500         dst_release(dst);
 501         return err;
 502 }
 503
 504 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
 505 {
 506         int err;
 507         const struct iphdr *rxiph;
 508         struct sk_buff *skb;
 509         struct dst_entry *dst;
 510         struct net *net = dev_net(skb_dst(rxskb)->dev);
 511         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
 512
 513         /* Never send a reset in response to a reset. */
 514         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 515                 return;
 516
 517         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 518                 return;
 519
 520         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 521         if (dst == NULL)
 522                 return;
 523
 524         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 525         if (skb == NULL)
 526                 goto out;
 527
 528         rxiph = ip_hdr(rxskb);
 529         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 530                                                                  rxiph->daddr);
 531         skb_dst_set(skb, dst_clone(dst));
 532
 533         bh_lock_sock(ctl_sk);
 534         err = ip_build_and_send_pkt(skb, ctl_sk,
 535                                     rxiph->daddr, rxiph->saddr, NULL);
 536         bh_unlock_sock(ctl_sk);
 537
 538         if (net_xmit_eval(err) == 0) {
 539                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 540                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 541         }
 542 out:
 543          dst_release(dst);
 544 }
 545
 546 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 547 {
 548         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 549         kfree(inet_rsk(req)->opt);
 550 }
 551
 552 void dccp_syn_ack_timeout(const struct request_sock *req)
 553 {
 554 }
 555 EXPORT_SYMBOL(dccp_syn_ack_timeout);
 556
 557 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 558         .family         = PF_INET,
 559         .obj_size       = sizeof(struct dccp_request_sock),
 560         .rtx_syn_ack    = dccp_v4_send_response,
 561         .send_ack       = dccp_reqsk_send_ack,
 562         .destructor     = dccp_v4_reqsk_destructor,
 563         .send_reset     = dccp_v4_ctl_send_reset,
 564         .syn_ack_timeout = dccp_syn_ack_timeout,
 565 };
 566
 567 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 568 {
 569         struct inet_request_sock *ireq;
 570         struct request_sock *req;
 571         struct dccp_request_sock *dreq;
 572         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 573         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 574
 575         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 576         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 577                 return 0;       /* discard, don't send a reset here */
 578
 579         if (dccp_bad_service_code(sk, service)) {
 580                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 581                 goto drop;
 582         }
 583         /*
 584          * TW buckets are converted to open requests without
 585          * limitations, they conserve resources and peer is
 586          * evidently real one.
 587          */
 588         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 589         if (inet_csk_reqsk_queue_is_full(sk))
 590                 goto drop;
 591
 592         /*
 593          * Accept backlog is full. If we have already queued enough
 594          * of warm entries in syn queue, drop request. It is better than
 595          * clogging syn queue with openreqs with exponentially increasing
 596          * timeout.
 597          */
 598         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 599                 goto drop;
 600
 601         req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
 602         if (req == NULL)
 603                 goto drop;
 604
 605         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 606                 goto drop_and_free;
 607
 608         dreq = dccp_rsk(req);
 609         if (dccp_parse_options(sk, dreq, skb))
 610                 goto drop_and_free;
 611
 612         if (security_inet_conn_request(sk, skb, req))
 613                 goto drop_and_free;
 614
 615         ireq = inet_rsk(req);
 616         sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
 617         sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
 618         ireq->ireq_family = AF_INET;
 619         ireq->ir_iif = sk->sk_bound_dev_if;
 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          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 627          */
 628         dreq->dreq_isr     = dcb->dccpd_seq;
 629         dreq->dreq_gsr     = dreq->dreq_isr;
 630         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 631         dreq->dreq_gss     = dreq->dreq_iss;
 632         dreq->dreq_service = service;
 633
 634         if (dccp_v4_send_response(sk, req))
 635                 goto drop_and_free;
 636
 637         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 638         return 0;
 639
 640 drop_and_free:
 641         reqsk_free(req);
 642 drop:
 643         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 644         return -1;
 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
 682         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 683                 goto reset;
 684         return 0;
 685
 686 reset:
 687         dccp_v4_ctl_send_reset(sk, skb);
 688         kfree_skb(skb);
 689         return 0;
 690 }
 691 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 692
 693 /**
 694  *      dccp_invalid_packet  -  check for malformed packets
 695  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 696  *      Packets that fail these checks are ignored and do not receive Resets.
 697  */
 698 int dccp_invalid_packet(struct sk_buff *skb)
 699 {
 700         const struct dccp_hdr *dh;
 701         unsigned int cscov;
 702         u8 dccph_doff;
 703
 704         if (skb->pkt_type != PACKET_HOST)
 705                 return 1;
 706
 707         /* If the packet is shorter than 12 bytes, drop packet and return */
 708         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 709                 DCCP_WARN("pskb_may_pull failed\n");
 710                 return 1;
 711         }
 712
 713         dh = dccp_hdr(skb);
 714
 715         /* If P.type is not understood, drop packet and return */
 716         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 717                 DCCP_WARN("invalid packet type\n");
 718                 return 1;
 719         }
 720
 721         /*
 722          * If P.Data Offset is too small for packet type, drop packet and return
 723          */
 724         dccph_doff = dh->dccph_doff;
 725         if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 726                 DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
 727                 return 1;
 728         }
 729         /*
 730          * If P.Data Offset is too too large for packet, drop packet and return
 731          */
 732         if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
 733                 DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
 734                 return 1;
 735         }
 736         dh = dccp_hdr(skb);
 737         /*
 738          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 739          * has short sequence numbers), drop packet and return
 740          */
 741         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 742             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 743                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 744                           dccp_packet_name(dh->dccph_type));
 745                 return 1;
 746         }
 747
 748         /*
 749          * If P.CsCov is too large for the packet size, drop packet and return.
 750          * This must come _before_ checksumming (not as RFC 4340 suggests).
 751          */
 752         cscov = dccp_csum_coverage(skb);
 753         if (cscov > skb->len) {
 754                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 755                           dh->dccph_cscov, skb->len);
 756                 return 1;
 757         }
 758
 759         /* If header checksum is incorrect, drop packet and return.
 760          * (This step is completed in the AF-dependent functions.) */
 761         skb->csum = skb_checksum(skb, 0, cscov, 0);
 762
 763         return 0;
 764 }
 765 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 766
 767 /* this is called when real data arrives */
 768 static int dccp_v4_rcv(struct sk_buff *skb)
 769 {
 770         const struct dccp_hdr *dh;
 771         const struct iphdr *iph;
 772         struct sock *sk;
 773         int min_cov;
 774
 775         /* Step 1: Check header basics */
 776
 777         if (dccp_invalid_packet(skb))
 778                 goto discard_it;
 779
 780         iph = ip_hdr(skb);
 781         /* Step 1: If header checksum is incorrect, drop packet and return */
 782         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 783                 DCCP_WARN("dropped packet with invalid checksum\n");
 784                 goto discard_it;
 785         }
 786
 787         dh = dccp_hdr(skb);
 788
 789         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 790         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 791
 792         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 793                       dccp_packet_name(dh->dccph_type),
 794                       &iph->saddr, ntohs(dh->dccph_sport),
 795                       &iph->daddr, ntohs(dh->dccph_dport),
 796                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 797
 798         if (dccp_packet_without_ack(skb)) {
 799                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 800                 dccp_pr_debug_cat("\n");
 801         } else {
 802                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 803                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 804                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 805         }
 806
 807 lookup:
 808         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
 809                                dh->dccph_sport, dh->dccph_dport);
 810         if (!sk) {
 811                 dccp_pr_debug("failed to look up flow ID in table and "
 812                               "get corresponding socket\n");
 813                 goto no_dccp_socket;
 814         }
 815
 816         /*
 817          * Step 2:
 818          *      ... or S.state == TIMEWAIT,
 819          *              Generate Reset(No Connection) unless P.type == Reset
 820          *              Drop packet and return
 821          */
 822         if (sk->sk_state == DCCP_TIME_WAIT) {
 823                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 824                 inet_twsk_put(inet_twsk(sk));
 825                 goto no_dccp_socket;
 826         }
 827
 828         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 829                 struct request_sock *req = inet_reqsk(sk);
 830                 struct sock *nsk;
 831
 832                 sk = req->rsk_listener;
 833                 if (unlikely(sk->sk_state != DCCP_LISTEN)) {
 834                         inet_csk_reqsk_queue_drop_and_put(sk, req);
 835                         goto lookup;
 836                 }
 837                 sock_hold(sk);
 838                 nsk = dccp_check_req(sk, skb, req);
 839                 if (!nsk) {
 840                         reqsk_put(req);
 841                         goto discard_and_relse;
 842                 }
 843                 if (nsk == sk) {
 844                         reqsk_put(req);
 845                 } else if (dccp_child_process(sk, nsk, skb)) {
 846                         dccp_v4_ctl_send_reset(sk, skb);
 847                         goto discard_and_relse;
 848                 } else {
 849                         sock_put(sk);
 850                         return 0;
 851                 }
 852         }
 853         /*
 854          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 855          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 856          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 857          */
 858         min_cov = dccp_sk(sk)->dccps_pcrlen;
 859         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 860                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 861                               dh->dccph_cscov, min_cov);
 862                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 863                  *         options (Section 11.7) with Drop Code 0, Protocol
 864                  *         Constraints."                                     */
 865                 goto discard_and_relse;
 866         }
 867
 868         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 869                 goto discard_and_relse;
 870         nf_reset(skb);
 871
 872         return sk_receive_skb(sk, skb, 1);
 873
 874 no_dccp_socket:
 875         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 876                 goto discard_it;
 877         /*
 878          * Step 2:
 879          *      If no socket ...
 880          *              Generate Reset(No Connection) unless P.type == Reset
 881          *              Drop packet and return
 882          */
 883         if (dh->dccph_type != DCCP_PKT_RESET) {
 884                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 885                                         DCCP_RESET_CODE_NO_CONNECTION;
 886                 dccp_v4_ctl_send_reset(sk, skb);
 887         }
 888
 889 discard_it:
 890         kfree_skb(skb);
 891         return 0;
 892
 893 discard_and_relse:
 894         sock_put(sk);
 895         goto discard_it;
 896 }
 897
 898 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 899         .queue_xmit        = ip_queue_xmit,
 900         .send_check        = dccp_v4_send_check,
 901         .rebuild_header    = inet_sk_rebuild_header,
 902         .conn_request      = dccp_v4_conn_request,
 903         .syn_recv_sock     = dccp_v4_request_recv_sock,
 904         .net_header_len    = sizeof(struct iphdr),
 905         .setsockopt        = ip_setsockopt,
 906         .getsockopt        = ip_getsockopt,
 907         .addr2sockaddr     = inet_csk_addr2sockaddr,
 908         .sockaddr_len      = sizeof(struct sockaddr_in),
 909         .bind_conflict     = inet_csk_bind_conflict,
 910 #ifdef CONFIG_COMPAT
 911         .compat_setsockopt = compat_ip_setsockopt,
 912         .compat_getsockopt = compat_ip_getsockopt,
 913 #endif
 914 };
 915
 916 static int dccp_v4_init_sock(struct sock *sk)
 917 {
 918         static __u8 dccp_v4_ctl_sock_initialized;
 919         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 920
 921         if (err == 0) {
 922                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 923                         dccp_v4_ctl_sock_initialized = 1;
 924                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 925         }
 926
 927         return err;
 928 }
 929
 930 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 931         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 932 };
 933
 934 static struct proto dccp_v4_prot = {
 935         .name                   = "DCCP",
 936         .owner                  = THIS_MODULE,
 937         .close                  = dccp_close,
 938         .connect                = dccp_v4_connect,
 939         .disconnect             = dccp_disconnect,
 940         .ioctl                  = dccp_ioctl,
 941         .init                   = dccp_v4_init_sock,
 942         .setsockopt             = dccp_setsockopt,
 943         .getsockopt             = dccp_getsockopt,
 944         .sendmsg                = dccp_sendmsg,
 945         .recvmsg                = dccp_recvmsg,
 946         .backlog_rcv            = dccp_v4_do_rcv,
 947         .hash                   = inet_hash,
 948         .unhash                 = inet_unhash,
 949         .accept                 = inet_csk_accept,
 950         .get_port               = inet_csk_get_port,
 951         .shutdown               = dccp_shutdown,
 952         .destroy                = dccp_destroy_sock,
 953         .orphan_count           = &dccp_orphan_count,
 954         .max_header             = MAX_DCCP_HEADER,
 955         .obj_size               = sizeof(struct dccp_sock),
 956         .slab_flags             = SLAB_DESTROY_BY_RCU,
 957         .rsk_prot               = &dccp_request_sock_ops,
 958         .twsk_prot              = &dccp_timewait_sock_ops,
 959         .h.hashinfo             = &dccp_hashinfo,
 960 #ifdef CONFIG_COMPAT
 961         .compat_setsockopt      = compat_dccp_setsockopt,
 962         .compat_getsockopt      = compat_dccp_getsockopt,
 963 #endif
 964 };
 965
 966 static const struct net_protocol dccp_v4_protocol = {
 967         .handler        = dccp_v4_rcv,
 968         .err_handler    = dccp_v4_err,
 969         .no_policy      = 1,
 970         .netns_ok       = 1,
 971         .icmp_strict_tag_validation = 1,
 972 };
 973
 974 static const struct proto_ops inet_dccp_ops = {
 975         .family            = PF_INET,
 976         .owner             = THIS_MODULE,
 977         .release           = inet_release,
 978         .bind              = inet_bind,
 979         .connect           = inet_stream_connect,
 980         .socketpair        = sock_no_socketpair,
 981         .accept            = inet_accept,
 982         .getname           = inet_getname,
 983         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
 984         .poll              = dccp_poll,
 985         .ioctl             = inet_ioctl,
 986         /* FIXME: work on inet_listen to rename it to sock_common_listen */
 987         .listen            = inet_dccp_listen,
 988         .shutdown          = inet_shutdown,
 989         .setsockopt        = sock_common_setsockopt,
 990         .getsockopt        = sock_common_getsockopt,
 991         .sendmsg           = inet_sendmsg,
 992         .recvmsg           = sock_common_recvmsg,
 993         .mmap              = sock_no_mmap,
 994         .sendpage          = sock_no_sendpage,
 995 #ifdef CONFIG_COMPAT
 996         .compat_setsockopt = compat_sock_common_setsockopt,
 997         .compat_getsockopt = compat_sock_common_getsockopt,
 998 #endif
 999 };
1000
1001 static struct inet_protosw dccp_v4_protosw = {
1002         .type           = SOCK_DCCP,
1003         .protocol       = IPPROTO_DCCP,
1004         .prot           = &dccp_v4_prot,
1005         .ops            = &inet_dccp_ops,
1006         .flags          = INET_PROTOSW_ICSK,
1007 };
1008
1009 static int __net_init dccp_v4_init_net(struct net *net)
1010 {
1011         if (dccp_hashinfo.bhash == NULL)
1012                 return -ESOCKTNOSUPPORT;
1013
1014         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1015                                     SOCK_DCCP, IPPROTO_DCCP, net);
1016 }
1017
1018 static void __net_exit dccp_v4_exit_net(struct net *net)
1019 {
1020         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1021 }
1022
1023 static struct pernet_operations dccp_v4_ops = {
1024         .init   = dccp_v4_init_net,
1025         .exit   = dccp_v4_exit_net,
1026 };
1027
1028 static int __init dccp_v4_init(void)
1029 {
1030         int err = proto_register(&dccp_v4_prot, 1);
1031
1032         if (err != 0)
1033                 goto out;
1034
1035         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1036         if (err != 0)
1037                 goto out_proto_unregister;
1038
1039         inet_register_protosw(&dccp_v4_protosw);
1040
1041         err = register_pernet_subsys(&dccp_v4_ops);
1042         if (err)
1043                 goto out_destroy_ctl_sock;
1044 out:
1045         return err;
1046 out_destroy_ctl_sock:
1047         inet_unregister_protosw(&dccp_v4_protosw);
1048         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1049 out_proto_unregister:
1050         proto_unregister(&dccp_v4_prot);
1051         goto out;
1052 }
1053
1054 static void __exit dccp_v4_exit(void)
1055 {
1056         unregister_pernet_subsys(&dccp_v4_ops);
1057         inet_unregister_protosw(&dccp_v4_protosw);
1058         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1059         proto_unregister(&dccp_v4_prot);
1060 }
1061
1062 module_init(dccp_v4_init);
1063 module_exit(dccp_v4_exit);
1064
1065 /*
1066  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1067  * values directly, Also cover the case where the protocol is not specified,
1068  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1069  */
1070 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1071 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1072 MODULE_LICENSE("GPL");
1073 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1074 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");