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 = prandom_u32();
 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         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
 138                 inet_reset_saddr(sk);
 139         ip_rt_put(rt);
 140         sk->sk_route_caps = 0;
 141         inet->inet_dport = 0;
 142         goto out;
 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             ip_sk_accept_pmtu(sk) &&
 178             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 179                 dccp_sync_mss(sk, mtu);
 180
 181                 /*
 182                  * From RFC 4340, sec. 14.1:
 183                  *
 184                  *      DCCP-Sync packets are the best choice for upward
 185                  *      probing, since DCCP-Sync probes do not risk application
 186                  *      data loss.
 187                  */
 188                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 189         } /* else let the usual retransmit timer handle it */
 190 }
 191
 192 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
 193 {
 194         struct dst_entry *dst = __sk_dst_check(sk, 0);
 195
 196         if (dst)
 197                 dst->ops->redirect(dst, sk, skb);
 198 }
 199
 200 void dccp_req_err(struct sock *sk, u64 seq)
 201         {
 202         struct request_sock *req = inet_reqsk(sk);
 203         struct net *net = sock_net(sk);
 204
 205         /*
 206          * ICMPs are not backlogged, hence we cannot get an established
 207          * socket here.
 208          */
 209         if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
 210                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 211         } else {
 212                 /*
 213                  * Still in RESPOND, just remove it silently.
 214                  * There is no good way to pass the error to the newly
 215                  * created socket, and POSIX does not want network
 216                  * errors returned from accept().
 217                  */
 218                 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
 219         }
 220         reqsk_put(req);
 221 }
 222 EXPORT_SYMBOL(dccp_req_err);
 223
 224 /*
 225  * This routine is called by the ICMP module when it gets some sort of error
 226  * condition. If err < 0 then the socket should be closed and the error
 227  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 228  * After adjustment header points to the first 8 bytes of the tcp header. We
 229  * need to find the appropriate port.
 230  *
 231  * The locking strategy used here is very "optimistic". When someone else
 232  * accesses the socket the ICMP is just dropped and for some paths there is no
 233  * check at all. A more general error queue to queue errors for later handling
 234  * is probably better.
 235  */
 236 static void dccp_v4_err(struct sk_buff *skb, u32 info)
 237 {
 238         const struct iphdr *iph = (struct iphdr *)skb->data;
 239         const u8 offset = iph->ihl << 2;
 240         const struct dccp_hdr *dh;
 241         struct dccp_sock *dp;
 242         struct inet_sock *inet;
 243         const int type = icmp_hdr(skb)->type;
 244         const int code = icmp_hdr(skb)->code;
 245         struct sock *sk;
 246         __u64 seq;
 247         int err;
 248         struct net *net = dev_net(skb->dev);
 249
 250         /* Only need dccph_dport & dccph_sport which are the first
 251          * 4 bytes in dccp header.
 252          * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us.
 253          */
 254         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_sport) > 8);
 255         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_dport) > 8);
 256         dh = (struct dccp_hdr *)(skb->data + offset);
 257
 258         sk = __inet_lookup_established(net, &dccp_hashinfo,
 259                                        iph->daddr, dh->dccph_dport,
 260                                        iph->saddr, ntohs(dh->dccph_sport),
 261                                        inet_iif(skb), 0);
 262         if (!sk) {
 263                 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 264                 return;
 265         }
 266
 267         if (sk->sk_state == DCCP_TIME_WAIT) {
 268                 inet_twsk_put(inet_twsk(sk));
 269                 return;
 270         }
 271         seq = dccp_hdr_seq(dh);
 272         if (sk->sk_state == DCCP_NEW_SYN_RECV)
 273                 return dccp_req_err(sk, seq);
 274
 275         bh_lock_sock(sk);
 276         /* If too many ICMPs get dropped on busy
 277          * servers this needs to be solved differently.
 278          */
 279         if (sock_owned_by_user(sk))
 280                 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
 281
 282         if (sk->sk_state == DCCP_CLOSED)
 283                 goto out;
 284
 285         dp = dccp_sk(sk);
 286         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 287             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 288                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 289                 goto out;
 290         }
 291
 292         switch (type) {
 293         case ICMP_REDIRECT:
 294                 if (!sock_owned_by_user(sk))
 295                         dccp_do_redirect(skb, sk);
 296                 goto out;
 297         case ICMP_SOURCE_QUENCH:
 298                 /* Just silently ignore these. */
 299                 goto out;
 300         case ICMP_PARAMETERPROB:
 301                 err = EPROTO;
 302                 break;
 303         case ICMP_DEST_UNREACH:
 304                 if (code > NR_ICMP_UNREACH)
 305                         goto out;
 306
 307                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 308                         if (!sock_owned_by_user(sk))
 309                                 dccp_do_pmtu_discovery(sk, iph, info);
 310                         goto out;
 311                 }
 312
 313                 err = icmp_err_convert[code].errno;
 314                 break;
 315         case ICMP_TIME_EXCEEDED:
 316                 err = EHOSTUNREACH;
 317                 break;
 318         default:
 319                 goto out;
 320         }
 321
 322         switch (sk->sk_state) {
 323         case DCCP_REQUESTING:
 324         case DCCP_RESPOND:
 325                 if (!sock_owned_by_user(sk)) {
 326                         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 327                         sk->sk_err = err;
 328
 329                         sk->sk_error_report(sk);
 330
 331                         dccp_done(sk);
 332                 } else
 333                         sk->sk_err_soft = err;
 334                 goto out;
 335         }
 336
 337         /* If we've already connected we will keep trying
 338          * until we time out, or the user gives up.
 339          *
 340          * rfc1122 4.2.3.9 allows to consider as hard errors
 341          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 342          * but it is obsoleted by pmtu discovery).
 343          *
 344          * Note, that in modern internet, where routing is unreliable
 345          * and in each dark corner broken firewalls sit, sending random
 346          * errors ordered by their masters even this two messages finally lose
 347          * their original sense (even Linux sends invalid PORT_UNREACHs)
 348          *
 349          * Now we are in compliance with RFCs.
 350          *                                                      --ANK (980905)
 351          */
 352
 353         inet = inet_sk(sk);
 354         if (!sock_owned_by_user(sk) && inet->recverr) {
 355                 sk->sk_err = err;
 356                 sk->sk_error_report(sk);
 357         } else /* Only an error on timeout */
 358                 sk->sk_err_soft = err;
 359 out:
 360         bh_unlock_sock(sk);
 361         sock_put(sk);
 362 }
 363
 364 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 365                                       __be32 src, __be32 dst)
 366 {
 367         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 368 }
 369
 370 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 371 {
 372         const struct inet_sock *inet = inet_sk(sk);
 373         struct dccp_hdr *dh = dccp_hdr(skb);
 374
 375         dccp_csum_outgoing(skb);
 376         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 377                                                  inet->inet_saddr,
 378                                                  inet->inet_daddr);
 379 }
 380 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 381
 382 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 383 {
 384         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 385                                            ip_hdr(skb)->saddr,
 386                                            dccp_hdr(skb)->dccph_dport,
 387                                            dccp_hdr(skb)->dccph_sport);
 388 }
 389
 390 /*
 391  * The three way handshake has completed - we got a valid ACK or DATAACK -
 392  * now create the new socket.
 393  *
 394  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 395  */
 396 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
 397                                        struct sk_buff *skb,
 398                                        struct request_sock *req,
 399                                        struct dst_entry *dst,
 400                                        struct request_sock *req_unhash,
 401                                        bool *own_req)
 402 {
 403         struct inet_request_sock *ireq;
 404         struct inet_sock *newinet;
 405         struct sock *newsk;
 406
 407         if (sk_acceptq_is_full(sk))
 408                 goto exit_overflow;
 409
 410         newsk = dccp_create_openreq_child(sk, req, skb);
 411         if (newsk == NULL)
 412                 goto exit_nonewsk;
 413
 414         newinet            = inet_sk(newsk);
 415         ireq               = inet_rsk(req);
 416         sk_daddr_set(newsk, ireq->ir_rmt_addr);
 417         sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
 418         newinet->inet_saddr     = ireq->ir_loc_addr;
 419         RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt));
 420         newinet->mc_index  = inet_iif(skb);
 421         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 422         newinet->inet_id   = prandom_u32();
 423
 424         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
 425                 goto put_and_exit;
 426
 427         sk_setup_caps(newsk, dst);
 428
 429         dccp_sync_mss(newsk, dst_mtu(dst));
 430
 431         if (__inet_inherit_port(sk, newsk) < 0)
 432                 goto put_and_exit;
 433         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), NULL);
 434         if (*own_req)
 435                 ireq->ireq_opt = NULL;
 436         else
 437                 newinet->inet_opt = NULL;
 438         return newsk;
 439
 440 exit_overflow:
 441         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 442 exit_nonewsk:
 443         dst_release(dst);
 444 exit:
 445         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
 446         return NULL;
 447 put_and_exit:
 448         newinet->inet_opt = NULL;
 449         inet_csk_prepare_forced_close(newsk);
 450         dccp_done(newsk);
 451         goto exit;
 452 }
 453 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 454
 455 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 456                                            struct sk_buff *skb)
 457 {
 458         struct rtable *rt;
 459         const struct iphdr *iph = ip_hdr(skb);
 460         struct flowi4 fl4 = {
 461                 .flowi4_oif = inet_iif(skb),
 462                 .daddr = iph->saddr,
 463                 .saddr = iph->daddr,
 464                 .flowi4_tos = RT_CONN_FLAGS(sk),
 465                 .flowi4_proto = sk->sk_protocol,
 466                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
 467                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
 468         };
 469
 470         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 471         rt = ip_route_output_flow(net, &fl4, sk);
 472         if (IS_ERR(rt)) {
 473                 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 474                 return NULL;
 475         }
 476
 477         return &rt->dst;
 478 }
 479
 480 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
 481 {
 482         int err = -1;
 483         struct sk_buff *skb;
 484         struct dst_entry *dst;
 485         struct flowi4 fl4;
 486
 487         dst = inet_csk_route_req(sk, &fl4, req);
 488         if (dst == NULL)
 489                 goto out;
 490
 491         skb = dccp_make_response(sk, dst, req);
 492         if (skb != NULL) {
 493                 const struct inet_request_sock *ireq = inet_rsk(req);
 494                 struct dccp_hdr *dh = dccp_hdr(skb);
 495
 496                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
 497                                                               ireq->ir_rmt_addr);
 498                 rcu_read_lock();
 499                 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 500                                             ireq->ir_rmt_addr,
 501                                             rcu_dereference(ireq->ireq_opt));
 502                 rcu_read_unlock();
 503                 err = net_xmit_eval(err);
 504         }
 505
 506 out:
 507         dst_release(dst);
 508         return err;
 509 }
 510
 511 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
 512 {
 513         int err;
 514         const struct iphdr *rxiph;
 515         struct sk_buff *skb;
 516         struct dst_entry *dst;
 517         struct net *net = dev_net(skb_dst(rxskb)->dev);
 518         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
 519
 520         /* Never send a reset in response to a reset. */
 521         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 522                 return;
 523
 524         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 525                 return;
 526
 527         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 528         if (dst == NULL)
 529                 return;
 530
 531         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 532         if (skb == NULL)
 533                 goto out;
 534
 535         rxiph = ip_hdr(rxskb);
 536         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 537                                                                  rxiph->daddr);
 538         skb_dst_set(skb, dst_clone(dst));
 539
 540         local_bh_disable();
 541         bh_lock_sock(ctl_sk);
 542         err = ip_build_and_send_pkt(skb, ctl_sk,
 543                                     rxiph->daddr, rxiph->saddr, NULL);
 544         bh_unlock_sock(ctl_sk);
 545
 546         if (net_xmit_eval(err) == 0) {
 547                 __DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 548                 __DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
 549         }
 550         local_bh_enable();
 551 out:
 552         dst_release(dst);
 553 }
 554
 555 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 556 {
 557         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 558         kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
 559 }
 560
 561 void dccp_syn_ack_timeout(const struct request_sock *req)
 562 {
 563 }
 564 EXPORT_SYMBOL(dccp_syn_ack_timeout);
 565
 566 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 567         .family         = PF_INET,
 568         .obj_size       = sizeof(struct dccp_request_sock),
 569         .rtx_syn_ack    = dccp_v4_send_response,
 570         .send_ack       = dccp_reqsk_send_ack,
 571         .destructor     = dccp_v4_reqsk_destructor,
 572         .send_reset     = dccp_v4_ctl_send_reset,
 573         .syn_ack_timeout = dccp_syn_ack_timeout,
 574 };
 575
 576 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 577 {
 578         struct inet_request_sock *ireq;
 579         struct request_sock *req;
 580         struct dccp_request_sock *dreq;
 581         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 582         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 583
 584         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 585         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 586                 return 0;       /* discard, don't send a reset here */
 587
 588         if (dccp_bad_service_code(sk, service)) {
 589                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 590                 goto drop;
 591         }
 592         /*
 593          * TW buckets are converted to open requests without
 594          * limitations, they conserve resources and peer is
 595          * evidently real one.
 596          */
 597         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 598         if (inet_csk_reqsk_queue_is_full(sk))
 599                 goto drop;
 600
 601         if (sk_acceptq_is_full(sk))
 602                 goto drop;
 603
 604         req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
 605         if (req == NULL)
 606                 goto drop;
 607
 608         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 609                 goto drop_and_free;
 610
 611         dreq = dccp_rsk(req);
 612         if (dccp_parse_options(sk, dreq, skb))
 613                 goto drop_and_free;
 614
 615         if (security_inet_conn_request(sk, skb, req))
 616                 goto drop_and_free;
 617
 618         ireq = inet_rsk(req);
 619         sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
 620         sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
 621         ireq->ir_mark = inet_request_mark(sk, skb);
 622         ireq->ireq_family = AF_INET;
 623         ireq->ir_iif = sk->sk_bound_dev_if;
 624
 625         /*
 626          * Step 3: Process LISTEN state
 627          *
 628          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 629          *
 630          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 631          */
 632         dreq->dreq_isr     = dcb->dccpd_seq;
 633         dreq->dreq_gsr     = dreq->dreq_isr;
 634         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 635         dreq->dreq_gss     = dreq->dreq_iss;
 636         dreq->dreq_service = service;
 637
 638         if (dccp_v4_send_response(sk, req))
 639                 goto drop_and_free;
 640
 641         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 642         reqsk_put(req);
 643         return 0;
 644
 645 drop_and_free:
 646         reqsk_free(req);
 647 drop:
 648         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 649         return -1;
 650 }
 651 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 652
 653 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 654 {
 655         struct dccp_hdr *dh = dccp_hdr(skb);
 656
 657         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 658                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 659                         goto reset;
 660                 return 0;
 661         }
 662
 663         /*
 664          *  Step 3: Process LISTEN state
 665          *       If P.type == Request or P contains a valid Init Cookie option,
 666          *            (* Must scan the packet's options to check for Init
 667          *               Cookies.  Only Init Cookies are processed here,
 668          *               however; other options are processed in Step 8.  This
 669          *               scan need only be performed if the endpoint uses Init
 670          *               Cookies *)
 671          *            (* Generate a new socket and switch to that socket *)
 672          *            Set S := new socket for this port pair
 673          *            S.state = RESPOND
 674          *            Choose S.ISS (initial seqno) or set from Init Cookies
 675          *            Initialize S.GAR := S.ISS
 676          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 677          *            Continue with S.state == RESPOND
 678          *            (* A Response packet will be generated in Step 11 *)
 679          *       Otherwise,
 680          *            Generate Reset(No Connection) unless P.type == Reset
 681          *            Drop packet and return
 682          *
 683          * NOTE: the check for the packet types is done in
 684          *       dccp_rcv_state_process
 685          */
 686
 687         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 688                 goto reset;
 689         return 0;
 690
 691 reset:
 692         dccp_v4_ctl_send_reset(sk, skb);
 693         kfree_skb(skb);
 694         return 0;
 695 }
 696 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 697
 698 /**
 699  *      dccp_invalid_packet  -  check for malformed packets
 700  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 701  *      Packets that fail these checks are ignored and do not receive Resets.
 702  */
 703 int dccp_invalid_packet(struct sk_buff *skb)
 704 {
 705         const struct dccp_hdr *dh;
 706         unsigned int cscov;
 707         u8 dccph_doff;
 708
 709         if (skb->pkt_type != PACKET_HOST)
 710                 return 1;
 711
 712         /* If the packet is shorter than 12 bytes, drop packet and return */
 713         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 714                 DCCP_WARN("pskb_may_pull failed\n");
 715                 return 1;
 716         }
 717
 718         dh = dccp_hdr(skb);
 719
 720         /* If P.type is not understood, drop packet and return */
 721         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 722                 DCCP_WARN("invalid packet type\n");
 723                 return 1;
 724         }
 725
 726         /*
 727          * If P.Data Offset is too small for packet type, drop packet and return
 728          */
 729         dccph_doff = dh->dccph_doff;
 730         if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 731                 DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
 732                 return 1;
 733         }
 734         /*
 735          * If P.Data Offset is too too large for packet, drop packet and return
 736          */
 737         if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
 738                 DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
 739                 return 1;
 740         }
 741         dh = dccp_hdr(skb);
 742         /*
 743          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 744          * has short sequence numbers), drop packet and return
 745          */
 746         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 747             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 748                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 749                           dccp_packet_name(dh->dccph_type));
 750                 return 1;
 751         }
 752
 753         /*
 754          * If P.CsCov is too large for the packet size, drop packet and return.
 755          * This must come _before_ checksumming (not as RFC 4340 suggests).
 756          */
 757         cscov = dccp_csum_coverage(skb);
 758         if (cscov > skb->len) {
 759                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 760                           dh->dccph_cscov, skb->len);
 761                 return 1;
 762         }
 763
 764         /* If header checksum is incorrect, drop packet and return.
 765          * (This step is completed in the AF-dependent functions.) */
 766         skb->csum = skb_checksum(skb, 0, cscov, 0);
 767
 768         return 0;
 769 }
 770 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 771
 772 /* this is called when real data arrives */
 773 static int dccp_v4_rcv(struct sk_buff *skb)
 774 {
 775         const struct dccp_hdr *dh;
 776         const struct iphdr *iph;
 777         bool refcounted;
 778         struct sock *sk;
 779         int min_cov;
 780
 781         /* Step 1: Check header basics */
 782
 783         if (dccp_invalid_packet(skb))
 784                 goto discard_it;
 785
 786         iph = ip_hdr(skb);
 787         /* Step 1: If header checksum is incorrect, drop packet and return */
 788         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 789                 DCCP_WARN("dropped packet with invalid checksum\n");
 790                 goto discard_it;
 791         }
 792
 793         dh = dccp_hdr(skb);
 794
 795         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 796         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 797
 798         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 799                       dccp_packet_name(dh->dccph_type),
 800                       &iph->saddr, ntohs(dh->dccph_sport),
 801                       &iph->daddr, ntohs(dh->dccph_dport),
 802                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 803
 804         if (dccp_packet_without_ack(skb)) {
 805                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 806                 dccp_pr_debug_cat("\n");
 807         } else {
 808                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 809                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 810                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 811         }
 812
 813 lookup:
 814         sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
 815                                dh->dccph_sport, dh->dccph_dport, 0, &refcounted);
 816         if (!sk) {
 817                 dccp_pr_debug("failed to look up flow ID in table and "
 818                               "get corresponding socket\n");
 819                 goto no_dccp_socket;
 820         }
 821
 822         /*
 823          * Step 2:
 824          *      ... or S.state == TIMEWAIT,
 825          *              Generate Reset(No Connection) unless P.type == Reset
 826          *              Drop packet and return
 827          */
 828         if (sk->sk_state == DCCP_TIME_WAIT) {
 829                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 830                 inet_twsk_put(inet_twsk(sk));
 831                 goto no_dccp_socket;
 832         }
 833
 834         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 835                 struct request_sock *req = inet_reqsk(sk);
 836                 struct sock *nsk;
 837
 838                 sk = req->rsk_listener;
 839                 if (unlikely(sk->sk_state != DCCP_LISTEN)) {
 840                         inet_csk_reqsk_queue_drop_and_put(sk, req);
 841                         goto lookup;
 842                 }
 843                 sock_hold(sk);
 844                 refcounted = true;
 845                 nsk = dccp_check_req(sk, skb, req);
 846                 if (!nsk) {
 847                         reqsk_put(req);
 848                         goto discard_and_relse;
 849                 }
 850                 if (nsk == sk) {
 851                         reqsk_put(req);
 852                 } else if (dccp_child_process(sk, nsk, skb)) {
 853                         dccp_v4_ctl_send_reset(sk, skb);
 854                         goto discard_and_relse;
 855                 } else {
 856                         sock_put(sk);
 857                         return 0;
 858                 }
 859         }
 860         /*
 861          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 862          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 863          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 864          */
 865         min_cov = dccp_sk(sk)->dccps_pcrlen;
 866         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 867                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 868                               dh->dccph_cscov, min_cov);
 869                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 870                  *         options (Section 11.7) with Drop Code 0, Protocol
 871                  *         Constraints."                                     */
 872                 goto discard_and_relse;
 873         }
 874
 875         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 876                 goto discard_and_relse;
 877         nf_reset(skb);
 878
 879         return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
 880
 881 no_dccp_socket:
 882         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 883                 goto discard_it;
 884         /*
 885          * Step 2:
 886          *      If no socket ...
 887          *              Generate Reset(No Connection) unless P.type == Reset
 888          *              Drop packet and return
 889          */
 890         if (dh->dccph_type != DCCP_PKT_RESET) {
 891                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 892                                         DCCP_RESET_CODE_NO_CONNECTION;
 893                 dccp_v4_ctl_send_reset(sk, skb);
 894         }
 895
 896 discard_it:
 897         kfree_skb(skb);
 898         return 0;
 899
 900 discard_and_relse:
 901         if (refcounted)
 902                 sock_put(sk);
 903         goto discard_it;
 904 }
 905
 906 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 907         .queue_xmit        = ip_queue_xmit,
 908         .send_check        = dccp_v4_send_check,
 909         .rebuild_header    = inet_sk_rebuild_header,
 910         .conn_request      = dccp_v4_conn_request,
 911         .syn_recv_sock     = dccp_v4_request_recv_sock,
 912         .net_header_len    = sizeof(struct iphdr),
 913         .setsockopt        = ip_setsockopt,
 914         .getsockopt        = ip_getsockopt,
 915         .addr2sockaddr     = inet_csk_addr2sockaddr,
 916         .sockaddr_len      = sizeof(struct sockaddr_in),
 917 #ifdef CONFIG_COMPAT
 918         .compat_setsockopt = compat_ip_setsockopt,
 919         .compat_getsockopt = compat_ip_getsockopt,
 920 #endif
 921 };
 922
 923 static int dccp_v4_init_sock(struct sock *sk)
 924 {
 925         static __u8 dccp_v4_ctl_sock_initialized;
 926         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 927
 928         if (err == 0) {
 929                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 930                         dccp_v4_ctl_sock_initialized = 1;
 931                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 932         }
 933
 934         return err;
 935 }
 936
 937 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 938         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 939 };
 940
 941 static struct proto dccp_v4_prot = {
 942         .name                   = "DCCP",
 943         .owner                  = THIS_MODULE,
 944         .close                  = dccp_close,
 945         .connect                = dccp_v4_connect,
 946         .disconnect             = dccp_disconnect,
 947         .ioctl                  = dccp_ioctl,
 948         .init                   = dccp_v4_init_sock,
 949         .setsockopt             = dccp_setsockopt,
 950         .getsockopt             = dccp_getsockopt,
 951         .sendmsg                = dccp_sendmsg,
 952         .recvmsg                = dccp_recvmsg,
 953         .backlog_rcv            = dccp_v4_do_rcv,
 954         .hash                   = inet_hash,
 955         .unhash                 = inet_unhash,
 956         .accept                 = inet_csk_accept,
 957         .get_port               = inet_csk_get_port,
 958         .shutdown               = dccp_shutdown,
 959         .destroy                = dccp_destroy_sock,
 960         .orphan_count           = &dccp_orphan_count,
 961         .max_header             = MAX_DCCP_HEADER,
 962         .obj_size               = sizeof(struct dccp_sock),
 963         .slab_flags             = SLAB_TYPESAFE_BY_RCU,
 964         .rsk_prot               = &dccp_request_sock_ops,
 965         .twsk_prot              = &dccp_timewait_sock_ops,
 966         .h.hashinfo             = &dccp_hashinfo,
 967 #ifdef CONFIG_COMPAT
 968         .compat_setsockopt      = compat_dccp_setsockopt,
 969         .compat_getsockopt      = compat_dccp_getsockopt,
 970 #endif
 971 };
 972
 973 static const struct net_protocol dccp_v4_protocol = {
 974         .handler        = dccp_v4_rcv,
 975         .err_handler    = dccp_v4_err,
 976         .no_policy      = 1,
 977         .netns_ok       = 1,
 978         .icmp_strict_tag_validation = 1,
 979 };
 980
 981 static const struct proto_ops inet_dccp_ops = {
 982         .family            = PF_INET,
 983         .owner             = THIS_MODULE,
 984         .release           = inet_release,
 985         .bind              = inet_bind,
 986         .connect           = inet_stream_connect,
 987         .socketpair        = sock_no_socketpair,
 988         .accept            = inet_accept,
 989         .getname           = inet_getname,
 990         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
 991         .poll              = dccp_poll,
 992         .ioctl             = inet_ioctl,
 993         /* FIXME: work on inet_listen to rename it to sock_common_listen */
 994         .listen            = inet_dccp_listen,
 995         .shutdown          = inet_shutdown,
 996         .setsockopt        = sock_common_setsockopt,
 997         .getsockopt        = sock_common_getsockopt,
 998         .sendmsg           = inet_sendmsg,
 999         .recvmsg           = sock_common_recvmsg,
1000         .mmap              = sock_no_mmap,
1001         .sendpage          = sock_no_sendpage,
1002 #ifdef CONFIG_COMPAT
1003         .compat_setsockopt = compat_sock_common_setsockopt,
1004         .compat_getsockopt = compat_sock_common_getsockopt,
1005 #endif
1006 };
1007
1008 static struct inet_protosw dccp_v4_protosw = {
1009         .type           = SOCK_DCCP,
1010         .protocol       = IPPROTO_DCCP,
1011         .prot           = &dccp_v4_prot,
1012         .ops            = &inet_dccp_ops,
1013         .flags          = INET_PROTOSW_ICSK,
1014 };
1015
1016 static int __net_init dccp_v4_init_net(struct net *net)
1017 {
1018         if (dccp_hashinfo.bhash == NULL)
1019                 return -ESOCKTNOSUPPORT;
1020
1021         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1022                                     SOCK_DCCP, IPPROTO_DCCP, net);
1023 }
1024
1025 static void __net_exit dccp_v4_exit_net(struct net *net)
1026 {
1027         inet_ctl_sock_destroy(net->dccp.v4_ctl_sk);
1028 }
1029
1030 static void __net_exit dccp_v4_exit_batch(struct list_head *net_exit_list)
1031 {
1032         inet_twsk_purge(&dccp_hashinfo, AF_INET);
1033 }
1034
1035 static struct pernet_operations dccp_v4_ops = {
1036         .init   = dccp_v4_init_net,
1037         .exit   = dccp_v4_exit_net,
1038         .exit_batch = dccp_v4_exit_batch,
1039 };
1040
1041 static int __init dccp_v4_init(void)
1042 {
1043         int err = proto_register(&dccp_v4_prot, 1);
1044
1045         if (err)
1046                 goto out;
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
1054         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1055         if (err)
1056                 goto out_proto_unregister;
1057
1058 out:
1059         return err;
1060 out_proto_unregister:
1061         unregister_pernet_subsys(&dccp_v4_ops);
1062 out_destroy_ctl_sock:
1063         inet_unregister_protosw(&dccp_v4_protosw);
1064         proto_unregister(&dccp_v4_prot);
1065         goto out;
1066 }
1067
1068 static void __exit dccp_v4_exit(void)
1069 {
1070         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1071         unregister_pernet_subsys(&dccp_v4_ops);
1072         inet_unregister_protosw(&dccp_v4_protosw);
1073         proto_unregister(&dccp_v4_prot);
1074 }
1075
1076 module_init(dccp_v4_init);
1077 module_exit(dccp_v4_exit);
1078
1079 /*
1080  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1081  * values directly, Also cover the case where the protocol is not specified,
1082  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1083  */
1084 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1085 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1086 MODULE_LICENSE("GPL");
1087 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1088 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");