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_sock *inet = inet_sk(sk);
  49         struct dccp_sock *dp = dccp_sk(sk);
  50         __be16 orig_sport, orig_dport;
  51         __be32 daddr, nexthop;
  52         struct flowi4 *fl4;
  53         struct rtable *rt;
  54         int err;
  55         struct ip_options_rcu *inet_opt;
  56
  57         dp->dccps_role = DCCP_ROLE_CLIENT;
  58
  59         if (addr_len < sizeof(struct sockaddr_in))
  60                 return -EINVAL;
  61
  62         if (usin->sin_family != AF_INET)
  63                 return -EAFNOSUPPORT;
  64
  65         nexthop = daddr = usin->sin_addr.s_addr;
  66
  67         inet_opt = rcu_dereference_protected(inet->inet_opt,
  68                                              lockdep_sock_is_held(sk));
  69         if (inet_opt != NULL && inet_opt->opt.srr) {
  70                 if (daddr == 0)
  71                         return -EINVAL;
  72                 nexthop = inet_opt->opt.faddr;
  73         }
  74
  75         orig_sport = inet->inet_sport;
  76         orig_dport = usin->sin_port;
  77         fl4 = &inet->cork.fl.u.ip4;
  78         rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
  79                               RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  80                               IPPROTO_DCCP,
  81                               orig_sport, 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                 inet->inet_saddr = fl4->saddr;
  95         sk_rcv_saddr_set(sk, inet->inet_saddr);
  96         inet->inet_dport = usin->sin_port;
  97         sk_daddr_set(sk, daddr);
  98
  99         inet_csk(sk)->icsk_ext_hdr_len = 0;
 100         if (inet_opt)
 101                 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
 102         /*
 103          * Socket identity is still unknown (sport may be zero).
 104          * However we set state to DCCP_REQUESTING and not releasing socket
 105          * lock select source port, enter ourselves into the hash tables and
 106          * complete initialization after this.
 107          */
 108         dccp_set_state(sk, DCCP_REQUESTING);
 109         err = inet_hash_connect(&dccp_death_row, sk);
 110         if (err != 0)
 111                 goto failure;
 112
 113         rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 114                                inet->inet_sport, inet->inet_dport, sk);
 115         if (IS_ERR(rt)) {
 116                 err = PTR_ERR(rt);
 117                 rt = NULL;
 118                 goto failure;
 119         }
 120         /* OK, now commit destination to socket.  */
 121         sk_setup_caps(sk, &rt->dst);
 122
 123         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
 124                                                     inet->inet_daddr,
 125                                                     inet->inet_sport,
 126                                                     inet->inet_dport);
 127         inet->inet_id = prandom_u32();
 128
 129         err = dccp_connect(sk);
 130         rt = NULL;
 131         if (err != 0)
 132                 goto failure;
 133 out:
 134         return err;
 135 failure:
 136         /*
 137          * This unhashes the socket and releases the local port, if necessary.
 138          */
 139         dccp_set_state(sk, DCCP_CLOSED);
 140         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
 141                 inet_reset_saddr(sk);
 142         ip_rt_put(rt);
 143         sk->sk_route_caps = 0;
 144         inet->inet_dport = 0;
 145         goto out;
 146 }
 147 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 148
 149 /*
 150  * This routine does path mtu discovery as defined in RFC1191.
 151  */
 152 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 153                                           const struct iphdr *iph,
 154                                           u32 mtu)
 155 {
 156         struct dst_entry *dst;
 157         const struct inet_sock *inet = inet_sk(sk);
 158         const struct dccp_sock *dp = dccp_sk(sk);
 159
 160         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 161          * send out by Linux are always < 576bytes so they should go through
 162          * unfragmented).
 163          */
 164         if (sk->sk_state == DCCP_LISTEN)
 165                 return;
 166
 167         dst = inet_csk_update_pmtu(sk, mtu);
 168         if (!dst)
 169                 return;
 170
 171         /* Something is about to be wrong... Remember soft error
 172          * for the case, if this connection will not able to recover.
 173          */
 174         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 175                 sk->sk_err_soft = EMSGSIZE;
 176
 177         mtu = dst_mtu(dst);
 178
 179         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 180             ip_sk_accept_pmtu(sk) &&
 181             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 182                 dccp_sync_mss(sk, mtu);
 183
 184                 /*
 185                  * From RFC 4340, sec. 14.1:
 186                  *
 187                  *      DCCP-Sync packets are the best choice for upward
 188                  *      probing, since DCCP-Sync probes do not risk application
 189                  *      data loss.
 190                  */
 191                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 192         } /* else let the usual retransmit timer handle it */
 193 }
 194
 195 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
 196 {
 197         struct dst_entry *dst = __sk_dst_check(sk, 0);
 198
 199         if (dst)
 200                 dst->ops->redirect(dst, sk, skb);
 201 }
 202
 203 void dccp_req_err(struct sock *sk, u64 seq)
 204         {
 205         struct request_sock *req = inet_reqsk(sk);
 206         struct net *net = sock_net(sk);
 207
 208         /*
 209          * ICMPs are not backlogged, hence we cannot get an established
 210          * socket here.
 211          */
 212         if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
 213                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 214         } else {
 215                 /*
 216                  * Still in RESPOND, just remove it silently.
 217                  * There is no good way to pass the error to the newly
 218                  * created socket, and POSIX does not want network
 219                  * errors returned from accept().
 220                  */
 221                 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
 222         }
 223         reqsk_put(req);
 224 }
 225 EXPORT_SYMBOL(dccp_req_err);
 226
 227 /*
 228  * This routine is called by the ICMP module when it gets some sort of error
 229  * condition. If err < 0 then the socket should be closed and the error
 230  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 231  * After adjustment header points to the first 8 bytes of the tcp header. We
 232  * need to find the appropriate port.
 233  *
 234  * The locking strategy used here is very "optimistic". When someone else
 235  * accesses the socket the ICMP is just dropped and for some paths there is no
 236  * check at all. A more general error queue to queue errors for later handling
 237  * is probably better.
 238  */
 239 static int dccp_v4_err(struct sk_buff *skb, u32 info)
 240 {
 241         const struct iphdr *iph = (struct iphdr *)skb->data;
 242         const u8 offset = iph->ihl << 2;
 243         const struct dccp_hdr *dh;
 244         struct dccp_sock *dp;
 245         struct inet_sock *inet;
 246         const int type = icmp_hdr(skb)->type;
 247         const int code = icmp_hdr(skb)->code;
 248         struct sock *sk;
 249         __u64 seq;
 250         int err;
 251         struct net *net = dev_net(skb->dev);
 252
 253         /* Only need dccph_dport & dccph_sport which are the first
 254          * 4 bytes in dccp header.
 255          * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us.
 256          */
 257         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_sport) > 8);
 258         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_dport) > 8);
 259         dh = (struct dccp_hdr *)(skb->data + offset);
 260
 261         sk = __inet_lookup_established(net, &dccp_hashinfo,
 262                                        iph->daddr, dh->dccph_dport,
 263                                        iph->saddr, ntohs(dh->dccph_sport),
 264                                        inet_iif(skb), 0);
 265         if (!sk) {
 266                 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 267                 return -ENOENT;
 268         }
 269
 270         if (sk->sk_state == DCCP_TIME_WAIT) {
 271                 inet_twsk_put(inet_twsk(sk));
 272                 return 0;
 273         }
 274         seq = dccp_hdr_seq(dh);
 275         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 276                 dccp_req_err(sk, seq);
 277                 return 0;
 278         }
 279
 280         bh_lock_sock(sk);
 281         /* If too many ICMPs get dropped on busy
 282          * servers this needs to be solved differently.
 283          */
 284         if (sock_owned_by_user(sk))
 285                 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
 286
 287         if (sk->sk_state == DCCP_CLOSED)
 288                 goto out;
 289
 290         dp = dccp_sk(sk);
 291         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 292             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 293                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 294                 goto out;
 295         }
 296
 297         switch (type) {
 298         case ICMP_REDIRECT:
 299                 if (!sock_owned_by_user(sk))
 300                         dccp_do_redirect(skb, sk);
 301                 goto out;
 302         case ICMP_SOURCE_QUENCH:
 303                 /* Just silently ignore these. */
 304                 goto out;
 305         case ICMP_PARAMETERPROB:
 306                 err = EPROTO;
 307                 break;
 308         case ICMP_DEST_UNREACH:
 309                 if (code > NR_ICMP_UNREACH)
 310                         goto out;
 311
 312                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 313                         if (!sock_owned_by_user(sk))
 314                                 dccp_do_pmtu_discovery(sk, iph, info);
 315                         goto out;
 316                 }
 317
 318                 err = icmp_err_convert[code].errno;
 319                 break;
 320         case ICMP_TIME_EXCEEDED:
 321                 err = EHOSTUNREACH;
 322                 break;
 323         default:
 324                 goto out;
 325         }
 326
 327         switch (sk->sk_state) {
 328         case DCCP_REQUESTING:
 329         case DCCP_RESPOND:
 330                 if (!sock_owned_by_user(sk)) {
 331                         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 332                         sk->sk_err = err;
 333
 334                         sk_error_report(sk);
 335
 336                         dccp_done(sk);
 337                 } else
 338                         sk->sk_err_soft = err;
 339                 goto out;
 340         }
 341
 342         /* If we've already connected we will keep trying
 343          * until we time out, or the user gives up.
 344          *
 345          * rfc1122 4.2.3.9 allows to consider as hard errors
 346          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 347          * but it is obsoleted by pmtu discovery).
 348          *
 349          * Note, that in modern internet, where routing is unreliable
 350          * and in each dark corner broken firewalls sit, sending random
 351          * errors ordered by their masters even this two messages finally lose
 352          * their original sense (even Linux sends invalid PORT_UNREACHs)
 353          *
 354          * Now we are in compliance with RFCs.
 355          *                                                      --ANK (980905)
 356          */
 357
 358         inet = inet_sk(sk);
 359         if (!sock_owned_by_user(sk) && inet->recverr) {
 360                 sk->sk_err = err;
 361                 sk_error_report(sk);
 362         } else /* Only an error on timeout */
 363                 sk->sk_err_soft = err;
 364 out:
 365         bh_unlock_sock(sk);
 366         sock_put(sk);
 367         return 0;
 368 }
 369
 370 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 371                                       __be32 src, __be32 dst)
 372 {
 373         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 374 }
 375
 376 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 377 {
 378         const struct inet_sock *inet = inet_sk(sk);
 379         struct dccp_hdr *dh = dccp_hdr(skb);
 380
 381         dccp_csum_outgoing(skb);
 382         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 383                                                  inet->inet_saddr,
 384                                                  inet->inet_daddr);
 385 }
 386 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 387
 388 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 389 {
 390         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 391                                            ip_hdr(skb)->saddr,
 392                                            dccp_hdr(skb)->dccph_dport,
 393                                            dccp_hdr(skb)->dccph_sport);
 394 }
 395
 396 /*
 397  * The three way handshake has completed - we got a valid ACK or DATAACK -
 398  * now create the new socket.
 399  *
 400  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 401  */
 402 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
 403                                        struct sk_buff *skb,
 404                                        struct request_sock *req,
 405                                        struct dst_entry *dst,
 406                                        struct request_sock *req_unhash,
 407                                        bool *own_req)
 408 {
 409         struct inet_request_sock *ireq;
 410         struct inet_sock *newinet;
 411         struct sock *newsk;
 412
 413         if (sk_acceptq_is_full(sk))
 414                 goto exit_overflow;
 415
 416         newsk = dccp_create_openreq_child(sk, req, skb);
 417         if (newsk == NULL)
 418                 goto exit_nonewsk;
 419
 420         newinet            = inet_sk(newsk);
 421         ireq               = inet_rsk(req);
 422         sk_daddr_set(newsk, ireq->ir_rmt_addr);
 423         sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
 424         newinet->inet_saddr     = ireq->ir_loc_addr;
 425         RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt));
 426         newinet->mc_index  = inet_iif(skb);
 427         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 428         newinet->inet_id   = prandom_u32();
 429
 430         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
 431                 goto put_and_exit;
 432
 433         sk_setup_caps(newsk, dst);
 434
 435         dccp_sync_mss(newsk, dst_mtu(dst));
 436
 437         if (__inet_inherit_port(sk, newsk) < 0)
 438                 goto put_and_exit;
 439         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), NULL);
 440         if (*own_req)
 441                 ireq->ireq_opt = NULL;
 442         else
 443                 newinet->inet_opt = NULL;
 444         return newsk;
 445
 446 exit_overflow:
 447         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 448 exit_nonewsk:
 449         dst_release(dst);
 450 exit:
 451         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
 452         return NULL;
 453 put_and_exit:
 454         newinet->inet_opt = NULL;
 455         inet_csk_prepare_forced_close(newsk);
 456         dccp_done(newsk);
 457         goto exit;
 458 }
 459 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 460
 461 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 462                                            struct sk_buff *skb)
 463 {
 464         struct rtable *rt;
 465         const struct iphdr *iph = ip_hdr(skb);
 466         struct flowi4 fl4 = {
 467                 .flowi4_oif = inet_iif(skb),
 468                 .daddr = iph->saddr,
 469                 .saddr = iph->daddr,
 470                 .flowi4_tos = RT_CONN_FLAGS(sk),
 471                 .flowi4_proto = sk->sk_protocol,
 472                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
 473                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
 474         };
 475
 476         security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
 477         rt = ip_route_output_flow(net, &fl4, sk);
 478         if (IS_ERR(rt)) {
 479                 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 480                 return NULL;
 481         }
 482
 483         return &rt->dst;
 484 }
 485
 486 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
 487 {
 488         int err = -1;
 489         struct sk_buff *skb;
 490         struct dst_entry *dst;
 491         struct flowi4 fl4;
 492
 493         dst = inet_csk_route_req(sk, &fl4, req);
 494         if (dst == NULL)
 495                 goto out;
 496
 497         skb = dccp_make_response(sk, dst, req);
 498         if (skb != NULL) {
 499                 const struct inet_request_sock *ireq = inet_rsk(req);
 500                 struct dccp_hdr *dh = dccp_hdr(skb);
 501
 502                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
 503                                                               ireq->ir_rmt_addr);
 504                 rcu_read_lock();
 505                 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 506                                             ireq->ir_rmt_addr,
 507                                             rcu_dereference(ireq->ireq_opt),
 508                                             inet_sk(sk)->tos);
 509                 rcu_read_unlock();
 510                 err = net_xmit_eval(err);
 511         }
 512
 513 out:
 514         dst_release(dst);
 515         return err;
 516 }
 517
 518 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
 519 {
 520         int err;
 521         const struct iphdr *rxiph;
 522         struct sk_buff *skb;
 523         struct dst_entry *dst;
 524         struct net *net = dev_net(skb_dst(rxskb)->dev);
 525         struct dccp_v4_pernet *pn;
 526         struct sock *ctl_sk;
 527
 528         /* Never send a reset in response to a reset. */
 529         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 530                 return;
 531
 532         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 533                 return;
 534
 535         pn = net_generic(net, dccp_v4_pernet_id);
 536         ctl_sk = pn->v4_ctl_sk;
 537         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 538         if (dst == NULL)
 539                 return;
 540
 541         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 542         if (skb == NULL)
 543                 goto out;
 544
 545         rxiph = ip_hdr(rxskb);
 546         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 547                                                                  rxiph->daddr);
 548         skb_dst_set(skb, dst_clone(dst));
 549
 550         local_bh_disable();
 551         bh_lock_sock(ctl_sk);
 552         err = ip_build_and_send_pkt(skb, ctl_sk,
 553                                     rxiph->daddr, rxiph->saddr, NULL,
 554                                     inet_sk(ctl_sk)->tos);
 555         bh_unlock_sock(ctl_sk);
 556
 557         if (net_xmit_eval(err) == 0) {
 558                 __DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 559                 __DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
 560         }
 561         local_bh_enable();
 562 out:
 563         dst_release(dst);
 564 }
 565
 566 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 567 {
 568         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 569         kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
 570 }
 571
 572 void dccp_syn_ack_timeout(const struct request_sock *req)
 573 {
 574 }
 575 EXPORT_SYMBOL(dccp_syn_ack_timeout);
 576
 577 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 578         .family         = PF_INET,
 579         .obj_size       = sizeof(struct dccp_request_sock),
 580         .rtx_syn_ack    = dccp_v4_send_response,
 581         .send_ack       = dccp_reqsk_send_ack,
 582         .destructor     = dccp_v4_reqsk_destructor,
 583         .send_reset     = dccp_v4_ctl_send_reset,
 584         .syn_ack_timeout = dccp_syn_ack_timeout,
 585 };
 586
 587 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 588 {
 589         struct inet_request_sock *ireq;
 590         struct request_sock *req;
 591         struct dccp_request_sock *dreq;
 592         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 593         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 594
 595         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 596         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 597                 return 0;       /* discard, don't send a reset here */
 598
 599         if (dccp_bad_service_code(sk, service)) {
 600                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 601                 goto drop;
 602         }
 603         /*
 604          * TW buckets are converted to open requests without
 605          * limitations, they conserve resources and peer is
 606          * evidently real one.
 607          */
 608         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 609         if (inet_csk_reqsk_queue_is_full(sk))
 610                 goto drop;
 611
 612         if (sk_acceptq_is_full(sk))
 613                 goto drop;
 614
 615         req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
 616         if (req == NULL)
 617                 goto drop;
 618
 619         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 620                 goto drop_and_free;
 621
 622         dreq = dccp_rsk(req);
 623         if (dccp_parse_options(sk, dreq, skb))
 624                 goto drop_and_free;
 625
 626         if (security_inet_conn_request(sk, skb, req))
 627                 goto drop_and_free;
 628
 629         ireq = inet_rsk(req);
 630         sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
 631         sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
 632         ireq->ir_mark = inet_request_mark(sk, skb);
 633         ireq->ireq_family = AF_INET;
 634         ireq->ir_iif = sk->sk_bound_dev_if;
 635
 636         /*
 637          * Step 3: Process LISTEN state
 638          *
 639          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 640          *
 641          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 642          */
 643         dreq->dreq_isr     = dcb->dccpd_seq;
 644         dreq->dreq_gsr     = dreq->dreq_isr;
 645         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 646         dreq->dreq_gss     = dreq->dreq_iss;
 647         dreq->dreq_service = service;
 648
 649         if (dccp_v4_send_response(sk, req))
 650                 goto drop_and_free;
 651
 652         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 653         reqsk_put(req);
 654         return 0;
 655
 656 drop_and_free:
 657         reqsk_free(req);
 658 drop:
 659         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 660         return -1;
 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
 698         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 699                 goto reset;
 700         return 0;
 701
 702 reset:
 703         dccp_v4_ctl_send_reset(sk, skb);
 704         kfree_skb(skb);
 705         return 0;
 706 }
 707 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 708
 709 /**
 710  *      dccp_invalid_packet  -  check for malformed packets
 711  *      @skb: Packet to validate
 712  *
 713  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 714  *      Packets that fail these checks are ignored and do not receive Resets.
 715  */
 716 int dccp_invalid_packet(struct sk_buff *skb)
 717 {
 718         const struct dccp_hdr *dh;
 719         unsigned int cscov;
 720         u8 dccph_doff;
 721
 722         if (skb->pkt_type != PACKET_HOST)
 723                 return 1;
 724
 725         /* If the packet is shorter than 12 bytes, drop packet and return */
 726         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 727                 DCCP_WARN("pskb_may_pull failed\n");
 728                 return 1;
 729         }
 730
 731         dh = dccp_hdr(skb);
 732
 733         /* If P.type is not understood, drop packet and return */
 734         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 735                 DCCP_WARN("invalid packet type\n");
 736                 return 1;
 737         }
 738
 739         /*
 740          * If P.Data Offset is too small for packet type, drop packet and return
 741          */
 742         dccph_doff = dh->dccph_doff;
 743         if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 744                 DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
 745                 return 1;
 746         }
 747         /*
 748          * If P.Data Offset is too large for packet, drop packet and return
 749          */
 750         if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
 751                 DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
 752                 return 1;
 753         }
 754         dh = dccp_hdr(skb);
 755         /*
 756          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 757          * has short sequence numbers), drop packet and return
 758          */
 759         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 760             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 761                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 762                           dccp_packet_name(dh->dccph_type));
 763                 return 1;
 764         }
 765
 766         /*
 767          * If P.CsCov is too large for the packet size, drop packet and return.
 768          * This must come _before_ checksumming (not as RFC 4340 suggests).
 769          */
 770         cscov = dccp_csum_coverage(skb);
 771         if (cscov > skb->len) {
 772                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 773                           dh->dccph_cscov, skb->len);
 774                 return 1;
 775         }
 776
 777         /* If header checksum is incorrect, drop packet and return.
 778          * (This step is completed in the AF-dependent functions.) */
 779         skb->csum = skb_checksum(skb, 0, cscov, 0);
 780
 781         return 0;
 782 }
 783 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 784
 785 /* this is called when real data arrives */
 786 static int dccp_v4_rcv(struct sk_buff *skb)
 787 {
 788         const struct dccp_hdr *dh;
 789         const struct iphdr *iph;
 790         bool refcounted;
 791         struct sock *sk;
 792         int min_cov;
 793
 794         /* Step 1: Check header basics */
 795
 796         if (dccp_invalid_packet(skb))
 797                 goto discard_it;
 798
 799         iph = ip_hdr(skb);
 800         /* Step 1: If header checksum is incorrect, drop packet and return */
 801         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 802                 DCCP_WARN("dropped packet with invalid checksum\n");
 803                 goto discard_it;
 804         }
 805
 806         dh = dccp_hdr(skb);
 807
 808         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 809         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 810
 811         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 812                       dccp_packet_name(dh->dccph_type),
 813                       &iph->saddr, ntohs(dh->dccph_sport),
 814                       &iph->daddr, ntohs(dh->dccph_dport),
 815                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 816
 817         if (dccp_packet_without_ack(skb)) {
 818                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 819                 dccp_pr_debug_cat("\n");
 820         } else {
 821                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 822                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 823                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 824         }
 825
 826 lookup:
 827         sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
 828                                dh->dccph_sport, dh->dccph_dport, 0, &refcounted);
 829         if (!sk) {
 830                 dccp_pr_debug("failed to look up flow ID in table and "
 831                               "get corresponding socket\n");
 832                 goto no_dccp_socket;
 833         }
 834
 835         /*
 836          * Step 2:
 837          *      ... or S.state == TIMEWAIT,
 838          *              Generate Reset(No Connection) unless P.type == Reset
 839          *              Drop packet and return
 840          */
 841         if (sk->sk_state == DCCP_TIME_WAIT) {
 842                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 843                 inet_twsk_put(inet_twsk(sk));
 844                 goto no_dccp_socket;
 845         }
 846
 847         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 848                 struct request_sock *req = inet_reqsk(sk);
 849                 struct sock *nsk;
 850
 851                 sk = req->rsk_listener;
 852                 if (unlikely(sk->sk_state != DCCP_LISTEN)) {
 853                         inet_csk_reqsk_queue_drop_and_put(sk, req);
 854                         goto lookup;
 855                 }
 856                 sock_hold(sk);
 857                 refcounted = true;
 858                 nsk = dccp_check_req(sk, skb, req);
 859                 if (!nsk) {
 860                         reqsk_put(req);
 861                         goto discard_and_relse;
 862                 }
 863                 if (nsk == sk) {
 864                         reqsk_put(req);
 865                 } else if (dccp_child_process(sk, nsk, skb)) {
 866                         dccp_v4_ctl_send_reset(sk, skb);
 867                         goto discard_and_relse;
 868                 } else {
 869                         sock_put(sk);
 870                         return 0;
 871                 }
 872         }
 873         /*
 874          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 875          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 876          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 877          */
 878         min_cov = dccp_sk(sk)->dccps_pcrlen;
 879         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 880                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 881                               dh->dccph_cscov, min_cov);
 882                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 883                  *         options (Section 11.7) with Drop Code 0, Protocol
 884                  *         Constraints."                                     */
 885                 goto discard_and_relse;
 886         }
 887
 888         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 889                 goto discard_and_relse;
 890         nf_reset_ct(skb);
 891
 892         return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
 893
 894 no_dccp_socket:
 895         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 896                 goto discard_it;
 897         /*
 898          * Step 2:
 899          *      If no socket ...
 900          *              Generate Reset(No Connection) unless P.type == Reset
 901          *              Drop packet and return
 902          */
 903         if (dh->dccph_type != DCCP_PKT_RESET) {
 904                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 905                                         DCCP_RESET_CODE_NO_CONNECTION;
 906                 dccp_v4_ctl_send_reset(sk, skb);
 907         }
 908
 909 discard_it:
 910         kfree_skb(skb);
 911         return 0;
 912
 913 discard_and_relse:
 914         if (refcounted)
 915                 sock_put(sk);
 916         goto discard_it;
 917 }
 918
 919 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 920         .queue_xmit        = ip_queue_xmit,
 921         .send_check        = dccp_v4_send_check,
 922         .rebuild_header    = inet_sk_rebuild_header,
 923         .conn_request      = dccp_v4_conn_request,
 924         .syn_recv_sock     = dccp_v4_request_recv_sock,
 925         .net_header_len    = sizeof(struct iphdr),
 926         .setsockopt        = ip_setsockopt,
 927         .getsockopt        = ip_getsockopt,
 928         .addr2sockaddr     = inet_csk_addr2sockaddr,
 929         .sockaddr_len      = sizeof(struct sockaddr_in),
 930 };
 931
 932 static int dccp_v4_init_sock(struct sock *sk)
 933 {
 934         static __u8 dccp_v4_ctl_sock_initialized;
 935         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 936
 937         if (err == 0) {
 938                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 939                         dccp_v4_ctl_sock_initialized = 1;
 940                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 941         }
 942
 943         return err;
 944 }
 945
 946 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 947         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 948 };
 949
 950 static struct proto dccp_v4_prot = {
 951         .name                   = "DCCP",
 952         .owner                  = THIS_MODULE,
 953         .close                  = dccp_close,
 954         .connect                = dccp_v4_connect,
 955         .disconnect             = dccp_disconnect,
 956         .ioctl                  = dccp_ioctl,
 957         .init                   = dccp_v4_init_sock,
 958         .setsockopt             = dccp_setsockopt,
 959         .getsockopt             = dccp_getsockopt,
 960         .sendmsg                = dccp_sendmsg,
 961         .recvmsg                = dccp_recvmsg,
 962         .backlog_rcv            = dccp_v4_do_rcv,
 963         .hash                   = inet_hash,
 964         .unhash                 = inet_unhash,
 965         .accept                 = inet_csk_accept,
 966         .get_port               = inet_csk_get_port,
 967         .shutdown               = dccp_shutdown,
 968         .destroy                = dccp_destroy_sock,
 969         .orphan_count           = &dccp_orphan_count,
 970         .max_header             = MAX_DCCP_HEADER,
 971         .obj_size               = sizeof(struct dccp_sock),
 972         .slab_flags             = SLAB_TYPESAFE_BY_RCU,
 973         .rsk_prot               = &dccp_request_sock_ops,
 974         .twsk_prot              = &dccp_timewait_sock_ops,
 975         .h.hashinfo             = &dccp_hashinfo,
 976 };
 977
 978 static const struct net_protocol dccp_v4_protocol = {
 979         .handler        = dccp_v4_rcv,
 980         .err_handler    = dccp_v4_err,
 981         .no_policy      = 1,
 982         .icmp_strict_tag_validation = 1,
 983 };
 984
 985 static const struct proto_ops inet_dccp_ops = {
 986         .family            = PF_INET,
 987         .owner             = THIS_MODULE,
 988         .release           = inet_release,
 989         .bind              = inet_bind,
 990         .connect           = inet_stream_connect,
 991         .socketpair        = sock_no_socketpair,
 992         .accept            = inet_accept,
 993         .getname           = inet_getname,
 994         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
 995         .poll              = dccp_poll,
 996         .ioctl             = inet_ioctl,
 997         .gettstamp         = sock_gettstamp,
 998         /* FIXME: work on inet_listen to rename it to sock_common_listen */
 999         .listen            = inet_dccp_listen,
1000         .shutdown          = inet_shutdown,
1001         .setsockopt        = sock_common_setsockopt,
1002         .getsockopt        = sock_common_getsockopt,
1003         .sendmsg           = inet_sendmsg,
1004         .recvmsg           = sock_common_recvmsg,
1005         .mmap              = sock_no_mmap,
1006         .sendpage          = sock_no_sendpage,
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         .flags          = INET_PROTOSW_ICSK,
1015 };
1016
1017 static int __net_init dccp_v4_init_net(struct net *net)
1018 {
1019         struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1020
1021         if (dccp_hashinfo.bhash == NULL)
1022                 return -ESOCKTNOSUPPORT;
1023
1024         return inet_ctl_sock_create(&pn->v4_ctl_sk, PF_INET,
1025                                     SOCK_DCCP, IPPROTO_DCCP, net);
1026 }
1027
1028 static void __net_exit dccp_v4_exit_net(struct net *net)
1029 {
1030         struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1031
1032         inet_ctl_sock_destroy(pn->v4_ctl_sk);
1033 }
1034
1035 static void __net_exit dccp_v4_exit_batch(struct list_head *net_exit_list)
1036 {
1037         inet_twsk_purge(&dccp_hashinfo, AF_INET);
1038 }
1039
1040 static struct pernet_operations dccp_v4_ops = {
1041         .init   = dccp_v4_init_net,
1042         .exit   = dccp_v4_exit_net,
1043         .exit_batch = dccp_v4_exit_batch,
1044         .id     = &dccp_v4_pernet_id,
1045         .size   = sizeof(struct dccp_v4_pernet),
1046 };
1047
1048 static int __init dccp_v4_init(void)
1049 {
1050         int err = proto_register(&dccp_v4_prot, 1);
1051
1052         if (err)
1053                 goto out;
1054
1055         inet_register_protosw(&dccp_v4_protosw);
1056
1057         err = register_pernet_subsys(&dccp_v4_ops);
1058         if (err)
1059                 goto out_destroy_ctl_sock;
1060
1061         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1062         if (err)
1063                 goto out_proto_unregister;
1064
1065 out:
1066         return err;
1067 out_proto_unregister:
1068         unregister_pernet_subsys(&dccp_v4_ops);
1069 out_destroy_ctl_sock:
1070         inet_unregister_protosw(&dccp_v4_protosw);
1071         proto_unregister(&dccp_v4_prot);
1072         goto out;
1073 }
1074
1075 static void __exit dccp_v4_exit(void)
1076 {
1077         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1078         unregister_pernet_subsys(&dccp_v4_ops);
1079         inet_unregister_protosw(&dccp_v4_protosw);
1080         proto_unregister(&dccp_v4_prot);
1081 }
1082
1083 module_init(dccp_v4_init);
1084 module_exit(dccp_v4_exit);
1085
1086 /*
1087  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1088  * values directly, Also cover the case where the protocol is not specified,
1089  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1090  */
1091 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1092 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1093 MODULE_LICENSE("GPL");
1094 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1095 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");