net/dccp/ipv4.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  net/dccp/ipv4.c
   4  *
   5  *  An implementation of the DCCP protocol
   6  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
   7  */
   8
   9 #include <linux/dccp.h>
  10 #include <linux/icmp.h>
  11 #include <linux/slab.h>
  12 #include <linux/module.h>
  13 #include <linux/skbuff.h>
  14 #include <linux/random.h>
  15
  16 #include <net/icmp.h>
  17 #include <net/inet_common.h>
  18 #include <net/inet_hashtables.h>
  19 #include <net/inet_sock.h>
  20 #include <net/protocol.h>
  21 #include <net/sock.h>
  22 #include <net/timewait_sock.h>
  23 #include <net/tcp_states.h>
  24 #include <net/xfrm.h>
  25 #include <net/secure_seq.h>
  26 #include <net/netns/generic.h>
  27
  28 #include "ackvec.h"
  29 #include "ccid.h"
  30 #include "dccp.h"
  31 #include "feat.h"
  32
  33 struct dccp_v4_pernet {
  34         struct sock *v4_ctl_sk;
  35 };
  36
  37 static unsigned int dccp_v4_pernet_id __read_mostly;
  38
  39 /*
  40  * The per-net v4_ctl_sk socket is used for responding to
  41  * the Out-of-the-blue (OOTB) packets. A control sock will be created
  42  * for this socket at the initialization time.
  43  */
  44
  45 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  46 {
  47         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  48         struct inet_bind_hashbucket *prev_addr_hashbucket = NULL;
  49         __be32 daddr, nexthop, prev_sk_rcv_saddr;
  50         struct inet_sock *inet = inet_sk(sk);
  51         struct dccp_sock *dp = dccp_sk(sk);
  52         __be16 orig_sport, orig_dport;
  53         struct flowi4 *fl4;
  54         struct rtable *rt;
  55         int err;
  56         struct ip_options_rcu *inet_opt;
  57
  58         dp->dccps_role = DCCP_ROLE_CLIENT;
  59
  60         if (addr_len < sizeof(struct sockaddr_in))
  61                 return -EINVAL;
  62
  63         if (usin->sin_family != AF_INET)
  64                 return -EAFNOSUPPORT;
  65
  66         nexthop = daddr = usin->sin_addr.s_addr;
  67
  68         inet_opt = rcu_dereference_protected(inet->inet_opt,
  69                                              lockdep_sock_is_held(sk));
  70         if (inet_opt != NULL && inet_opt->opt.srr) {
  71                 if (daddr == 0)
  72                         return -EINVAL;
  73                 nexthop = inet_opt->opt.faddr;
  74         }
  75
  76         orig_sport = inet->inet_sport;
  77         orig_dport = usin->sin_port;
  78         fl4 = &inet->cork.fl.u.ip4;
  79         rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
  80                               sk->sk_bound_dev_if, IPPROTO_DCCP, orig_sport,
  81                               orig_dport, sk);
  82         if (IS_ERR(rt))
  83                 return PTR_ERR(rt);
  84
  85         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  86                 ip_rt_put(rt);
  87                 return -ENETUNREACH;
  88         }
  89
  90         if (inet_opt == NULL || !inet_opt->opt.srr)
  91                 daddr = fl4->daddr;
  92
  93         if (inet->inet_saddr == 0) {
  94                 if (inet_csk(sk)->icsk_bind2_hash) {
  95                         prev_addr_hashbucket =
  96                                 inet_bhashfn_portaddr(&dccp_hashinfo, sk,
  97                                                       sock_net(sk),
  98                                                       inet->inet_num);
  99                         prev_sk_rcv_saddr = sk->sk_rcv_saddr;
 100                 }
 101                 inet->inet_saddr = fl4->saddr;
 102         }
 103
 104         sk_rcv_saddr_set(sk, inet->inet_saddr);
 105
 106         if (prev_addr_hashbucket) {
 107                 err = inet_bhash2_update_saddr(prev_addr_hashbucket, sk);
 108                 if (err) {
 109                         inet->inet_saddr = 0;
 110                         sk_rcv_saddr_set(sk, prev_sk_rcv_saddr);
 111                         ip_rt_put(rt);
 112                         return err;
 113                 }
 114         }
 115
 116         inet->inet_dport = usin->sin_port;
 117         sk_daddr_set(sk, daddr);
 118
 119         inet_csk(sk)->icsk_ext_hdr_len = 0;
 120         if (inet_opt)
 121                 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
 122         /*
 123          * Socket identity is still unknown (sport may be zero).
 124          * However we set state to DCCP_REQUESTING and not releasing socket
 125          * lock select source port, enter ourselves into the hash tables and
 126          * complete initialization after this.
 127          */
 128         dccp_set_state(sk, DCCP_REQUESTING);
 129         err = inet_hash_connect(&dccp_death_row, sk);
 130         if (err != 0)
 131                 goto failure;
 132
 133         rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 134                                inet->inet_sport, inet->inet_dport, sk);
 135         if (IS_ERR(rt)) {
 136                 err = PTR_ERR(rt);
 137                 rt = NULL;
 138                 goto failure;
 139         }
 140         /* OK, now commit destination to socket.  */
 141         sk_setup_caps(sk, &rt->dst);
 142
 143         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
 144                                                     inet->inet_daddr,
 145                                                     inet->inet_sport,
 146                                                     inet->inet_dport);
 147         inet->inet_id = get_random_u16();
 148
 149         err = dccp_connect(sk);
 150         rt = NULL;
 151         if (err != 0)
 152                 goto failure;
 153 out:
 154         return err;
 155 failure:
 156         /*
 157          * This unhashes the socket and releases the local port, if necessary.
 158          */
 159         dccp_set_state(sk, DCCP_CLOSED);
 160         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
 161                 inet_reset_saddr(sk);
 162         ip_rt_put(rt);
 163         sk->sk_route_caps = 0;
 164         inet->inet_dport = 0;
 165         goto out;
 166 }
 167 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 168
 169 /*
 170  * This routine does path mtu discovery as defined in RFC1191.
 171  */
 172 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 173                                           const struct iphdr *iph,
 174                                           u32 mtu)
 175 {
 176         struct dst_entry *dst;
 177         const struct inet_sock *inet = inet_sk(sk);
 178         const struct dccp_sock *dp = dccp_sk(sk);
 179
 180         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 181          * send out by Linux are always < 576bytes so they should go through
 182          * unfragmented).
 183          */
 184         if (sk->sk_state == DCCP_LISTEN)
 185                 return;
 186
 187         dst = inet_csk_update_pmtu(sk, mtu);
 188         if (!dst)
 189                 return;
 190
 191         /* Something is about to be wrong... Remember soft error
 192          * for the case, if this connection will not able to recover.
 193          */
 194         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 195                 sk->sk_err_soft = EMSGSIZE;
 196
 197         mtu = dst_mtu(dst);
 198
 199         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 200             ip_sk_accept_pmtu(sk) &&
 201             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 202                 dccp_sync_mss(sk, mtu);
 203
 204                 /*
 205                  * From RFC 4340, sec. 14.1:
 206                  *
 207                  *      DCCP-Sync packets are the best choice for upward
 208                  *      probing, since DCCP-Sync probes do not risk application
 209                  *      data loss.
 210                  */
 211                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 212         } /* else let the usual retransmit timer handle it */
 213 }
 214
 215 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
 216 {
 217         struct dst_entry *dst = __sk_dst_check(sk, 0);
 218
 219         if (dst)
 220                 dst->ops->redirect(dst, sk, skb);
 221 }
 222
 223 void dccp_req_err(struct sock *sk, u64 seq)
 224         {
 225         struct request_sock *req = inet_reqsk(sk);
 226         struct net *net = sock_net(sk);
 227
 228         /*
 229          * ICMPs are not backlogged, hence we cannot get an established
 230          * socket here.
 231          */
 232         if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
 233                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 234         } else {
 235                 /*
 236                  * Still in RESPOND, just remove it silently.
 237                  * There is no good way to pass the error to the newly
 238                  * created socket, and POSIX does not want network
 239                  * errors returned from accept().
 240                  */
 241                 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
 242         }
 243         reqsk_put(req);
 244 }
 245 EXPORT_SYMBOL(dccp_req_err);
 246
 247 /*
 248  * This routine is called by the ICMP module when it gets some sort of error
 249  * condition. If err < 0 then the socket should be closed and the error
 250  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 251  * After adjustment header points to the first 8 bytes of the tcp header. We
 252  * need to find the appropriate port.
 253  *
 254  * The locking strategy used here is very "optimistic". When someone else
 255  * accesses the socket the ICMP is just dropped and for some paths there is no
 256  * check at all. A more general error queue to queue errors for later handling
 257  * is probably better.
 258  */
 259 static int dccp_v4_err(struct sk_buff *skb, u32 info)
 260 {
 261         const struct iphdr *iph = (struct iphdr *)skb->data;
 262         const u8 offset = iph->ihl << 2;
 263         const struct dccp_hdr *dh;
 264         struct dccp_sock *dp;
 265         struct inet_sock *inet;
 266         const int type = icmp_hdr(skb)->type;
 267         const int code = icmp_hdr(skb)->code;
 268         struct sock *sk;
 269         __u64 seq;
 270         int err;
 271         struct net *net = dev_net(skb->dev);
 272
 273         /* Only need dccph_dport & dccph_sport which are the first
 274          * 4 bytes in dccp header.
 275          * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us.
 276          */
 277         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_sport) > 8);
 278         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_dport) > 8);
 279         dh = (struct dccp_hdr *)(skb->data + offset);
 280
 281         sk = __inet_lookup_established(net, &dccp_hashinfo,
 282                                        iph->daddr, dh->dccph_dport,
 283                                        iph->saddr, ntohs(dh->dccph_sport),
 284                                        inet_iif(skb), 0);
 285         if (!sk) {
 286                 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 287                 return -ENOENT;
 288         }
 289
 290         if (sk->sk_state == DCCP_TIME_WAIT) {
 291                 inet_twsk_put(inet_twsk(sk));
 292                 return 0;
 293         }
 294         seq = dccp_hdr_seq(dh);
 295         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 296                 dccp_req_err(sk, seq);
 297                 return 0;
 298         }
 299
 300         bh_lock_sock(sk);
 301         /* If too many ICMPs get dropped on busy
 302          * servers this needs to be solved differently.
 303          */
 304         if (sock_owned_by_user(sk))
 305                 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
 306
 307         if (sk->sk_state == DCCP_CLOSED)
 308                 goto out;
 309
 310         dp = dccp_sk(sk);
 311         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 312             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 313                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 314                 goto out;
 315         }
 316
 317         switch (type) {
 318         case ICMP_REDIRECT:
 319                 if (!sock_owned_by_user(sk))
 320                         dccp_do_redirect(skb, sk);
 321                 goto out;
 322         case ICMP_SOURCE_QUENCH:
 323                 /* Just silently ignore these. */
 324                 goto out;
 325         case ICMP_PARAMETERPROB:
 326                 err = EPROTO;
 327                 break;
 328         case ICMP_DEST_UNREACH:
 329                 if (code > NR_ICMP_UNREACH)
 330                         goto out;
 331
 332                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 333                         if (!sock_owned_by_user(sk))
 334                                 dccp_do_pmtu_discovery(sk, iph, info);
 335                         goto out;
 336                 }
 337
 338                 err = icmp_err_convert[code].errno;
 339                 break;
 340         case ICMP_TIME_EXCEEDED:
 341                 err = EHOSTUNREACH;
 342                 break;
 343         default:
 344                 goto out;
 345         }
 346
 347         switch (sk->sk_state) {
 348         case DCCP_REQUESTING:
 349         case DCCP_RESPOND:
 350                 if (!sock_owned_by_user(sk)) {
 351                         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 352                         sk->sk_err = err;
 353
 354                         sk_error_report(sk);
 355
 356                         dccp_done(sk);
 357                 } else
 358                         sk->sk_err_soft = err;
 359                 goto out;
 360         }
 361
 362         /* If we've already connected we will keep trying
 363          * until we time out, or the user gives up.
 364          *
 365          * rfc1122 4.2.3.9 allows to consider as hard errors
 366          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 367          * but it is obsoleted by pmtu discovery).
 368          *
 369          * Note, that in modern internet, where routing is unreliable
 370          * and in each dark corner broken firewalls sit, sending random
 371          * errors ordered by their masters even this two messages finally lose
 372          * their original sense (even Linux sends invalid PORT_UNREACHs)
 373          *
 374          * Now we are in compliance with RFCs.
 375          *                                                      --ANK (980905)
 376          */
 377
 378         inet = inet_sk(sk);
 379         if (!sock_owned_by_user(sk) && inet->recverr) {
 380                 sk->sk_err = err;
 381                 sk_error_report(sk);
 382         } else /* Only an error on timeout */
 383                 sk->sk_err_soft = err;
 384 out:
 385         bh_unlock_sock(sk);
 386         sock_put(sk);
 387         return 0;
 388 }
 389
 390 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 391                                       __be32 src, __be32 dst)
 392 {
 393         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 394 }
 395
 396 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 397 {
 398         const struct inet_sock *inet = inet_sk(sk);
 399         struct dccp_hdr *dh = dccp_hdr(skb);
 400
 401         dccp_csum_outgoing(skb);
 402         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 403                                                  inet->inet_saddr,
 404                                                  inet->inet_daddr);
 405 }
 406 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 407
 408 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 409 {
 410         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 411                                            ip_hdr(skb)->saddr,
 412                                            dccp_hdr(skb)->dccph_dport,
 413                                            dccp_hdr(skb)->dccph_sport);
 414 }
 415
 416 /*
 417  * The three way handshake has completed - we got a valid ACK or DATAACK -
 418  * now create the new socket.
 419  *
 420  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 421  */
 422 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
 423                                        struct sk_buff *skb,
 424                                        struct request_sock *req,
 425                                        struct dst_entry *dst,
 426                                        struct request_sock *req_unhash,
 427                                        bool *own_req)
 428 {
 429         struct inet_request_sock *ireq;
 430         struct inet_sock *newinet;
 431         struct sock *newsk;
 432
 433         if (sk_acceptq_is_full(sk))
 434                 goto exit_overflow;
 435
 436         newsk = dccp_create_openreq_child(sk, req, skb);
 437         if (newsk == NULL)
 438                 goto exit_nonewsk;
 439
 440         newinet            = inet_sk(newsk);
 441         ireq               = inet_rsk(req);
 442         sk_daddr_set(newsk, ireq->ir_rmt_addr);
 443         sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
 444         newinet->inet_saddr     = ireq->ir_loc_addr;
 445         RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt));
 446         newinet->mc_index  = inet_iif(skb);
 447         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 448         newinet->inet_id   = get_random_u16();
 449
 450         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
 451                 goto put_and_exit;
 452
 453         sk_setup_caps(newsk, dst);
 454
 455         dccp_sync_mss(newsk, dst_mtu(dst));
 456
 457         if (__inet_inherit_port(sk, newsk) < 0)
 458                 goto put_and_exit;
 459         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), NULL);
 460         if (*own_req)
 461                 ireq->ireq_opt = NULL;
 462         else
 463                 newinet->inet_opt = NULL;
 464         return newsk;
 465
 466 exit_overflow:
 467         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 468 exit_nonewsk:
 469         dst_release(dst);
 470 exit:
 471         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
 472         return NULL;
 473 put_and_exit:
 474         newinet->inet_opt = NULL;
 475         inet_csk_prepare_forced_close(newsk);
 476         dccp_done(newsk);
 477         goto exit;
 478 }
 479 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 480
 481 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 482                                            struct sk_buff *skb)
 483 {
 484         struct rtable *rt;
 485         const struct iphdr *iph = ip_hdr(skb);
 486         struct flowi4 fl4 = {
 487                 .flowi4_oif = inet_iif(skb),
 488                 .daddr = iph->saddr,
 489                 .saddr = iph->daddr,
 490                 .flowi4_tos = RT_CONN_FLAGS(sk),
 491                 .flowi4_proto = sk->sk_protocol,
 492                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
 493                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
 494         };
 495
 496         security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
 497         rt = ip_route_output_flow(net, &fl4, sk);
 498         if (IS_ERR(rt)) {
 499                 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 500                 return NULL;
 501         }
 502
 503         return &rt->dst;
 504 }
 505
 506 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
 507 {
 508         int err = -1;
 509         struct sk_buff *skb;
 510         struct dst_entry *dst;
 511         struct flowi4 fl4;
 512
 513         dst = inet_csk_route_req(sk, &fl4, req);
 514         if (dst == NULL)
 515                 goto out;
 516
 517         skb = dccp_make_response(sk, dst, req);
 518         if (skb != NULL) {
 519                 const struct inet_request_sock *ireq = inet_rsk(req);
 520                 struct dccp_hdr *dh = dccp_hdr(skb);
 521
 522                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
 523                                                               ireq->ir_rmt_addr);
 524                 rcu_read_lock();
 525                 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 526                                             ireq->ir_rmt_addr,
 527                                             rcu_dereference(ireq->ireq_opt),
 528                                             inet_sk(sk)->tos);
 529                 rcu_read_unlock();
 530                 err = net_xmit_eval(err);
 531         }
 532
 533 out:
 534         dst_release(dst);
 535         return err;
 536 }
 537
 538 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
 539 {
 540         int err;
 541         const struct iphdr *rxiph;
 542         struct sk_buff *skb;
 543         struct dst_entry *dst;
 544         struct net *net = dev_net(skb_dst(rxskb)->dev);
 545         struct dccp_v4_pernet *pn;
 546         struct sock *ctl_sk;
 547
 548         /* Never send a reset in response to a reset. */
 549         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 550                 return;
 551
 552         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 553                 return;
 554
 555         pn = net_generic(net, dccp_v4_pernet_id);
 556         ctl_sk = pn->v4_ctl_sk;
 557         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 558         if (dst == NULL)
 559                 return;
 560
 561         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 562         if (skb == NULL)
 563                 goto out;
 564
 565         rxiph = ip_hdr(rxskb);
 566         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 567                                                                  rxiph->daddr);
 568         skb_dst_set(skb, dst_clone(dst));
 569
 570         local_bh_disable();
 571         bh_lock_sock(ctl_sk);
 572         err = ip_build_and_send_pkt(skb, ctl_sk,
 573                                     rxiph->daddr, rxiph->saddr, NULL,
 574                                     inet_sk(ctl_sk)->tos);
 575         bh_unlock_sock(ctl_sk);
 576
 577         if (net_xmit_eval(err) == 0) {
 578                 __DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 579                 __DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
 580         }
 581         local_bh_enable();
 582 out:
 583         dst_release(dst);
 584 }
 585
 586 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 587 {
 588         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 589         kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
 590 }
 591
 592 void dccp_syn_ack_timeout(const struct request_sock *req)
 593 {
 594 }
 595 EXPORT_SYMBOL(dccp_syn_ack_timeout);
 596
 597 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 598         .family         = PF_INET,
 599         .obj_size       = sizeof(struct dccp_request_sock),
 600         .rtx_syn_ack    = dccp_v4_send_response,
 601         .send_ack       = dccp_reqsk_send_ack,
 602         .destructor     = dccp_v4_reqsk_destructor,
 603         .send_reset     = dccp_v4_ctl_send_reset,
 604         .syn_ack_timeout = dccp_syn_ack_timeout,
 605 };
 606
 607 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 608 {
 609         struct inet_request_sock *ireq;
 610         struct request_sock *req;
 611         struct dccp_request_sock *dreq;
 612         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 613         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 614
 615         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 616         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 617                 return 0;       /* discard, don't send a reset here */
 618
 619         if (dccp_bad_service_code(sk, service)) {
 620                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 621                 goto drop;
 622         }
 623         /*
 624          * TW buckets are converted to open requests without
 625          * limitations, they conserve resources and peer is
 626          * evidently real one.
 627          */
 628         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 629         if (inet_csk_reqsk_queue_is_full(sk))
 630                 goto drop;
 631
 632         if (sk_acceptq_is_full(sk))
 633                 goto drop;
 634
 635         req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
 636         if (req == NULL)
 637                 goto drop;
 638
 639         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 640                 goto drop_and_free;
 641
 642         dreq = dccp_rsk(req);
 643         if (dccp_parse_options(sk, dreq, skb))
 644                 goto drop_and_free;
 645
 646         if (security_inet_conn_request(sk, skb, req))
 647                 goto drop_and_free;
 648
 649         ireq = inet_rsk(req);
 650         sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
 651         sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
 652         ireq->ir_mark = inet_request_mark(sk, skb);
 653         ireq->ireq_family = AF_INET;
 654         ireq->ir_iif = READ_ONCE(sk->sk_bound_dev_if);
 655
 656         /*
 657          * Step 3: Process LISTEN state
 658          *
 659          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 660          *
 661          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 662          */
 663         dreq->dreq_isr     = dcb->dccpd_seq;
 664         dreq->dreq_gsr     = dreq->dreq_isr;
 665         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 666         dreq->dreq_gss     = dreq->dreq_iss;
 667         dreq->dreq_service = service;
 668
 669         if (dccp_v4_send_response(sk, req))
 670                 goto drop_and_free;
 671
 672         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 673         reqsk_put(req);
 674         return 0;
 675
 676 drop_and_free:
 677         reqsk_free(req);
 678 drop:
 679         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 680         return -1;
 681 }
 682 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 683
 684 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 685 {
 686         struct dccp_hdr *dh = dccp_hdr(skb);
 687
 688         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 689                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 690                         goto reset;
 691                 return 0;
 692         }
 693
 694         /*
 695          *  Step 3: Process LISTEN state
 696          *       If P.type == Request or P contains a valid Init Cookie option,
 697          *            (* Must scan the packet's options to check for Init
 698          *               Cookies.  Only Init Cookies are processed here,
 699          *               however; other options are processed in Step 8.  This
 700          *               scan need only be performed if the endpoint uses Init
 701          *               Cookies *)
 702          *            (* Generate a new socket and switch to that socket *)
 703          *            Set S := new socket for this port pair
 704          *            S.state = RESPOND
 705          *            Choose S.ISS (initial seqno) or set from Init Cookies
 706          *            Initialize S.GAR := S.ISS
 707          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 708          *            Continue with S.state == RESPOND
 709          *            (* A Response packet will be generated in Step 11 *)
 710          *       Otherwise,
 711          *            Generate Reset(No Connection) unless P.type == Reset
 712          *            Drop packet and return
 713          *
 714          * NOTE: the check for the packet types is done in
 715          *       dccp_rcv_state_process
 716          */
 717
 718         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 719                 goto reset;
 720         return 0;
 721
 722 reset:
 723         dccp_v4_ctl_send_reset(sk, skb);
 724         kfree_skb(skb);
 725         return 0;
 726 }
 727 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 728
 729 /**
 730  *      dccp_invalid_packet  -  check for malformed packets
 731  *      @skb: Packet to validate
 732  *
 733  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 734  *      Packets that fail these checks are ignored and do not receive Resets.
 735  */
 736 int dccp_invalid_packet(struct sk_buff *skb)
 737 {
 738         const struct dccp_hdr *dh;
 739         unsigned int cscov;
 740         u8 dccph_doff;
 741
 742         if (skb->pkt_type != PACKET_HOST)
 743                 return 1;
 744
 745         /* If the packet is shorter than 12 bytes, drop packet and return */
 746         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 747                 DCCP_WARN("pskb_may_pull failed\n");
 748                 return 1;
 749         }
 750
 751         dh = dccp_hdr(skb);
 752
 753         /* If P.type is not understood, drop packet and return */
 754         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 755                 DCCP_WARN("invalid packet type\n");
 756                 return 1;
 757         }
 758
 759         /*
 760          * If P.Data Offset is too small for packet type, drop packet and return
 761          */
 762         dccph_doff = dh->dccph_doff;
 763         if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 764                 DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
 765                 return 1;
 766         }
 767         /*
 768          * If P.Data Offset is too large for packet, drop packet and return
 769          */
 770         if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
 771                 DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
 772                 return 1;
 773         }
 774         dh = dccp_hdr(skb);
 775         /*
 776          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 777          * has short sequence numbers), drop packet and return
 778          */
 779         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 780             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 781                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 782                           dccp_packet_name(dh->dccph_type));
 783                 return 1;
 784         }
 785
 786         /*
 787          * If P.CsCov is too large for the packet size, drop packet and return.
 788          * This must come _before_ checksumming (not as RFC 4340 suggests).
 789          */
 790         cscov = dccp_csum_coverage(skb);
 791         if (cscov > skb->len) {
 792                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 793                           dh->dccph_cscov, skb->len);
 794                 return 1;
 795         }
 796
 797         /* If header checksum is incorrect, drop packet and return.
 798          * (This step is completed in the AF-dependent functions.) */
 799         skb->csum = skb_checksum(skb, 0, cscov, 0);
 800
 801         return 0;
 802 }
 803 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 804
 805 /* this is called when real data arrives */
 806 static int dccp_v4_rcv(struct sk_buff *skb)
 807 {
 808         const struct dccp_hdr *dh;
 809         const struct iphdr *iph;
 810         bool refcounted;
 811         struct sock *sk;
 812         int min_cov;
 813
 814         /* Step 1: Check header basics */
 815
 816         if (dccp_invalid_packet(skb))
 817                 goto discard_it;
 818
 819         iph = ip_hdr(skb);
 820         /* Step 1: If header checksum is incorrect, drop packet and return */
 821         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 822                 DCCP_WARN("dropped packet with invalid checksum\n");
 823                 goto discard_it;
 824         }
 825
 826         dh = dccp_hdr(skb);
 827
 828         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 829         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 830
 831         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 832                       dccp_packet_name(dh->dccph_type),
 833                       &iph->saddr, ntohs(dh->dccph_sport),
 834                       &iph->daddr, ntohs(dh->dccph_dport),
 835                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 836
 837         if (dccp_packet_without_ack(skb)) {
 838                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 839                 dccp_pr_debug_cat("\n");
 840         } else {
 841                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 842                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 843                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 844         }
 845
 846 lookup:
 847         sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
 848                                dh->dccph_sport, dh->dccph_dport, 0, &refcounted);
 849         if (!sk) {
 850                 dccp_pr_debug("failed to look up flow ID in table and "
 851                               "get corresponding socket\n");
 852                 goto no_dccp_socket;
 853         }
 854
 855         /*
 856          * Step 2:
 857          *      ... or S.state == TIMEWAIT,
 858          *              Generate Reset(No Connection) unless P.type == Reset
 859          *              Drop packet and return
 860          */
 861         if (sk->sk_state == DCCP_TIME_WAIT) {
 862                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 863                 inet_twsk_put(inet_twsk(sk));
 864                 goto no_dccp_socket;
 865         }
 866
 867         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 868                 struct request_sock *req = inet_reqsk(sk);
 869                 struct sock *nsk;
 870
 871                 sk = req->rsk_listener;
 872                 if (unlikely(sk->sk_state != DCCP_LISTEN)) {
 873                         inet_csk_reqsk_queue_drop_and_put(sk, req);
 874                         goto lookup;
 875                 }
 876                 sock_hold(sk);
 877                 refcounted = true;
 878                 nsk = dccp_check_req(sk, skb, req);
 879                 if (!nsk) {
 880                         reqsk_put(req);
 881                         goto discard_and_relse;
 882                 }
 883                 if (nsk == sk) {
 884                         reqsk_put(req);
 885                 } else if (dccp_child_process(sk, nsk, skb)) {
 886                         dccp_v4_ctl_send_reset(sk, skb);
 887                         goto discard_and_relse;
 888                 } else {
 889                         sock_put(sk);
 890                         return 0;
 891                 }
 892         }
 893         /*
 894          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 895          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 896          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 897          */
 898         min_cov = dccp_sk(sk)->dccps_pcrlen;
 899         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 900                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 901                               dh->dccph_cscov, min_cov);
 902                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 903                  *         options (Section 11.7) with Drop Code 0, Protocol
 904                  *         Constraints."                                     */
 905                 goto discard_and_relse;
 906         }
 907
 908         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 909                 goto discard_and_relse;
 910         nf_reset_ct(skb);
 911
 912         return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
 913
 914 no_dccp_socket:
 915         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 916                 goto discard_it;
 917         /*
 918          * Step 2:
 919          *      If no socket ...
 920          *              Generate Reset(No Connection) unless P.type == Reset
 921          *              Drop packet and return
 922          */
 923         if (dh->dccph_type != DCCP_PKT_RESET) {
 924                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 925                                         DCCP_RESET_CODE_NO_CONNECTION;
 926                 dccp_v4_ctl_send_reset(sk, skb);
 927         }
 928
 929 discard_it:
 930         kfree_skb(skb);
 931         return 0;
 932
 933 discard_and_relse:
 934         if (refcounted)
 935                 sock_put(sk);
 936         goto discard_it;
 937 }
 938
 939 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 940         .queue_xmit        = ip_queue_xmit,
 941         .send_check        = dccp_v4_send_check,
 942         .rebuild_header    = inet_sk_rebuild_header,
 943         .conn_request      = dccp_v4_conn_request,
 944         .syn_recv_sock     = dccp_v4_request_recv_sock,
 945         .net_header_len    = sizeof(struct iphdr),
 946         .setsockopt        = ip_setsockopt,
 947         .getsockopt        = ip_getsockopt,
 948         .addr2sockaddr     = inet_csk_addr2sockaddr,
 949         .sockaddr_len      = sizeof(struct sockaddr_in),
 950 };
 951
 952 static int dccp_v4_init_sock(struct sock *sk)
 953 {
 954         static __u8 dccp_v4_ctl_sock_initialized;
 955         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 956
 957         if (err == 0) {
 958                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 959                         dccp_v4_ctl_sock_initialized = 1;
 960                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 961         }
 962
 963         return err;
 964 }
 965
 966 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 967         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 968 };
 969
 970 static struct proto dccp_v4_prot = {
 971         .name                   = "DCCP",
 972         .owner                  = THIS_MODULE,
 973         .close                  = dccp_close,
 974         .connect                = dccp_v4_connect,
 975         .disconnect             = dccp_disconnect,
 976         .ioctl                  = dccp_ioctl,
 977         .init                   = dccp_v4_init_sock,
 978         .setsockopt             = dccp_setsockopt,
 979         .getsockopt             = dccp_getsockopt,
 980         .sendmsg                = dccp_sendmsg,
 981         .recvmsg                = dccp_recvmsg,
 982         .backlog_rcv            = dccp_v4_do_rcv,
 983         .hash                   = inet_hash,
 984         .unhash                 = inet_unhash,
 985         .accept                 = inet_csk_accept,
 986         .get_port               = inet_csk_get_port,
 987         .shutdown               = dccp_shutdown,
 988         .destroy                = dccp_destroy_sock,
 989         .orphan_count           = &dccp_orphan_count,
 990         .max_header             = MAX_DCCP_HEADER,
 991         .obj_size               = sizeof(struct dccp_sock),
 992         .slab_flags             = SLAB_TYPESAFE_BY_RCU,
 993         .rsk_prot               = &dccp_request_sock_ops,
 994         .twsk_prot              = &dccp_timewait_sock_ops,
 995         .h.hashinfo             = &dccp_hashinfo,
 996 };
 997
 998 static const struct net_protocol dccp_v4_protocol = {
 999         .handler        = dccp_v4_rcv,
1000         .err_handler    = dccp_v4_err,
1001         .no_policy      = 1,
1002         .icmp_strict_tag_validation = 1,
1003 };
1004
1005 static const struct proto_ops inet_dccp_ops = {
1006         .family            = PF_INET,
1007         .owner             = THIS_MODULE,
1008         .release           = inet_release,
1009         .bind              = inet_bind,
1010         .connect           = inet_stream_connect,
1011         .socketpair        = sock_no_socketpair,
1012         .accept            = inet_accept,
1013         .getname           = inet_getname,
1014         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1015         .poll              = dccp_poll,
1016         .ioctl             = inet_ioctl,
1017         .gettstamp         = sock_gettstamp,
1018         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1019         .listen            = inet_dccp_listen,
1020         .shutdown          = inet_shutdown,
1021         .setsockopt        = sock_common_setsockopt,
1022         .getsockopt        = sock_common_getsockopt,
1023         .sendmsg           = inet_sendmsg,
1024         .recvmsg           = sock_common_recvmsg,
1025         .mmap              = sock_no_mmap,
1026         .sendpage          = sock_no_sendpage,
1027 };
1028
1029 static struct inet_protosw dccp_v4_protosw = {
1030         .type           = SOCK_DCCP,
1031         .protocol       = IPPROTO_DCCP,
1032         .prot           = &dccp_v4_prot,
1033         .ops            = &inet_dccp_ops,
1034         .flags          = INET_PROTOSW_ICSK,
1035 };
1036
1037 static int __net_init dccp_v4_init_net(struct net *net)
1038 {
1039         struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1040
1041         if (dccp_hashinfo.bhash == NULL)
1042                 return -ESOCKTNOSUPPORT;
1043
1044         return inet_ctl_sock_create(&pn->v4_ctl_sk, PF_INET,
1045                                     SOCK_DCCP, IPPROTO_DCCP, net);
1046 }
1047
1048 static void __net_exit dccp_v4_exit_net(struct net *net)
1049 {
1050         struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1051
1052         inet_ctl_sock_destroy(pn->v4_ctl_sk);
1053 }
1054
1055 static void __net_exit dccp_v4_exit_batch(struct list_head *net_exit_list)
1056 {
1057         inet_twsk_purge(&dccp_hashinfo, AF_INET);
1058 }
1059
1060 static struct pernet_operations dccp_v4_ops = {
1061         .init   = dccp_v4_init_net,
1062         .exit   = dccp_v4_exit_net,
1063         .exit_batch = dccp_v4_exit_batch,
1064         .id     = &dccp_v4_pernet_id,
1065         .size   = sizeof(struct dccp_v4_pernet),
1066 };
1067
1068 static int __init dccp_v4_init(void)
1069 {
1070         int err = proto_register(&dccp_v4_prot, 1);
1071
1072         if (err)
1073                 goto out;
1074
1075         inet_register_protosw(&dccp_v4_protosw);
1076
1077         err = register_pernet_subsys(&dccp_v4_ops);
1078         if (err)
1079                 goto out_destroy_ctl_sock;
1080
1081         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1082         if (err)
1083                 goto out_proto_unregister;
1084
1085 out:
1086         return err;
1087 out_proto_unregister:
1088         unregister_pernet_subsys(&dccp_v4_ops);
1089 out_destroy_ctl_sock:
1090         inet_unregister_protosw(&dccp_v4_protosw);
1091         proto_unregister(&dccp_v4_prot);
1092         goto out;
1093 }
1094
1095 static void __exit dccp_v4_exit(void)
1096 {
1097         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1098         unregister_pernet_subsys(&dccp_v4_ops);
1099         inet_unregister_protosw(&dccp_v4_protosw);
1100         proto_unregister(&dccp_v4_prot);
1101 }
1102
1103 module_init(dccp_v4_init);
1104 module_exit(dccp_v4_exit);
1105
1106 /*
1107  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1108  * values directly, Also cover the case where the protocol is not specified,
1109  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1110  */
1111 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1112 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1113 MODULE_LICENSE("GPL");
1114 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1115 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");