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         /* We don't check in the destentry if pmtu discovery is forbidden
 165          * on this route. We just assume that no packet_to_big packets
 166          * are send back when pmtu discovery is not active.
 167          * There is a small race when the user changes this flag in the
 168          * route, but I think that's acceptable.
 169          */
 170         if ((dst = __sk_dst_check(sk, 0)) == NULL)
 171                 return;
 172
 173         dst->ops->update_pmtu(dst, mtu);
 174
 175         /* Something is about to be wrong... Remember soft error
 176          * for the case, if this connection will not able to recover.
 177          */
 178         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 179                 sk->sk_err_soft = EMSGSIZE;
 180
 181         mtu = dst_mtu(dst);
 182
 183         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 184             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 185                 dccp_sync_mss(sk, mtu);
 186
 187                 /*
 188                  * From RFC 4340, sec. 14.1:
 189                  *
 190                  *      DCCP-Sync packets are the best choice for upward
 191                  *      probing, since DCCP-Sync probes do not risk application
 192                  *      data loss.
 193                  */
 194                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 195         } /* else let the usual retransmit timer handle it */
 196 }
 197
 198 /*
 199  * This routine is called by the ICMP module when it gets some sort of error
 200  * condition. If err < 0 then the socket should be closed and the error
 201  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 202  * After adjustment header points to the first 8 bytes of the tcp header. We
 203  * need to find the appropriate port.
 204  *
 205  * The locking strategy used here is very "optimistic". When someone else
 206  * accesses the socket the ICMP is just dropped and for some paths there is no
 207  * check at all. A more general error queue to queue errors for later handling
 208  * is probably better.
 209  */
 210 static void dccp_v4_err(struct sk_buff *skb, u32 info)
 211 {
 212         const struct iphdr *iph = (struct iphdr *)skb->data;
 213         const u8 offset = iph->ihl << 2;
 214         const struct dccp_hdr *dh = (struct dccp_hdr *)(skb->data + offset);
 215         struct dccp_sock *dp;
 216         struct inet_sock *inet;
 217         const int type = icmp_hdr(skb)->type;
 218         const int code = icmp_hdr(skb)->code;
 219         struct sock *sk;
 220         __u64 seq;
 221         int err;
 222         struct net *net = dev_net(skb->dev);
 223
 224         if (skb->len < offset + sizeof(*dh) ||
 225             skb->len < offset + __dccp_basic_hdr_len(dh)) {
 226                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 227                 return;
 228         }
 229
 230         sk = inet_lookup(net, &dccp_hashinfo,
 231                         iph->daddr, dh->dccph_dport,
 232                         iph->saddr, dh->dccph_sport, inet_iif(skb));
 233         if (sk == NULL) {
 234                 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 235                 return;
 236         }
 237
 238         if (sk->sk_state == DCCP_TIME_WAIT) {
 239                 inet_twsk_put(inet_twsk(sk));
 240                 return;
 241         }
 242
 243         bh_lock_sock(sk);
 244         /* If too many ICMPs get dropped on busy
 245          * servers this needs to be solved differently.
 246          */
 247         if (sock_owned_by_user(sk))
 248                 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 249
 250         if (sk->sk_state == DCCP_CLOSED)
 251                 goto out;
 252
 253         dp = dccp_sk(sk);
 254         seq = dccp_hdr_seq(dh);
 255         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 256             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 257                 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 258                 goto out;
 259         }
 260
 261         switch (type) {
 262         case ICMP_SOURCE_QUENCH:
 263                 /* Just silently ignore these. */
 264                 goto out;
 265         case ICMP_PARAMETERPROB:
 266                 err = EPROTO;
 267                 break;
 268         case ICMP_DEST_UNREACH:
 269                 if (code > NR_ICMP_UNREACH)
 270                         goto out;
 271
 272                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 273                         if (!sock_owned_by_user(sk))
 274                                 dccp_do_pmtu_discovery(sk, iph, info);
 275                         goto out;
 276                 }
 277
 278                 err = icmp_err_convert[code].errno;
 279                 break;
 280         case ICMP_TIME_EXCEEDED:
 281                 err = EHOSTUNREACH;
 282                 break;
 283         default:
 284                 goto out;
 285         }
 286
 287         switch (sk->sk_state) {
 288                 struct request_sock *req , **prev;
 289         case DCCP_LISTEN:
 290                 if (sock_owned_by_user(sk))
 291                         goto out;
 292                 req = inet_csk_search_req(sk, &prev, dh->dccph_dport,
 293                                           iph->daddr, iph->saddr);
 294                 if (!req)
 295                         goto out;
 296
 297                 /*
 298                  * ICMPs are not backlogged, hence we cannot get an established
 299                  * socket here.
 300                  */
 301                 WARN_ON(req->sk);
 302
 303                 if (seq != dccp_rsk(req)->dreq_iss) {
 304                         NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 305                         goto out;
 306                 }
 307                 /*
 308                  * Still in RESPOND, just remove it silently.
 309                  * There is no good way to pass the error to the newly
 310                  * created socket, and POSIX does not want network
 311                  * errors returned from accept().
 312                  */
 313                 inet_csk_reqsk_queue_drop(sk, req, prev);
 314                 goto out;
 315
 316         case DCCP_REQUESTING:
 317         case DCCP_RESPOND:
 318                 if (!sock_owned_by_user(sk)) {
 319                         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 320                         sk->sk_err = err;
 321
 322                         sk->sk_error_report(sk);
 323
 324                         dccp_done(sk);
 325                 } else
 326                         sk->sk_err_soft = err;
 327                 goto out;
 328         }
 329
 330         /* If we've already connected we will keep trying
 331          * until we time out, or the user gives up.
 332          *
 333          * rfc1122 4.2.3.9 allows to consider as hard errors
 334          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 335          * but it is obsoleted by pmtu discovery).
 336          *
 337          * Note, that in modern internet, where routing is unreliable
 338          * and in each dark corner broken firewalls sit, sending random
 339          * errors ordered by their masters even this two messages finally lose
 340          * their original sense (even Linux sends invalid PORT_UNREACHs)
 341          *
 342          * Now we are in compliance with RFCs.
 343          *                                                      --ANK (980905)
 344          */
 345
 346         inet = inet_sk(sk);
 347         if (!sock_owned_by_user(sk) && inet->recverr) {
 348                 sk->sk_err = err;
 349                 sk->sk_error_report(sk);
 350         } else /* Only an error on timeout */
 351                 sk->sk_err_soft = err;
 352 out:
 353         bh_unlock_sock(sk);
 354         sock_put(sk);
 355 }
 356
 357 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 358                                       __be32 src, __be32 dst)
 359 {
 360         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 361 }
 362
 363 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 364 {
 365         const struct inet_sock *inet = inet_sk(sk);
 366         struct dccp_hdr *dh = dccp_hdr(skb);
 367
 368         dccp_csum_outgoing(skb);
 369         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 370                                                  inet->inet_saddr,
 371                                                  inet->inet_daddr);
 372 }
 373
 374 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 375
 376 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 377 {
 378         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 379                                            ip_hdr(skb)->saddr,
 380                                            dccp_hdr(skb)->dccph_dport,
 381                                            dccp_hdr(skb)->dccph_sport);
 382 }
 383
 384 /*
 385  * The three way handshake has completed - we got a valid ACK or DATAACK -
 386  * now create the new socket.
 387  *
 388  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 389  */
 390 struct sock *dccp_v4_request_recv_sock(struct sock *sk, struct sk_buff *skb,
 391                                        struct request_sock *req,
 392                                        struct dst_entry *dst)
 393 {
 394         struct inet_request_sock *ireq;
 395         struct inet_sock *newinet;
 396         struct sock *newsk;
 397
 398         if (sk_acceptq_is_full(sk))
 399                 goto exit_overflow;
 400
 401         newsk = dccp_create_openreq_child(sk, req, skb);
 402         if (newsk == NULL)
 403                 goto exit_nonewsk;
 404
 405         newinet            = inet_sk(newsk);
 406         ireq               = inet_rsk(req);
 407         newinet->inet_daddr     = ireq->rmt_addr;
 408         newinet->inet_rcv_saddr = ireq->loc_addr;
 409         newinet->inet_saddr     = ireq->loc_addr;
 410         newinet->inet_opt       = ireq->opt;
 411         ireq->opt          = NULL;
 412         newinet->mc_index  = inet_iif(skb);
 413         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 414         newinet->inet_id   = jiffies;
 415
 416         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
 417                 goto put_and_exit;
 418
 419         sk_setup_caps(newsk, dst);
 420
 421         dccp_sync_mss(newsk, dst_mtu(dst));
 422
 423         if (__inet_inherit_port(sk, newsk) < 0)
 424                 goto put_and_exit;
 425         __inet_hash_nolisten(newsk, NULL);
 426
 427         return newsk;
 428
 429 exit_overflow:
 430         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 431 exit_nonewsk:
 432         dst_release(dst);
 433 exit:
 434         NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 435         return NULL;
 436 put_and_exit:
 437         bh_unlock_sock(newsk);
 438         sock_put(newsk);
 439         goto exit;
 440 }
 441
 442 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 443
 444 static struct sock *dccp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
 445 {
 446         const struct dccp_hdr *dh = dccp_hdr(skb);
 447         const struct iphdr *iph = ip_hdr(skb);
 448         struct sock *nsk;
 449         struct request_sock **prev;
 450         /* Find possible connection requests. */
 451         struct request_sock *req = inet_csk_search_req(sk, &prev,
 452                                                        dh->dccph_sport,
 453                                                        iph->saddr, iph->daddr);
 454         if (req != NULL)
 455                 return dccp_check_req(sk, skb, req, prev);
 456
 457         nsk = inet_lookup_established(sock_net(sk), &dccp_hashinfo,
 458                                       iph->saddr, dh->dccph_sport,
 459                                       iph->daddr, dh->dccph_dport,
 460                                       inet_iif(skb));
 461         if (nsk != NULL) {
 462                 if (nsk->sk_state != DCCP_TIME_WAIT) {
 463                         bh_lock_sock(nsk);
 464                         return nsk;
 465                 }
 466                 inet_twsk_put(inet_twsk(nsk));
 467                 return NULL;
 468         }
 469
 470         return sk;
 471 }
 472
 473 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 474                                            struct sk_buff *skb)
 475 {
 476         struct rtable *rt;
 477         const struct iphdr *iph = ip_hdr(skb);
 478         struct flowi4 fl4 = {
 479                 .flowi4_oif = skb_rtable(skb)->rt_iif,
 480                 .daddr = iph->saddr,
 481                 .saddr = iph->daddr,
 482                 .flowi4_tos = RT_CONN_FLAGS(sk),
 483                 .flowi4_proto = sk->sk_protocol,
 484                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
 485                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
 486         };
 487
 488         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 489         rt = ip_route_output_flow(net, &fl4, sk);
 490         if (IS_ERR(rt)) {
 491                 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 492                 return NULL;
 493         }
 494
 495         return &rt->dst;
 496 }
 497
 498 static int dccp_v4_send_response(struct sock *sk, struct request_sock *req,
 499                                  struct request_values *rv_unused)
 500 {
 501         int err = -1;
 502         struct sk_buff *skb;
 503         struct dst_entry *dst;
 504         struct flowi4 fl4;
 505
 506         dst = inet_csk_route_req(sk, &fl4, req);
 507         if (dst == NULL)
 508                 goto out;
 509
 510         skb = dccp_make_response(sk, dst, req);
 511         if (skb != NULL) {
 512                 const struct inet_request_sock *ireq = inet_rsk(req);
 513                 struct dccp_hdr *dh = dccp_hdr(skb);
 514
 515                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->loc_addr,
 516                                                               ireq->rmt_addr);
 517                 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
 518                                             ireq->rmt_addr,
 519                                             ireq->opt);
 520                 err = net_xmit_eval(err);
 521         }
 522
 523 out:
 524         dst_release(dst);
 525         return err;
 526 }
 527
 528 static void dccp_v4_ctl_send_reset(struct sock *sk, struct sk_buff *rxskb)
 529 {
 530         int err;
 531         const struct iphdr *rxiph;
 532         struct sk_buff *skb;
 533         struct dst_entry *dst;
 534         struct net *net = dev_net(skb_dst(rxskb)->dev);
 535         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
 536
 537         /* Never send a reset in response to a reset. */
 538         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 539                 return;
 540
 541         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 542                 return;
 543
 544         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 545         if (dst == NULL)
 546                 return;
 547
 548         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 549         if (skb == NULL)
 550                 goto out;
 551
 552         rxiph = ip_hdr(rxskb);
 553         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 554                                                                  rxiph->daddr);
 555         skb_dst_set(skb, dst_clone(dst));
 556
 557         bh_lock_sock(ctl_sk);
 558         err = ip_build_and_send_pkt(skb, ctl_sk,
 559                                     rxiph->daddr, rxiph->saddr, NULL);
 560         bh_unlock_sock(ctl_sk);
 561
 562         if (net_xmit_eval(err) == 0) {
 563                 DCCP_INC_STATS_BH(DCCP_MIB_OUTSEGS);
 564                 DCCP_INC_STATS_BH(DCCP_MIB_OUTRSTS);
 565         }
 566 out:
 567          dst_release(dst);
 568 }
 569
 570 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 571 {
 572         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 573         kfree(inet_rsk(req)->opt);
 574 }
 575
 576 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 577         .family         = PF_INET,
 578         .obj_size       = sizeof(struct dccp_request_sock),
 579         .rtx_syn_ack    = dccp_v4_send_response,
 580         .send_ack       = dccp_reqsk_send_ack,
 581         .destructor     = dccp_v4_reqsk_destructor,
 582         .send_reset     = dccp_v4_ctl_send_reset,
 583 };
 584
 585 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 586 {
 587         struct inet_request_sock *ireq;
 588         struct request_sock *req;
 589         struct dccp_request_sock *dreq;
 590         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 591         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 592
 593         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 594         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 595                 return 0;       /* discard, don't send a reset here */
 596
 597         if (dccp_bad_service_code(sk, service)) {
 598                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 599                 goto drop;
 600         }
 601         /*
 602          * TW buckets are converted to open requests without
 603          * limitations, they conserve resources and peer is
 604          * evidently real one.
 605          */
 606         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 607         if (inet_csk_reqsk_queue_is_full(sk))
 608                 goto drop;
 609
 610         /*
 611          * Accept backlog is full. If we have already queued enough
 612          * of warm entries in syn queue, drop request. It is better than
 613          * clogging syn queue with openreqs with exponentially increasing
 614          * timeout.
 615          */
 616         if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
 617                 goto drop;
 618
 619         req = inet_reqsk_alloc(&dccp_request_sock_ops);
 620         if (req == NULL)
 621                 goto drop;
 622
 623         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 624                 goto drop_and_free;
 625
 626         dreq = dccp_rsk(req);
 627         if (dccp_parse_options(sk, dreq, skb))
 628                 goto drop_and_free;
 629
 630         if (security_inet_conn_request(sk, skb, req))
 631                 goto drop_and_free;
 632
 633         ireq = inet_rsk(req);
 634         ireq->loc_addr = ip_hdr(skb)->daddr;
 635         ireq->rmt_addr = ip_hdr(skb)->saddr;
 636
 637         /*
 638          * Step 3: Process LISTEN state
 639          *
 640          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 641          *
 642          * In fact we defer setting S.GSR, S.SWL, S.SWH to
 643          * dccp_create_openreq_child.
 644          */
 645         dreq->dreq_isr     = dcb->dccpd_seq;
 646         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 647         dreq->dreq_service = service;
 648
 649         if (dccp_v4_send_response(sk, req, NULL))
 650                 goto drop_and_free;
 651
 652         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 653         return 0;
 654
 655 drop_and_free:
 656         reqsk_free(req);
 657 drop:
 658         DCCP_INC_STATS_BH(DCCP_MIB_ATTEMPTFAILS);
 659         return -1;
 660 }
 661
 662 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 663
 664 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 665 {
 666         struct dccp_hdr *dh = dccp_hdr(skb);
 667
 668         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 669                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 670                         goto reset;
 671                 return 0;
 672         }
 673
 674         /*
 675          *  Step 3: Process LISTEN state
 676          *       If P.type == Request or P contains a valid Init Cookie option,
 677          *            (* Must scan the packet's options to check for Init
 678          *               Cookies.  Only Init Cookies are processed here,
 679          *               however; other options are processed in Step 8.  This
 680          *               scan need only be performed if the endpoint uses Init
 681          *               Cookies *)
 682          *            (* Generate a new socket and switch to that socket *)
 683          *            Set S := new socket for this port pair
 684          *            S.state = RESPOND
 685          *            Choose S.ISS (initial seqno) or set from Init Cookies
 686          *            Initialize S.GAR := S.ISS
 687          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 688          *            Continue with S.state == RESPOND
 689          *            (* A Response packet will be generated in Step 11 *)
 690          *       Otherwise,
 691          *            Generate Reset(No Connection) unless P.type == Reset
 692          *            Drop packet and return
 693          *
 694          * NOTE: the check for the packet types is done in
 695          *       dccp_rcv_state_process
 696          */
 697         if (sk->sk_state == DCCP_LISTEN) {
 698                 struct sock *nsk = dccp_v4_hnd_req(sk, skb);
 699
 700                 if (nsk == NULL)
 701                         goto discard;
 702
 703                 if (nsk != sk) {
 704                         if (dccp_child_process(sk, nsk, skb))
 705                                 goto reset;
 706                         return 0;
 707                 }
 708         }
 709
 710         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 711                 goto reset;
 712         return 0;
 713
 714 reset:
 715         dccp_v4_ctl_send_reset(sk, skb);
 716 discard:
 717         kfree_skb(skb);
 718         return 0;
 719 }
 720
 721 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 722
 723 /**
 724  *      dccp_invalid_packet  -  check for malformed packets
 725  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 726  *      Packets that fail these checks are ignored and do not receive Resets.
 727  */
 728 int dccp_invalid_packet(struct sk_buff *skb)
 729 {
 730         const struct dccp_hdr *dh;
 731         unsigned int cscov;
 732
 733         if (skb->pkt_type != PACKET_HOST)
 734                 return 1;
 735
 736         /* If the packet is shorter than 12 bytes, drop packet and return */
 737         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 738                 DCCP_WARN("pskb_may_pull failed\n");
 739                 return 1;
 740         }
 741
 742         dh = dccp_hdr(skb);
 743
 744         /* If P.type is not understood, drop packet and return */
 745         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 746                 DCCP_WARN("invalid packet type\n");
 747                 return 1;
 748         }
 749
 750         /*
 751          * If P.Data Offset is too small for packet type, drop packet and return
 752          */
 753         if (dh->dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 754                 DCCP_WARN("P.Data Offset(%u) too small\n", dh->dccph_doff);
 755                 return 1;
 756         }
 757         /*
 758          * If P.Data Offset is too too large for packet, drop packet and return
 759          */
 760         if (!pskb_may_pull(skb, dh->dccph_doff * sizeof(u32))) {
 761                 DCCP_WARN("P.Data Offset(%u) too large\n", dh->dccph_doff);
 762                 return 1;
 763         }
 764
 765         /*
 766          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 767          * has short sequence numbers), drop packet and return
 768          */
 769         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 770             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 771                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 772                           dccp_packet_name(dh->dccph_type));
 773                 return 1;
 774         }
 775
 776         /*
 777          * If P.CsCov is too large for the packet size, drop packet and return.
 778          * This must come _before_ checksumming (not as RFC 4340 suggests).
 779          */
 780         cscov = dccp_csum_coverage(skb);
 781         if (cscov > skb->len) {
 782                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 783                           dh->dccph_cscov, skb->len);
 784                 return 1;
 785         }
 786
 787         /* If header checksum is incorrect, drop packet and return.
 788          * (This step is completed in the AF-dependent functions.) */
 789         skb->csum = skb_checksum(skb, 0, cscov, 0);
 790
 791         return 0;
 792 }
 793
 794 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 795
 796 /* this is called when real data arrives */
 797 static int dccp_v4_rcv(struct sk_buff *skb)
 798 {
 799         const struct dccp_hdr *dh;
 800         const struct iphdr *iph;
 801         struct sock *sk;
 802         int min_cov;
 803
 804         /* Step 1: Check header basics */
 805
 806         if (dccp_invalid_packet(skb))
 807                 goto discard_it;
 808
 809         iph = ip_hdr(skb);
 810         /* Step 1: If header checksum is incorrect, drop packet and return */
 811         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 812                 DCCP_WARN("dropped packet with invalid checksum\n");
 813                 goto discard_it;
 814         }
 815
 816         dh = dccp_hdr(skb);
 817
 818         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 819         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 820
 821         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 822                       dccp_packet_name(dh->dccph_type),
 823                       &iph->saddr, ntohs(dh->dccph_sport),
 824                       &iph->daddr, ntohs(dh->dccph_dport),
 825                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 826
 827         if (dccp_packet_without_ack(skb)) {
 828                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 829                 dccp_pr_debug_cat("\n");
 830         } else {
 831                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 832                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 833                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 834         }
 835
 836         /* Step 2:
 837          *      Look up flow ID in table and get corresponding socket */
 838         sk = __inet_lookup_skb(&dccp_hashinfo, skb,
 839                                dh->dccph_sport, dh->dccph_dport);
 840         /*
 841          * Step 2:
 842          *      If no socket ...
 843          */
 844         if (sk == NULL) {
 845                 dccp_pr_debug("failed to look up flow ID in table and "
 846                               "get corresponding socket\n");
 847                 goto no_dccp_socket;
 848         }
 849
 850         /*
 851          * Step 2:
 852          *      ... or S.state == TIMEWAIT,
 853          *              Generate Reset(No Connection) unless P.type == Reset
 854          *              Drop packet and return
 855          */
 856         if (sk->sk_state == DCCP_TIME_WAIT) {
 857                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 858                 inet_twsk_put(inet_twsk(sk));
 859                 goto no_dccp_socket;
 860         }
 861
 862         /*
 863          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 864          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 865          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 866          */
 867         min_cov = dccp_sk(sk)->dccps_pcrlen;
 868         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 869                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 870                               dh->dccph_cscov, min_cov);
 871                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 872                  *         options (Section 11.7) with Drop Code 0, Protocol
 873                  *         Constraints."                                     */
 874                 goto discard_and_relse;
 875         }
 876
 877         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 878                 goto discard_and_relse;
 879         nf_reset(skb);
 880
 881         return sk_receive_skb(sk, skb, 1);
 882
 883 no_dccp_socket:
 884         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 885                 goto discard_it;
 886         /*
 887          * Step 2:
 888          *      If no socket ...
 889          *              Generate Reset(No Connection) unless P.type == Reset
 890          *              Drop packet and return
 891          */
 892         if (dh->dccph_type != DCCP_PKT_RESET) {
 893                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 894                                         DCCP_RESET_CODE_NO_CONNECTION;
 895                 dccp_v4_ctl_send_reset(sk, skb);
 896         }
 897
 898 discard_it:
 899         kfree_skb(skb);
 900         return 0;
 901
 902 discard_and_relse:
 903         sock_put(sk);
 904         goto discard_it;
 905 }
 906
 907 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 908         .queue_xmit        = ip_queue_xmit,
 909         .send_check        = dccp_v4_send_check,
 910         .rebuild_header    = inet_sk_rebuild_header,
 911         .conn_request      = dccp_v4_conn_request,
 912         .syn_recv_sock     = dccp_v4_request_recv_sock,
 913         .net_header_len    = sizeof(struct iphdr),
 914         .setsockopt        = ip_setsockopt,
 915         .getsockopt        = ip_getsockopt,
 916         .addr2sockaddr     = inet_csk_addr2sockaddr,
 917         .sockaddr_len      = sizeof(struct sockaddr_in),
 918         .bind_conflict     = inet_csk_bind_conflict,
 919 #ifdef CONFIG_COMPAT
 920         .compat_setsockopt = compat_ip_setsockopt,
 921         .compat_getsockopt = compat_ip_getsockopt,
 922 #endif
 923 };
 924
 925 static int dccp_v4_init_sock(struct sock *sk)
 926 {
 927         static __u8 dccp_v4_ctl_sock_initialized;
 928         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 929
 930         if (err == 0) {
 931                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 932                         dccp_v4_ctl_sock_initialized = 1;
 933                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 934         }
 935
 936         return err;
 937 }
 938
 939 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 940         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 941 };
 942
 943 static struct proto dccp_v4_prot = {
 944         .name                   = "DCCP",
 945         .owner                  = THIS_MODULE,
 946         .close                  = dccp_close,
 947         .connect                = dccp_v4_connect,
 948         .disconnect             = dccp_disconnect,
 949         .ioctl                  = dccp_ioctl,
 950         .init                   = dccp_v4_init_sock,
 951         .setsockopt             = dccp_setsockopt,
 952         .getsockopt             = dccp_getsockopt,
 953         .sendmsg                = dccp_sendmsg,
 954         .recvmsg                = dccp_recvmsg,
 955         .backlog_rcv            = dccp_v4_do_rcv,
 956         .hash                   = inet_hash,
 957         .unhash                 = inet_unhash,
 958         .accept                 = inet_csk_accept,
 959         .get_port               = inet_csk_get_port,
 960         .shutdown               = dccp_shutdown,
 961         .destroy                = dccp_destroy_sock,
 962         .orphan_count           = &dccp_orphan_count,
 963         .max_header             = MAX_DCCP_HEADER,
 964         .obj_size               = sizeof(struct dccp_sock),
 965         .slab_flags             = SLAB_DESTROY_BY_RCU,
 966         .rsk_prot               = &dccp_request_sock_ops,
 967         .twsk_prot              = &dccp_timewait_sock_ops,
 968         .h.hashinfo             = &dccp_hashinfo,
 969 #ifdef CONFIG_COMPAT
 970         .compat_setsockopt      = compat_dccp_setsockopt,
 971         .compat_getsockopt      = compat_dccp_getsockopt,
 972 #endif
 973 };
 974
 975 static const struct net_protocol dccp_v4_protocol = {
 976         .handler        = dccp_v4_rcv,
 977         .err_handler    = dccp_v4_err,
 978         .no_policy      = 1,
 979         .netns_ok       = 1,
 980 };
 981
 982 static const struct proto_ops inet_dccp_ops = {
 983         .family            = PF_INET,
 984         .owner             = THIS_MODULE,
 985         .release           = inet_release,
 986         .bind              = inet_bind,
 987         .connect           = inet_stream_connect,
 988         .socketpair        = sock_no_socketpair,
 989         .accept            = inet_accept,
 990         .getname           = inet_getname,
 991         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
 992         .poll              = dccp_poll,
 993         .ioctl             = inet_ioctl,
 994         /* FIXME: work on inet_listen to rename it to sock_common_listen */
 995         .listen            = inet_dccp_listen,
 996         .shutdown          = inet_shutdown,
 997         .setsockopt        = sock_common_setsockopt,
 998         .getsockopt        = sock_common_getsockopt,
 999         .sendmsg           = inet_sendmsg,
1000         .recvmsg           = sock_common_recvmsg,
1001         .mmap              = sock_no_mmap,
1002         .sendpage          = sock_no_sendpage,
1003 #ifdef CONFIG_COMPAT
1004         .compat_setsockopt = compat_sock_common_setsockopt,
1005         .compat_getsockopt = compat_sock_common_getsockopt,
1006 #endif
1007 };
1008
1009 static struct inet_protosw dccp_v4_protosw = {
1010         .type           = SOCK_DCCP,
1011         .protocol       = IPPROTO_DCCP,
1012         .prot           = &dccp_v4_prot,
1013         .ops            = &inet_dccp_ops,
1014         .no_check       = 0,
1015         .flags          = INET_PROTOSW_ICSK,
1016 };
1017
1018 static int __net_init dccp_v4_init_net(struct net *net)
1019 {
1020         if (dccp_hashinfo.bhash == NULL)
1021                 return -ESOCKTNOSUPPORT;
1022
1023         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1024                                     SOCK_DCCP, IPPROTO_DCCP, net);
1025 }
1026
1027 static void __net_exit dccp_v4_exit_net(struct net *net)
1028 {
1029         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1030 }
1031
1032 static struct pernet_operations dccp_v4_ops = {
1033         .init   = dccp_v4_init_net,
1034         .exit   = dccp_v4_exit_net,
1035 };
1036
1037 static int __init dccp_v4_init(void)
1038 {
1039         int err = proto_register(&dccp_v4_prot, 1);
1040
1041         if (err != 0)
1042                 goto out;
1043
1044         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1045         if (err != 0)
1046                 goto out_proto_unregister;
1047
1048         inet_register_protosw(&dccp_v4_protosw);
1049
1050         err = register_pernet_subsys(&dccp_v4_ops);
1051         if (err)
1052                 goto out_destroy_ctl_sock;
1053 out:
1054         return err;
1055 out_destroy_ctl_sock:
1056         inet_unregister_protosw(&dccp_v4_protosw);
1057         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1058 out_proto_unregister:
1059         proto_unregister(&dccp_v4_prot);
1060         goto out;
1061 }
1062
1063 static void __exit dccp_v4_exit(void)
1064 {
1065         unregister_pernet_subsys(&dccp_v4_ops);
1066         inet_unregister_protosw(&dccp_v4_protosw);
1067         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1068         proto_unregister(&dccp_v4_prot);
1069 }
1070
1071 module_init(dccp_v4_init);
1072 module_exit(dccp_v4_exit);
1073
1074 /*
1075  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1076  * values directly, Also cover the case where the protocol is not specified,
1077  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1078  */
1079 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1080 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1081 MODULE_LICENSE("GPL");
1082 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1083 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");