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                                              lockdep_sock_is_held(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(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(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(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(net, LINUX_MIB_OUTOFWINDOWICMPS);
 287                 goto out;
 288         }
 289
 290         switch (type) {
 291         case ICMP_REDIRECT:
 292                 dccp_do_redirect(skb, sk);
 293                 goto out;
 294         case ICMP_SOURCE_QUENCH:
 295                 /* Just silently ignore these. */
 296                 goto out;
 297         case ICMP_PARAMETERPROB:
 298                 err = EPROTO;
 299                 break;
 300         case ICMP_DEST_UNREACH:
 301                 if (code > NR_ICMP_UNREACH)
 302                         goto out;
 303
 304                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 305                         if (!sock_owned_by_user(sk))
 306                                 dccp_do_pmtu_discovery(sk, iph, info);
 307                         goto out;
 308                 }
 309
 310                 err = icmp_err_convert[code].errno;
 311                 break;
 312         case ICMP_TIME_EXCEEDED:
 313                 err = EHOSTUNREACH;
 314                 break;
 315         default:
 316                 goto out;
 317         }
 318
 319         switch (sk->sk_state) {
 320         case DCCP_REQUESTING:
 321         case DCCP_RESPOND:
 322                 if (!sock_owned_by_user(sk)) {
 323                         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 324                         sk->sk_err = err;
 325
 326                         sk->sk_error_report(sk);
 327
 328                         dccp_done(sk);
 329                 } else
 330                         sk->sk_err_soft = err;
 331                 goto out;
 332         }
 333
 334         /* If we've already connected we will keep trying
 335          * until we time out, or the user gives up.
 336          *
 337          * rfc1122 4.2.3.9 allows to consider as hard errors
 338          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 339          * but it is obsoleted by pmtu discovery).
 340          *
 341          * Note, that in modern internet, where routing is unreliable
 342          * and in each dark corner broken firewalls sit, sending random
 343          * errors ordered by their masters even this two messages finally lose
 344          * their original sense (even Linux sends invalid PORT_UNREACHs)
 345          *
 346          * Now we are in compliance with RFCs.
 347          *                                                      --ANK (980905)
 348          */
 349
 350         inet = inet_sk(sk);
 351         if (!sock_owned_by_user(sk) && inet->recverr) {
 352                 sk->sk_err = err;
 353                 sk->sk_error_report(sk);
 354         } else /* Only an error on timeout */
 355                 sk->sk_err_soft = err;
 356 out:
 357         bh_unlock_sock(sk);
 358         sock_put(sk);
 359 }
 360
 361 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 362                                       __be32 src, __be32 dst)
 363 {
 364         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 365 }
 366
 367 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 368 {
 369         const struct inet_sock *inet = inet_sk(sk);
 370         struct dccp_hdr *dh = dccp_hdr(skb);
 371
 372         dccp_csum_outgoing(skb);
 373         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 374                                                  inet->inet_saddr,
 375                                                  inet->inet_daddr);
 376 }
 377 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 378
 379 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 380 {
 381         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 382                                            ip_hdr(skb)->saddr,
 383                                            dccp_hdr(skb)->dccph_dport,
 384                                            dccp_hdr(skb)->dccph_sport);
 385 }
 386
 387 /*
 388  * The three way handshake has completed - we got a valid ACK or DATAACK -
 389  * now create the new socket.
 390  *
 391  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 392  */
 393 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
 394                                        struct sk_buff *skb,
 395                                        struct request_sock *req,
 396                                        struct dst_entry *dst,
 397                                        struct request_sock *req_unhash,
 398                                        bool *own_req)
 399 {
 400         struct inet_request_sock *ireq;
 401         struct inet_sock *newinet;
 402         struct sock *newsk;
 403
 404         if (sk_acceptq_is_full(sk))
 405                 goto exit_overflow;
 406
 407         newsk = dccp_create_openreq_child(sk, req, skb);
 408         if (newsk == NULL)
 409                 goto exit_nonewsk;
 410
 411         newinet            = inet_sk(newsk);
 412         ireq               = inet_rsk(req);
 413         sk_daddr_set(newsk, ireq->ir_rmt_addr);
 414         sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
 415         newinet->inet_saddr     = ireq->ir_loc_addr;
 416         newinet->inet_opt       = ireq->opt;
 417         ireq->opt          = NULL;
 418         newinet->mc_index  = inet_iif(skb);
 419         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 420         newinet->inet_id   = jiffies;
 421
 422         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
 423                 goto put_and_exit;
 424
 425         sk_setup_caps(newsk, dst);
 426
 427         dccp_sync_mss(newsk, dst_mtu(dst));
 428
 429         if (__inet_inherit_port(sk, newsk) < 0)
 430                 goto put_and_exit;
 431         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
 432
 433         return newsk;
 434
 435 exit_overflow:
 436         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 437 exit_nonewsk:
 438         dst_release(dst);
 439 exit:
 440         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
 441         return NULL;
 442 put_and_exit:
 443         inet_csk_prepare_forced_close(newsk);
 444         dccp_done(newsk);
 445         goto exit;
 446 }
 447 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 448
 449 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 450                                            struct sk_buff *skb)
 451 {
 452         struct rtable *rt;
 453         const struct iphdr *iph = ip_hdr(skb);
 454         struct flowi4 fl4 = {
 455                 .flowi4_oif = inet_iif(skb),
 456                 .daddr = iph->saddr,
 457                 .saddr = iph->daddr,
 458                 .flowi4_tos = RT_CONN_FLAGS(sk),
 459                 .flowi4_proto = sk->sk_protocol,
 460                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
 461                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
 462         };
 463
 464         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 465         rt = ip_route_output_flow(net, &fl4, sk);
 466         if (IS_ERR(rt)) {
 467                 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 468                 return NULL;
 469         }
 470
 471         return &rt->dst;
 472 }
 473
 474 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
 475 {
 476         int err = -1;
 477         struct sk_buff *skb;
 478         struct dst_entry *dst;
 479         struct flowi4 fl4;
 480
 481         dst = inet_csk_route_req(sk, &fl4, req);
 482         if (dst == NULL)
 483                 goto out;
 484
 485         skb = dccp_make_response(sk, dst, req);
 486         if (skb != NULL) {
 487                 const struct inet_request_sock *ireq = inet_rsk(req);
 488                 struct dccp_hdr *dh = dccp_hdr(skb);
 489
 490                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
 491                                                               ireq->ir_rmt_addr);
 492                 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 493                                             ireq->ir_rmt_addr,
 494                                             ireq->opt);
 495                 err = net_xmit_eval(err);
 496         }
 497
 498 out:
 499         dst_release(dst);
 500         return err;
 501 }
 502
 503 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
 504 {
 505         int err;
 506         const struct iphdr *rxiph;
 507         struct sk_buff *skb;
 508         struct dst_entry *dst;
 509         struct net *net = dev_net(skb_dst(rxskb)->dev);
 510         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
 511
 512         /* Never send a reset in response to a reset. */
 513         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 514                 return;
 515
 516         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 517                 return;
 518
 519         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 520         if (dst == NULL)
 521                 return;
 522
 523         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 524         if (skb == NULL)
 525                 goto out;
 526
 527         rxiph = ip_hdr(rxskb);
 528         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 529                                                                  rxiph->daddr);
 530         skb_dst_set(skb, dst_clone(dst));
 531
 532         local_bh_disable();
 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(DCCP_MIB_OUTSEGS);
 540                 __DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
 541         }
 542         local_bh_enable();
 543 out:
 544         dst_release(dst);
 545 }
 546
 547 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 548 {
 549         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 550         kfree(inet_rsk(req)->opt);
 551 }
 552
 553 void dccp_syn_ack_timeout(const struct request_sock *req)
 554 {
 555 }
 556 EXPORT_SYMBOL(dccp_syn_ack_timeout);
 557
 558 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 559         .family         = PF_INET,
 560         .obj_size       = sizeof(struct dccp_request_sock),
 561         .rtx_syn_ack    = dccp_v4_send_response,
 562         .send_ack       = dccp_reqsk_send_ack,
 563         .destructor     = dccp_v4_reqsk_destructor,
 564         .send_reset     = dccp_v4_ctl_send_reset,
 565         .syn_ack_timeout = dccp_syn_ack_timeout,
 566 };
 567
 568 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 569 {
 570         struct inet_request_sock *ireq;
 571         struct request_sock *req;
 572         struct dccp_request_sock *dreq;
 573         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 574         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 575
 576         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 577         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 578                 return 0;       /* discard, don't send a reset here */
 579
 580         if (dccp_bad_service_code(sk, service)) {
 581                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 582                 goto drop;
 583         }
 584         /*
 585          * TW buckets are converted to open requests without
 586          * limitations, they conserve resources and peer is
 587          * evidently real one.
 588          */
 589         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 590         if (inet_csk_reqsk_queue_is_full(sk))
 591                 goto drop;
 592
 593         /*
 594          * Accept backlog is full. If we have already queued enough
 595          * of warm entries in syn queue, drop request. It is better than
 596          * clogging syn queue with openreqs with exponentially increasing
 597          * timeout.
 598          */
 599         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 600                 goto drop;
 601
 602         req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
 603         if (req == NULL)
 604                 goto drop;
 605
 606         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 607                 goto drop_and_free;
 608
 609         dreq = dccp_rsk(req);
 610         if (dccp_parse_options(sk, dreq, skb))
 611                 goto drop_and_free;
 612
 613         if (security_inet_conn_request(sk, skb, req))
 614                 goto drop_and_free;
 615
 616         ireq = inet_rsk(req);
 617         sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
 618         sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
 619         ireq->ireq_family = AF_INET;
 620         ireq->ir_iif = sk->sk_bound_dev_if;
 621
 622         /*
 623          * Step 3: Process LISTEN state
 624          *
 625          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 626          *
 627          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 628          */
 629         dreq->dreq_isr     = dcb->dccpd_seq;
 630         dreq->dreq_gsr     = dreq->dreq_isr;
 631         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 632         dreq->dreq_gss     = dreq->dreq_iss;
 633         dreq->dreq_service = service;
 634
 635         if (dccp_v4_send_response(sk, req))
 636                 goto drop_and_free;
 637
 638         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 639         return 0;
 640
 641 drop_and_free:
 642         reqsk_free(req);
 643 drop:
 644         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 645         return -1;
 646 }
 647 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 648
 649 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 650 {
 651         struct dccp_hdr *dh = dccp_hdr(skb);
 652
 653         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 654                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 655                         goto reset;
 656                 return 0;
 657         }
 658
 659         /*
 660          *  Step 3: Process LISTEN state
 661          *       If P.type == Request or P contains a valid Init Cookie option,
 662          *            (* Must scan the packet's options to check for Init
 663          *               Cookies.  Only Init Cookies are processed here,
 664          *               however; other options are processed in Step 8.  This
 665          *               scan need only be performed if the endpoint uses Init
 666          *               Cookies *)
 667          *            (* Generate a new socket and switch to that socket *)
 668          *            Set S := new socket for this port pair
 669          *            S.state = RESPOND
 670          *            Choose S.ISS (initial seqno) or set from Init Cookies
 671          *            Initialize S.GAR := S.ISS
 672          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 673          *            Continue with S.state == RESPOND
 674          *            (* A Response packet will be generated in Step 11 *)
 675          *       Otherwise,
 676          *            Generate Reset(No Connection) unless P.type == Reset
 677          *            Drop packet and return
 678          *
 679          * NOTE: the check for the packet types is done in
 680          *       dccp_rcv_state_process
 681          */
 682
 683         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 684                 goto reset;
 685         return 0;
 686
 687 reset:
 688         dccp_v4_ctl_send_reset(sk, skb);
 689         kfree_skb(skb);
 690         return 0;
 691 }
 692 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 693
 694 /**
 695  *      dccp_invalid_packet  -  check for malformed packets
 696  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 697  *      Packets that fail these checks are ignored and do not receive Resets.
 698  */
 699 int dccp_invalid_packet(struct sk_buff *skb)
 700 {
 701         const struct dccp_hdr *dh;
 702         unsigned int cscov;
 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         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 725                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
 726                 return 1;
 727         }
 728         /*
 729          * If P.Data Offset is too too large for packet, drop packet and return
 730          */
 731         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 732                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
 733                 return 1;
 734         }
 735
 736         /*
 737          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 738          * has short sequence numbers), drop packet and return
 739          */
 740         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 741             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 742                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 743                           dccp_packet_name(dh->dccph_type));
 744                 return 1;
 745         }
 746
 747         /*
 748          * If P.CsCov is too large for the packet size, drop packet and return.
 749          * This must come _before_ checksumming (not as RFC 4340 suggests).
 750          */
 751         cscov = dccp_csum_coverage(skb);
 752         if (cscov > skb->len) {
 753                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 754                           dh->dccph_cscov, skb->len);
 755                 return 1;
 756         }
 757
 758         /* If header checksum is incorrect, drop packet and return.
 759          * (This step is completed in the AF-dependent functions.) */
 760         skb->csum = skb_checksum(skb, 0, cscov, 0);
 761
 762         return 0;
 763 }
 764 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 765
 766 /* this is called when real data arrives */
 767 static int dccp_v4_rcv(struct sk_buff *skb)
 768 {
 769         const struct dccp_hdr *dh;
 770         const struct iphdr *iph;
 771         bool refcounted;
 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, __dccp_hdr_len(dh),
 809                                dh->dccph_sport, dh->dccph_dport, &refcounted);
 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                 refcounted = true;
 839                 nsk = dccp_check_req(sk, skb, req);
 840                 if (!nsk) {
 841                         reqsk_put(req);
 842                         goto discard_and_relse;
 843                 }
 844                 if (nsk == sk) {
 845                         reqsk_put(req);
 846                 } else if (dccp_child_process(sk, nsk, skb)) {
 847                         dccp_v4_ctl_send_reset(sk, skb);
 848                         goto discard_and_relse;
 849                 } else {
 850                         sock_put(sk);
 851                         return 0;
 852                 }
 853         }
 854         /*
 855          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 856          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 857          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 858          */
 859         min_cov = dccp_sk(sk)->dccps_pcrlen;
 860         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 861                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 862                               dh->dccph_cscov, min_cov);
 863                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 864                  *         options (Section 11.7) with Drop Code 0, Protocol
 865                  *         Constraints."                                     */
 866                 goto discard_and_relse;
 867         }
 868
 869         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 870                 goto discard_and_relse;
 871         nf_reset(skb);
 872
 873         return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
 874
 875 no_dccp_socket:
 876         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 877                 goto discard_it;
 878         /*
 879          * Step 2:
 880          *      If no socket ...
 881          *              Generate Reset(No Connection) unless P.type == Reset
 882          *              Drop packet and return
 883          */
 884         if (dh->dccph_type != DCCP_PKT_RESET) {
 885                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 886                                         DCCP_RESET_CODE_NO_CONNECTION;
 887                 dccp_v4_ctl_send_reset(sk, skb);
 888         }
 889
 890 discard_it:
 891         kfree_skb(skb);
 892         return 0;
 893
 894 discard_and_relse:
 895         if (refcounted)
 896                 sock_put(sk);
 897         goto discard_it;
 898 }
 899
 900 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 901         .queue_xmit        = ip_queue_xmit,
 902         .send_check        = dccp_v4_send_check,
 903         .rebuild_header    = inet_sk_rebuild_header,
 904         .conn_request      = dccp_v4_conn_request,
 905         .syn_recv_sock     = dccp_v4_request_recv_sock,
 906         .net_header_len    = sizeof(struct iphdr),
 907         .setsockopt        = ip_setsockopt,
 908         .getsockopt        = ip_getsockopt,
 909         .addr2sockaddr     = inet_csk_addr2sockaddr,
 910         .sockaddr_len      = sizeof(struct sockaddr_in),
 911         .bind_conflict     = inet_csk_bind_conflict,
 912 #ifdef CONFIG_COMPAT
 913         .compat_setsockopt = compat_ip_setsockopt,
 914         .compat_getsockopt = compat_ip_getsockopt,
 915 #endif
 916 };
 917
 918 static int dccp_v4_init_sock(struct sock *sk)
 919 {
 920         static __u8 dccp_v4_ctl_sock_initialized;
 921         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 922
 923         if (err == 0) {
 924                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 925                         dccp_v4_ctl_sock_initialized = 1;
 926                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 927         }
 928
 929         return err;
 930 }
 931
 932 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 933         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 934 };
 935
 936 static struct proto dccp_v4_prot = {
 937         .name                   = "DCCP",
 938         .owner                  = THIS_MODULE,
 939         .close                  = dccp_close,
 940         .connect                = dccp_v4_connect,
 941         .disconnect             = dccp_disconnect,
 942         .ioctl                  = dccp_ioctl,
 943         .init                   = dccp_v4_init_sock,
 944         .setsockopt             = dccp_setsockopt,
 945         .getsockopt             = dccp_getsockopt,
 946         .sendmsg                = dccp_sendmsg,
 947         .recvmsg                = dccp_recvmsg,
 948         .backlog_rcv            = dccp_v4_do_rcv,
 949         .hash                   = inet_hash,
 950         .unhash                 = inet_unhash,
 951         .accept                 = inet_csk_accept,
 952         .get_port               = inet_csk_get_port,
 953         .shutdown               = dccp_shutdown,
 954         .destroy                = dccp_destroy_sock,
 955         .orphan_count           = &dccp_orphan_count,
 956         .max_header             = MAX_DCCP_HEADER,
 957         .obj_size               = sizeof(struct dccp_sock),
 958         .slab_flags             = SLAB_DESTROY_BY_RCU,
 959         .rsk_prot               = &dccp_request_sock_ops,
 960         .twsk_prot              = &dccp_timewait_sock_ops,
 961         .h.hashinfo             = &dccp_hashinfo,
 962 #ifdef CONFIG_COMPAT
 963         .compat_setsockopt      = compat_dccp_setsockopt,
 964         .compat_getsockopt      = compat_dccp_getsockopt,
 965 #endif
 966 };
 967
 968 static const struct net_protocol dccp_v4_protocol = {
 969         .handler        = dccp_v4_rcv,
 970         .err_handler    = dccp_v4_err,
 971         .no_policy      = 1,
 972         .netns_ok       = 1,
 973         .icmp_strict_tag_validation = 1,
 974 };
 975
 976 static const struct proto_ops inet_dccp_ops = {
 977         .family            = PF_INET,
 978         .owner             = THIS_MODULE,
 979         .release           = inet_release,
 980         .bind              = inet_bind,
 981         .connect           = inet_stream_connect,
 982         .socketpair        = sock_no_socketpair,
 983         .accept            = inet_accept,
 984         .getname           = inet_getname,
 985         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
 986         .poll              = dccp_poll,
 987         .ioctl             = inet_ioctl,
 988         /* FIXME: work on inet_listen to rename it to sock_common_listen */
 989         .listen            = inet_dccp_listen,
 990         .shutdown          = inet_shutdown,
 991         .setsockopt        = sock_common_setsockopt,
 992         .getsockopt        = sock_common_getsockopt,
 993         .sendmsg           = inet_sendmsg,
 994         .recvmsg           = sock_common_recvmsg,
 995         .mmap              = sock_no_mmap,
 996         .sendpage          = sock_no_sendpage,
 997 #ifdef CONFIG_COMPAT
 998         .compat_setsockopt = compat_sock_common_setsockopt,
 999         .compat_getsockopt = compat_sock_common_getsockopt,
1000 #endif
1001 };
1002
1003 static struct inet_protosw dccp_v4_protosw = {
1004         .type           = SOCK_DCCP,
1005         .protocol       = IPPROTO_DCCP,
1006         .prot           = &dccp_v4_prot,
1007         .ops            = &inet_dccp_ops,
1008         .flags          = INET_PROTOSW_ICSK,
1009 };
1010
1011 static int __net_init dccp_v4_init_net(struct net *net)
1012 {
1013         if (dccp_hashinfo.bhash == NULL)
1014                 return -ESOCKTNOSUPPORT;
1015
1016         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1017                                     SOCK_DCCP, IPPROTO_DCCP, net);
1018 }
1019
1020 static void __net_exit dccp_v4_exit_net(struct net *net)
1021 {
1022         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1023 }
1024
1025 static struct pernet_operations dccp_v4_ops = {
1026         .init   = dccp_v4_init_net,
1027         .exit   = dccp_v4_exit_net,
1028 };
1029
1030 static int __init dccp_v4_init(void)
1031 {
1032         int err = proto_register(&dccp_v4_prot, 1);
1033
1034         if (err != 0)
1035                 goto out;
1036
1037         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1038         if (err != 0)
1039                 goto out_proto_unregister;
1040
1041         inet_register_protosw(&dccp_v4_protosw);
1042
1043         err = register_pernet_subsys(&dccp_v4_ops);
1044         if (err)
1045                 goto out_destroy_ctl_sock;
1046 out:
1047         return err;
1048 out_destroy_ctl_sock:
1049         inet_unregister_protosw(&dccp_v4_protosw);
1050         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1051 out_proto_unregister:
1052         proto_unregister(&dccp_v4_prot);
1053         goto out;
1054 }
1055
1056 static void __exit dccp_v4_exit(void)
1057 {
1058         unregister_pernet_subsys(&dccp_v4_ops);
1059         inet_unregister_protosw(&dccp_v4_protosw);
1060         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1061         proto_unregister(&dccp_v4_prot);
1062 }
1063
1064 module_init(dccp_v4_init);
1065 module_exit(dccp_v4_exit);
1066
1067 /*
1068  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1069  * values directly, Also cover the case where the protocol is not specified,
1070  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1071  */
1072 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1073 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1074 MODULE_LICENSE("GPL");
1075 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1076 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");