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, true);
  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         inet->inet_rcv_saddr = inet->inet_saddr;
  93
  94         inet->inet_dport = usin->sin_port;
  95         inet->inet_daddr = daddr;
  96
  97         inet_csk(sk)->icsk_ext_hdr_len = 0;
  98         if (inet_opt)
  99                 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
 100         /*
 101          * Socket identity is still unknown (sport may be zero).
 102          * However we set state to DCCP_REQUESTING and not releasing socket
 103          * lock select source port, enter ourselves into the hash tables and
 104          * complete initialization after this.
 105          */
 106         dccp_set_state(sk, DCCP_REQUESTING);
 107         err = inet_hash_connect(&dccp_death_row, sk);
 108         if (err != 0)
 109                 goto failure;
 110
 111         rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 112                                inet->inet_sport, inet->inet_dport, sk);
 113         if (IS_ERR(rt)) {
 114                 err = PTR_ERR(rt);
 115                 rt = NULL;
 116                 goto failure;
 117         }
 118         /* OK, now commit destination to socket.  */
 119         sk_setup_caps(sk, &rt->dst);
 120
 121         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
 122                                                     inet->inet_daddr,
 123                                                     inet->inet_sport,
 124                                                     inet->inet_dport);
 125         inet->inet_id = dp->dccps_iss ^ jiffies;
 126
 127         err = dccp_connect(sk);
 128         rt = NULL;
 129         if (err != 0)
 130                 goto failure;
 131 out:
 132         return err;
 133 failure:
 134         /*
 135          * This unhashes the socket and releases the local port, if necessary.
 136          */
 137         dccp_set_state(sk, DCCP_CLOSED);
 138         ip_rt_put(rt);
 139         sk->sk_route_caps = 0;
 140         inet->inet_dport = 0;
 141         goto out;
 142 }
 143
 144 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 145
 146 /*
 147  * This routine does path mtu discovery as defined in RFC1191.
 148  */
 149 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 150                                           const struct iphdr *iph,
 151                                           u32 mtu)
 152 {
 153         struct dst_entry *dst;
 154         const struct inet_sock *inet = inet_sk(sk);
 155         const struct dccp_sock *dp = dccp_sk(sk);
 156
 157         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 158          * send out by Linux are always < 576bytes so they should go through
 159          * unfragmented).
 160          */
 161         if (sk->sk_state == DCCP_LISTEN)
 162                 return;
 163
 164         dst = inet_csk_update_pmtu(sk, mtu);
 165         if (!dst)
 166                 return;
 167
 168         /* Something is about to be wrong... Remember soft error
 169          * for the case, if this connection will not able to recover.
 170          */
 171         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 172                 sk->sk_err_soft = EMSGSIZE;
 173
 174         mtu = dst_mtu(dst);
 175
 176         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 177             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 178                 dccp_sync_mss(sk, mtu);
 179
 180                 /*
 181                  * From RFC 4340, sec. 14.1:
 182                  *
 183                  *      DCCP-Sync packets are the best choice for upward
 184                  *      probing, since DCCP-Sync probes do not risk application
 185                  *      data loss.
 186                  */
 187                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 188         } /* else let the usual retransmit timer handle it */
 189 }
 190
 191 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
 192 {
 193         struct dst_entry *dst = __sk_dst_check(sk, 0);
 194
 195         if (dst)
 196                 dst->ops->redirect(dst, sk, skb);
 197 }
 198
 199 /*
 200  * This routine is called by the ICMP module when it gets some sort of error
 201  * condition. If err < 0 then the socket should be closed and the error
 202  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 203  * After adjustment header points to the first 8 bytes of the tcp header. We
 204  * need to find the appropriate port.
 205  *
 206  * The locking strategy used here is very "optimistic". When someone else
 207  * accesses the socket the ICMP is just dropped and for some paths there is no
 208  * check at all. A more general error queue to queue errors for later handling
 209  * is probably better.
 210  */
 211 static void dccp_v4_err(struct sk_buff *skb, u32 info)
 212 {
 213         const struct iphdr *iph = (struct iphdr *)skb->data;
 214         const u8 offset = iph->ihl << 2;
 215         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
 216         struct dccp_sock *dp;
 217         struct inet_sock *inet;
 218         const int type = icmp_hdr(skb)->type;
 219         const int code = icmp_hdr(skb)->code;
 220         struct sock *sk;
 221         __u64 seq;
 222         int err;
 223         struct net *net = dev_net(skb->dev);
 224
 225         if (skb->len < offset + sizeof(*dh) ||
 226             skb->len < offset + __dccp_basic_hdr_len(dh)) {
 227                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 228                 return;
 229         }
 230
 231         sk = inet_lookup(net, &dccp_hashinfo,
 232                         iph->daddr, dh->dccph_dport,
 233                         iph->saddr, dh->dccph_sport, inet_iif(skb));
 234         if (sk == NULL) {
 235                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 236                 return;
 237         }
 238
 239         if (sk->sk_state == DCCP_TIME_WAIT) {
 240                 inet_twsk_put(inet_twsk(sk));
 241                 return;
 242         }
 243
 244         bh_lock_sock(sk);
 245         /* If too many ICMPs get dropped on busy
 246          * servers this needs to be solved differently.
 247          */
 248         if (sock_owned_by_user(sk))
 249                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 250
 251         if (sk->sk_state == DCCP_CLOSED)
 252                 goto out;
 253
 254         dp = dccp_sk(sk);
 255         seq = dccp_hdr_seq(dh);
 256         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 257             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 258                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 259                 goto out;
 260         }
 261
 262         switch (type) {
 263         case ICMP_REDIRECT:
 264                 dccp_do_redirect(skb, sk);
 265                 goto out;
 266         case ICMP_SOURCE_QUENCH:
 267                 /* Just silently ignore these. */
 268                 goto out;
 269         case ICMP_PARAMETERPROB:
 270                 err = EPROTO;
 271                 break;
 272         case ICMP_DEST_UNREACH:
 273                 if (code > NR_ICMP_UNREACH)
 274                         goto out;
 275
 276                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 277                         if (!sock_owned_by_user(sk))
 278                                 dccp_do_pmtu_discovery(sk, iph, info);
 279                         goto out;
 280                 }
 281
 282                 err = icmp_err_convert[code].errno;
 283                 break;
 284         case ICMP_TIME_EXCEEDED:
 285                 err = EHOSTUNREACH;
 286                 break;
 287         default:
 288                 goto out;
 289         }
 290
 291         switch (sk->sk_state) {
 292                 struct request_sock *req , **prev;
 293         case DCCP_LISTEN:
 294                 if (sock_owned_by_user(sk))
 295                         goto out;
 296                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 297                                           iph->daddr, iph->saddr);
 298                 if (!req)
 299                         goto out;
 300
 301                 /*
 302                  * ICMPs are not backlogged, hence we cannot get an established
 303                  * socket here.
 304                  */
 305                 WARN_ON(req->sk);
 306
 307                 if (!between48(seq, dccp_rsk(req)->dreq_iss,
 308                                     dccp_rsk(req)->dreq_gss)) {
 309                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 310                         goto out;
 311                 }
 312                 /*
 313                  * Still in RESPOND, just remove it silently.
 314                  * There is no good way to pass the error to the newly
 315                  * created socket, and POSIX does not want network
 316                  * errors returned from accept().
 317                  */
 318                 inet_csk_reqsk_queue_drop(sk, req, prev);
 319                 goto out;
 320
 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
 379 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 380
 381 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 382 {
 383         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 384                                            ip_hdr(skb)->saddr,
 385                                            dccp_hdr(skb)->dccph_dport,
 386                                            dccp_hdr(skb)->dccph_sport);
 387 }
 388
 389 /*
 390  * The three way handshake has completed - we got a valid ACK or DATAACK -
 391  * now create the new socket.
 392  *
 393  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 394  */
 395 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 396                                        struct request_sock *req,
 397                                        struct dst_entry *dst)
 398 {
 399         struct inet_request_sock *ireq;
 400         struct inet_sock *newinet;
 401         struct sock *newsk;
 402
 403         if (sk_acceptq_is_full(sk))
 404                 goto exit_overflow;
 405
 406         newsk = dccp_create_openreq_child(sk, req, skb);
 407         if (newsk == NULL)
 408                 goto exit_nonewsk;
 409
 410         newinet            = inet_sk(newsk);
 411         ireq               = inet_rsk(req);
 412         newinet->inet_daddr     = ireq->ir_rmt_addr;
 413         newinet->inet_rcv_saddr = ireq->ir_loc_addr;
 414         newinet->inet_saddr     = ireq->ir_loc_addr;
 415         newinet->inet_opt       = ireq->opt;
 416         ireq->opt          = NULL;
 417         newinet->mc_index  = inet_iif(skb);
 418         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 419         newinet->inet_id   = jiffies;
 420
 421         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
 422                 goto put_and_exit;
 423
 424         sk_setup_caps(newsk, dst);
 425
 426         dccp_sync_mss(newsk, dst_mtu(dst));
 427
 428         if (__inet_inherit_port(sk, newsk) < 0)
 429                 goto put_and_exit;
 430         __inet_hash_nolisten(newsk, NULL);
 431
 432         return newsk;
 433
 434 exit_overflow:
 435         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 436 exit_nonewsk:
 437         dst_release(dst);
 438 exit:
 439         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 440         return NULL;
 441 put_and_exit:
 442         inet_csk_prepare_forced_close(newsk);
 443         dccp_done(newsk);
 444         goto exit;
 445 }
 446
 447 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 448
 449 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 450 {
 451         const struct dccp_hdr *dh = dccp_hdr(skb);
 452         const struct iphdr *iph = ip_hdr(skb);
 453         struct sock *nsk;
 454         struct request_sock **prev;
 455         /* Find possible connection requests. */
 456         struct request_sock *req = inet_csk_search_req(sk, &prev,
 457                                                        dh->dccph_sport,
 458                                                        iph->saddr, iph->daddr);
 459         if (req != NULL)
 460                 return dccp_check_req(sk, skb, req, prev);
 461
 462         nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
 463                                       iph->saddr, dh->dccph_sport,
 464                                       iph->daddr, dh->dccph_dport,
 465                                       inet_iif(skb));
 466         if (nsk != NULL) {
 467                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 468                         bh_lock_sock(nsk);
 469                         return nsk;
 470                 }
 471                 inet_twsk_put(inet_twsk(nsk));
 472                 return NULL;
 473         }
 474
 475         return sk;
 476 }
 477
 478 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 479                                            struct sk_buff *skb)
 480 {
 481         struct rtable *rt;
 482         const struct iphdr *iph = ip_hdr(skb);
 483         struct flowi4 fl4 = {
 484                 .flowi4_oif = inet_iif(skb),
 485                 .daddr = iph->saddr,
 486                 .saddr = iph->daddr,
 487                 .flowi4_tos = RT_CONN_FLAGS(sk),
 488                 .flowi4_proto = sk->sk_protocol,
 489                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
 490                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
 491         };
 492
 493         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 494         rt = ip_route_output_flow(net, &fl4, sk);
 495         if (IS_ERR(rt)) {
 496                 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 497                 return NULL;
 498         }
 499
 500         return &rt->dst;
 501 }
 502
 503 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req)
 504 {
 505         int err = -1;
 506         struct sk_buff *skb;
 507         struct dst_entry *dst;
 508         struct flowi4 fl4;
 509
 510         dst = inet_csk_route_req(sk, &fl4, req);
 511         if (dst == NULL)
 512                 goto out;
 513
 514         skb = dccp_make_response(sk, dst, req);
 515         if (skb != NULL) {
 516                 const struct inet_request_sock *ireq = inet_rsk(req);
 517                 struct dccp_hdr *dh = dccp_hdr(skb);
 518
 519                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
 520                                                               ireq->ir_rmt_addr);
 521                 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 522                                             ireq->ir_rmt_addr,
 523                                             ireq->opt);
 524                 err = net_xmit_eval(err);
 525         }
 526
 527 out:
 528         dst_release(dst);
 529         return err;
 530 }
 531
 532 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
 533 {
 534         int err;
 535         const struct iphdr *rxiph;
 536         struct sk_buff *skb;
 537         struct dst_entry *dst;
 538         struct net *net = dev_net(skb_dst(rxskb)->dev);
 539         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
 540
 541         /* Never send a reset in response to a reset. */
 542         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 543                 return;
 544
 545         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 546                 return;
 547
 548         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 549         if (dst == NULL)
 550                 return;
 551
 552         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 553         if (skb == NULL)
 554                 goto out;
 555
 556         rxiph = ip_hdr(rxskb);
 557         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 558                                                                  rxiph->daddr);
 559         skb_dst_set(skb, dst_clone(dst));
 560
 561         bh_lock_sock(ctl_sk);
 562         err = ip_build_and_send_pkt(skb, ctl_sk,
 563                                     rxiph->daddr, rxiph->saddr, NULL);
 564         bh_unlock_sock(ctl_sk);
 565
 566         if (net_xmit_eval(err) == 0) {
 567                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 568                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 569         }
 570 out:
 571          dst_release(dst);
 572 }
 573
 574 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 575 {
 576         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 577         kfree(inet_rsk(req)->opt);
 578 }
 579
 580 void dccp_syn_ack_timeout(struct sock *sk, struct request_sock *req)
 581 {
 582 }
 583 EXPORT_SYMBOL(dccp_syn_ack_timeout);
 584
 585 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 586         .family         = PF_INET,
 587         .obj_size       = sizeof(struct dccp_request_sock),
 588         .rtx_syn_ack    = dccp_v4_send_response,
 589         .send_ack       = dccp_reqsk_send_ack,
 590         .destructor     = dccp_v4_reqsk_destructor,
 591         .send_reset     = dccp_v4_ctl_send_reset,
 592         .syn_ack_timeout = dccp_syn_ack_timeout,
 593 };
 594
 595 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 596 {
 597         struct inet_request_sock *ireq;
 598         struct request_sock *req;
 599         struct dccp_request_sock *dreq;
 600         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 601         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 602
 603         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 604         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 605                 return 0;       /* discard, don't send a reset here */
 606
 607         if (dccp_bad_service_code(sk, service)) {
 608                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 609                 goto drop;
 610         }
 611         /*
 612          * TW buckets are converted to open requests without
 613          * limitations, they conserve resources and peer is
 614          * evidently real one.
 615          */
 616         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 617         if (inet_csk_reqsk_queue_is_full(sk))
 618                 goto drop;
 619
 620         /*
 621          * Accept backlog is full. If we have already queued enough
 622          * of warm entries in syn queue, drop request. It is better than
 623          * clogging syn queue with openreqs with exponentially increasing
 624          * timeout.
 625          */
 626         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 627                 goto drop;
 628
 629         req = inet_reqsk_alloc(&dccp_request_sock_ops);
 630         if (req == NULL)
 631                 goto drop;
 632
 633         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 634                 goto drop_and_free;
 635
 636         dreq = dccp_rsk(req);
 637         if (dccp_parse_options(sk, dreq, skb))
 638                 goto drop_and_free;
 639
 640         if (security_inet_conn_request(sk, skb, req))
 641                 goto drop_and_free;
 642
 643         ireq = inet_rsk(req);
 644         ireq->ir_loc_addr = ip_hdr(skb)->daddr;
 645         ireq->ir_rmt_addr = ip_hdr(skb)->saddr;
 646
 647         /*
 648          * Step 3: Process LISTEN state
 649          *
 650          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 651          *
 652          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 653          */
 654         dreq->dreq_isr     = dcb->dccpd_seq;
 655         dreq->dreq_gsr     = dreq->dreq_isr;
 656         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 657         dreq->dreq_gss     = dreq->dreq_iss;
 658         dreq->dreq_service = service;
 659
 660         if (dccp_v4_send_response(sk, req))
 661                 goto drop_and_free;
 662
 663         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 664         return 0;
 665
 666 drop_and_free:
 667         reqsk_free(req);
 668 drop:
 669         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 670         return -1;
 671 }
 672
 673 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 674
 675 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 676 {
 677         struct dccp_hdr *dh = dccp_hdr(skb);
 678
 679         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 680                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 681                         goto reset;
 682                 return 0;
 683         }
 684
 685         /*
 686          *  Step 3: Process LISTEN state
 687          *       If P.type == Request or P contains a valid Init Cookie option,
 688          *            (* Must scan the packet's options to check for Init
 689          *               Cookies.  Only Init Cookies are processed here,
 690          *               however; other options are processed in Step 8.  This
 691          *               scan need only be performed if the endpoint uses Init
 692          *               Cookies *)
 693          *            (* Generate a new socket and switch to that socket *)
 694          *            Set S := new socket for this port pair
 695          *            S.state = RESPOND
 696          *            Choose S.ISS (initial seqno) or set from Init Cookies
 697          *            Initialize S.GAR := S.ISS
 698          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 699          *            Continue with S.state == RESPOND
 700          *            (* A Response packet will be generated in Step 11 *)
 701          *       Otherwise,
 702          *            Generate Reset(No Connection) unless P.type == Reset
 703          *            Drop packet and return
 704          *
 705          * NOTE: the check for the packet types is done in
 706          *       dccp_rcv_state_process
 707          */
 708         if (sk->sk_state == DCCP_LISTEN) {
 709                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 710
 711                 if (nsk == NULL)
 712                         goto discard;
 713
 714                 if (nsk != sk) {
 715                         if (dccp_child_process(sk, nsk, skb))
 716                                 goto reset;
 717                         return 0;
 718                 }
 719         }
 720
 721         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 722                 goto reset;
 723         return 0;
 724
 725 reset:
 726         dccp_v4_ctl_send_reset(sk, skb);
 727 discard:
 728         kfree_skb(skb);
 729         return 0;
 730 }
 731
 732 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 733
 734 /**
 735  *      dccp_invalid_packet  -  check for malformed packets
 736  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 737  *      Packets that fail these checks are ignored and do not receive Resets.
 738  */
 739 int dccp_invalid_packet(struct sk_buff *skb)
 740 {
 741         const struct dccp_hdr *dh;
 742         unsigned int cscov;
 743
 744         if (skb->pkt_type != PACKET_HOST)
 745                 return 1;
 746
 747         /* If the packet is shorter than 12 bytes, drop packet and return */
 748         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 749                 DCCP_WARN("pskb_may_pull failed\n");
 750                 return 1;
 751         }
 752
 753         dh = dccp_hdr(skb);
 754
 755         /* If P.type is not understood, drop packet and return */
 756         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 757                 DCCP_WARN("invalid packet type\n");
 758                 return 1;
 759         }
 760
 761         /*
 762          * If P.Data Offset is too small for packet type, drop packet and return
 763          */
 764         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 765                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
 766                 return 1;
 767         }
 768         /*
 769          * If P.Data Offset is too too large for packet, drop packet and return
 770          */
 771         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 772                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
 773                 return 1;
 774         }
 775
 776         /*
 777          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 778          * has short sequence numbers), drop packet and return
 779          */
 780         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 781             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 782                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 783                           dccp_packet_name(dh->dccph_type));
 784                 return 1;
 785         }
 786
 787         /*
 788          * If P.CsCov is too large for the packet size, drop packet and return.
 789          * This must come _before_ checksumming (not as RFC 4340 suggests).
 790          */
 791         cscov = dccp_csum_coverage(skb);
 792         if (cscov > skb->len) {
 793                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 794                           dh->dccph_cscov, skb->len);
 795                 return 1;
 796         }
 797
 798         /* If header checksum is incorrect, drop packet and return.
 799          * (This step is completed in the AF-dependent functions.) */
 800         skb->csum = skb_checksum(skb, 0, cscov, 0);
 801
 802         return 0;
 803 }
 804
 805 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 806
 807 /* this is called when real data arrives */
 808 static int dccp_v4_rcv(struct sk_buff *skb)
 809 {
 810         const struct dccp_hdr *dh;
 811         const struct iphdr *iph;
 812         struct sock *sk;
 813         int min_cov;
 814
 815         /* Step 1: Check header basics */
 816
 817         if (dccp_invalid_packet(skb))
 818                 goto discard_it;
 819
 820         iph = ip_hdr(skb);
 821         /* Step 1: If header checksum is incorrect, drop packet and return */
 822         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 823                 DCCP_WARN("dropped packet with invalid checksum\n");
 824                 goto discard_it;
 825         }
 826
 827         dh = dccp_hdr(skb);
 828
 829         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 830         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 831
 832         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 833                       dccp_packet_name(dh->dccph_type),
 834                       &iph->saddr, ntohs(dh->dccph_sport),
 835                       &iph->daddr, ntohs(dh->dccph_dport),
 836                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 837
 838         if (dccp_packet_without_ack(skb)) {
 839                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 840                 dccp_pr_debug_cat("\n");
 841         } else {
 842                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 843                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 844                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 845         }
 846
 847         /* Step 2:
 848          *      Look up flow ID in table and get corresponding socket */
 849         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
 850                                dh->dccph_sport, dh->dccph_dport);
 851         /*
 852          * Step 2:
 853          *      If no socket ...
 854          */
 855         if (sk == NULL) {
 856                 dccp_pr_debug("failed to look up flow ID in table and "
 857                               "get corresponding socket\n");
 858                 goto no_dccp_socket;
 859         }
 860
 861         /*
 862          * Step 2:
 863          *      ... or S.state == TIMEWAIT,
 864          *              Generate Reset(No Connection) unless P.type == Reset
 865          *              Drop packet and return
 866          */
 867         if (sk->sk_state == DCCP_TIME_WAIT) {
 868                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 869                 inet_twsk_put(inet_twsk(sk));
 870                 goto no_dccp_socket;
 871         }
 872
 873         /*
 874          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 875          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 876          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 877          */
 878         min_cov = dccp_sk(sk)->dccps_pcrlen;
 879         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 880                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 881                               dh->dccph_cscov, min_cov);
 882                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 883                  *         options (Section 11.7) with Drop Code 0, Protocol
 884                  *         Constraints."                                     */
 885                 goto discard_and_relse;
 886         }
 887
 888         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 889                 goto discard_and_relse;
 890         nf_reset(skb);
 891
 892         return sk_receive_skb(sk, skb, 1);
 893
 894 no_dccp_socket:
 895         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 896                 goto discard_it;
 897         /*
 898          * Step 2:
 899          *      If no socket ...
 900          *              Generate Reset(No Connection) unless P.type == Reset
 901          *              Drop packet and return
 902          */
 903         if (dh->dccph_type != DCCP_PKT_RESET) {
 904                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 905                                         DCCP_RESET_CODE_NO_CONNECTION;
 906                 dccp_v4_ctl_send_reset(sk, skb);
 907         }
 908
 909 discard_it:
 910         kfree_skb(skb);
 911         return 0;
 912
 913 discard_and_relse:
 914         sock_put(sk);
 915         goto discard_it;
 916 }
 917
 918 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 919         .queue_xmit        = ip_queue_xmit,
 920         .send_check        = dccp_v4_send_check,
 921         .rebuild_header    = inet_sk_rebuild_header,
 922         .conn_request      = dccp_v4_conn_request,
 923         .syn_recv_sock     = dccp_v4_request_recv_sock,
 924         .net_header_len    = sizeof(struct iphdr),
 925         .setsockopt        = ip_setsockopt,
 926         .getsockopt        = ip_getsockopt,
 927         .addr2sockaddr     = inet_csk_addr2sockaddr,
 928         .sockaddr_len      = sizeof(struct sockaddr_in),
 929         .bind_conflict     = inet_csk_bind_conflict,
 930 #ifdef CONFIG_COMPAT
 931         .compat_setsockopt = compat_ip_setsockopt,
 932         .compat_getsockopt = compat_ip_getsockopt,
 933 #endif
 934 };
 935
 936 static int dccp_v4_init_sock(struct sock *sk)
 937 {
 938         static __u8 dccp_v4_ctl_sock_initialized;
 939         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 940
 941         if (err == 0) {
 942                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 943                         dccp_v4_ctl_sock_initialized = 1;
 944                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 945         }
 946
 947         return err;
 948 }
 949
 950 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 951         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 952 };
 953
 954 static struct proto dccp_v4_prot = {
 955         .name                   = "DCCP",
 956         .owner                  = THIS_MODULE,
 957         .close                  = dccp_close,
 958         .connect                = dccp_v4_connect,
 959         .disconnect             = dccp_disconnect,
 960         .ioctl                  = dccp_ioctl,
 961         .init                   = dccp_v4_init_sock,
 962         .setsockopt             = dccp_setsockopt,
 963         .getsockopt             = dccp_getsockopt,
 964         .sendmsg                = dccp_sendmsg,
 965         .recvmsg                = dccp_recvmsg,
 966         .backlog_rcv            = dccp_v4_do_rcv,
 967         .hash                   = inet_hash,
 968         .unhash                 = inet_unhash,
 969         .accept                 = inet_csk_accept,
 970         .get_port               = inet_csk_get_port,
 971         .shutdown               = dccp_shutdown,
 972         .destroy                = dccp_destroy_sock,
 973         .orphan_count           = &dccp_orphan_count,
 974         .max_header             = MAX_DCCP_HEADER,
 975         .obj_size               = sizeof(struct dccp_sock),
 976         .slab_flags             = SLAB_DESTROY_BY_RCU,
 977         .rsk_prot               = &dccp_request_sock_ops,
 978         .twsk_prot              = &dccp_timewait_sock_ops,
 979         .h.hashinfo             = &dccp_hashinfo,
 980 #ifdef CONFIG_COMPAT
 981         .compat_setsockopt      = compat_dccp_setsockopt,
 982         .compat_getsockopt      = compat_dccp_getsockopt,
 983 #endif
 984 };
 985
 986 static const struct net_protocol dccp_v4_protocol = {
 987         .handler        = dccp_v4_rcv,
 988         .err_handler    = dccp_v4_err,
 989         .no_policy      = 1,
 990         .netns_ok       = 1,
 991 };
 992
 993 static const struct proto_ops inet_dccp_ops = {
 994         .family            = PF_INET,
 995         .owner             = THIS_MODULE,
 996         .release           = inet_release,
 997         .bind              = inet_bind,
 998         .connect           = inet_stream_connect,
 999         .socketpair        = sock_no_socketpair,
1000         .accept            = inet_accept,
1001         .getname           = inet_getname,
1002         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1003         .poll              = dccp_poll,
1004         .ioctl             = inet_ioctl,
1005         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1006         .listen            = inet_dccp_listen,
1007         .shutdown          = inet_shutdown,
1008         .setsockopt        = sock_common_setsockopt,
1009         .getsockopt        = sock_common_getsockopt,
1010         .sendmsg           = inet_sendmsg,
1011         .recvmsg           = sock_common_recvmsg,
1012         .mmap              = sock_no_mmap,
1013         .sendpage          = sock_no_sendpage,
1014 #ifdef CONFIG_COMPAT
1015         .compat_setsockopt = compat_sock_common_setsockopt,
1016         .compat_getsockopt = compat_sock_common_getsockopt,
1017 #endif
1018 };
1019
1020 static struct inet_protosw dccp_v4_protosw = {
1021         .type           = SOCK_DCCP,
1022         .protocol       = IPPROTO_DCCP,
1023         .prot           = &dccp_v4_prot,
1024         .ops            = &inet_dccp_ops,
1025         .no_check       = 0,
1026         .flags          = INET_PROTOSW_ICSK,
1027 };
1028
1029 static int __net_init dccp_v4_init_net(struct net *net)
1030 {
1031         if (dccp_hashinfo.bhash == NULL)
1032                 return -ESOCKTNOSUPPORT;
1033
1034         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1035                                     SOCK_DCCP, IPPROTO_DCCP, net);
1036 }
1037
1038 static void __net_exit dccp_v4_exit_net(struct net *net)
1039 {
1040         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1041 }
1042
1043 static struct pernet_operations dccp_v4_ops = {
1044         .init   = dccp_v4_init_net,
1045         .exit   = dccp_v4_exit_net,
1046 };
1047
1048 static int __init dccp_v4_init(void)
1049 {
1050         int err = proto_register(&dccp_v4_prot, 1);
1051
1052         if (err != 0)
1053                 goto out;
1054
1055         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1056         if (err != 0)
1057                 goto out_proto_unregister;
1058
1059         inet_register_protosw(&dccp_v4_protosw);
1060
1061         err = register_pernet_subsys(&dccp_v4_ops);
1062         if (err)
1063                 goto out_destroy_ctl_sock;
1064 out:
1065         return err;
1066 out_destroy_ctl_sock:
1067         inet_unregister_protosw(&dccp_v4_protosw);
1068         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1069 out_proto_unregister:
1070         proto_unregister(&dccp_v4_prot);
1071         goto out;
1072 }
1073
1074 static void __exit dccp_v4_exit(void)
1075 {
1076         unregister_pernet_subsys(&dccp_v4_ops);
1077         inet_unregister_protosw(&dccp_v4_protosw);
1078         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1079         proto_unregister(&dccp_v4_prot);
1080 }
1081
1082 module_init(dccp_v4_init);
1083 module_exit(dccp_v4_exit);
1084
1085 /*
1086  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1087  * values directly, Also cover the case where the protocol is not specified,
1088  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1089  */
1090 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1091 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1092 MODULE_LICENSE("GPL");
1093 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1094 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");