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
  27 #include "ackvec.h"
  28 #include "ccid.h"
  29 #include "dccp.h"
  30 #include "feat.h"
  31
  32 /*
  33  * The per-net dccp.v4_ctl_sk socket is used for responding to
  34  * the Out-of-the-blue (OOTB) packets. A control sock will be created
  35  * for this socket at the initialization time.
  36  */
  37
  38 int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
  39 {
  40         const struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
  41         struct inet_sock *inet = inet_sk(sk);
  42         struct dccp_sock *dp = dccp_sk(sk);
  43         __be16 orig_sport, orig_dport;
  44         __be32 daddr, nexthop;
  45         struct flowi4 *fl4;
  46         struct rtable *rt;
  47         int err;
  48         struct ip_options_rcu *inet_opt;
  49
  50         dp->dccps_role = DCCP_ROLE_CLIENT;
  51
  52         if (addr_len < sizeof(struct sockaddr_in))
  53                 return -EINVAL;
  54
  55         if (usin->sin_family != AF_INET)
  56                 return -EAFNOSUPPORT;
  57
  58         nexthop = daddr = usin->sin_addr.s_addr;
  59
  60         inet_opt = rcu_dereference_protected(inet->inet_opt,
  61                                              lockdep_sock_is_held(sk));
  62         if (inet_opt != NULL && inet_opt->opt.srr) {
  63                 if (daddr == 0)
  64                         return -EINVAL;
  65                 nexthop = inet_opt->opt.faddr;
  66         }
  67
  68         orig_sport = inet->inet_sport;
  69         orig_dport = usin->sin_port;
  70         fl4 = &inet->cork.fl.u.ip4;
  71         rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
  72                               RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
  73                               IPPROTO_DCCP,
  74                               orig_sport, orig_dport, sk);
  75         if (IS_ERR(rt))
  76                 return PTR_ERR(rt);
  77
  78         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  79                 ip_rt_put(rt);
  80                 return -ENETUNREACH;
  81         }
  82
  83         if (inet_opt == NULL || !inet_opt->opt.srr)
  84                 daddr = fl4->daddr;
  85
  86         if (inet->inet_saddr == 0)
  87                 inet->inet_saddr = fl4->saddr;
  88         sk_rcv_saddr_set(sk, inet->inet_saddr);
  89         inet->inet_dport = usin->sin_port;
  90         sk_daddr_set(sk, daddr);
  91
  92         inet_csk(sk)->icsk_ext_hdr_len = 0;
  93         if (inet_opt)
  94                 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
  95         /*
  96          * Socket identity is still unknown (sport may be zero).
  97          * However we set state to DCCP_REQUESTING and not releasing socket
  98          * lock select source port, enter ourselves into the hash tables and
  99          * complete initialization after this.
 100          */
 101         dccp_set_state(sk, DCCP_REQUESTING);
 102         err = inet_hash_connect(&dccp_death_row, sk);
 103         if (err != 0)
 104                 goto failure;
 105
 106         rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 107                                inet->inet_sport, inet->inet_dport, sk);
 108         if (IS_ERR(rt)) {
 109                 err = PTR_ERR(rt);
 110                 rt = NULL;
 111                 goto failure;
 112         }
 113         /* OK, now commit destination to socket.  */
 114         sk_setup_caps(sk, &rt->dst);
 115
 116         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
 117                                                     inet->inet_daddr,
 118                                                     inet->inet_sport,
 119                                                     inet->inet_dport);
 120         inet->inet_id = prandom_u32();
 121
 122         err = dccp_connect(sk);
 123         rt = NULL;
 124         if (err != 0)
 125                 goto failure;
 126 out:
 127         return err;
 128 failure:
 129         /*
 130          * This unhashes the socket and releases the local port, if necessary.
 131          */
 132         dccp_set_state(sk, DCCP_CLOSED);
 133         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
 134                 inet_reset_saddr(sk);
 135         ip_rt_put(rt);
 136         sk->sk_route_caps = 0;
 137         inet->inet_dport = 0;
 138         goto out;
 139 }
 140 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 141
 142 /*
 143  * This routine does path mtu discovery as defined in RFC1191.
 144  */
 145 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 146                                           const struct iphdr *iph,
 147                                           u32 mtu)
 148 {
 149         struct dst_entry *dst;
 150         const struct inet_sock *inet = inet_sk(sk);
 151         const struct dccp_sock *dp = dccp_sk(sk);
 152
 153         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 154          * send out by Linux are always < 576bytes so they should go through
 155          * unfragmented).
 156          */
 157         if (sk->sk_state == DCCP_LISTEN)
 158                 return;
 159
 160         dst = inet_csk_update_pmtu(sk, mtu);
 161         if (!dst)
 162                 return;
 163
 164         /* Something is about to be wrong... Remember soft error
 165          * for the case, if this connection will not able to recover.
 166          */
 167         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 168                 sk->sk_err_soft = EMSGSIZE;
 169
 170         mtu = dst_mtu(dst);
 171
 172         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 173             ip_sk_accept_pmtu(sk) &&
 174             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 175                 dccp_sync_mss(sk, mtu);
 176
 177                 /*
 178                  * From RFC 4340, sec. 14.1:
 179                  *
 180                  *      DCCP-Sync packets are the best choice for upward
 181                  *      probing, since DCCP-Sync probes do not risk application
 182                  *      data loss.
 183                  */
 184                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 185         } /* else let the usual retransmit timer handle it */
 186 }
 187
 188 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
 189 {
 190         struct dst_entry *dst = __sk_dst_check(sk, 0);
 191
 192         if (dst)
 193                 dst->ops->redirect(dst, sk, skb);
 194 }
 195
 196 void dccp_req_err(struct sock *sk, u64 seq)
 197         {
 198         struct request_sock *req = inet_reqsk(sk);
 199         struct net *net = sock_net(sk);
 200
 201         /*
 202          * ICMPs are not backlogged, hence we cannot get an established
 203          * socket here.
 204          */
 205         if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
 206                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 207         } else {
 208                 /*
 209                  * Still in RESPOND, just remove it silently.
 210                  * There is no good way to pass the error to the newly
 211                  * created socket, and POSIX does not want network
 212                  * errors returned from accept().
 213                  */
 214                 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
 215         }
 216         reqsk_put(req);
 217 }
 218 EXPORT_SYMBOL(dccp_req_err);
 219
 220 /*
 221  * This routine is called by the ICMP module when it gets some sort of error
 222  * condition. If err < 0 then the socket should be closed and the error
 223  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 224  * After adjustment header points to the first 8 bytes of the tcp header. We
 225  * need to find the appropriate port.
 226  *
 227  * The locking strategy used here is very "optimistic". When someone else
 228  * accesses the socket the ICMP is just dropped and for some paths there is no
 229  * check at all. A more general error queue to queue errors for later handling
 230  * is probably better.
 231  */
 232 static int dccp_v4_err(struct sk_buff *skb, u32 info)
 233 {
 234         const struct iphdr *iph = (struct iphdr *)skb->data;
 235         const u8 offset = iph->ihl << 2;
 236         const struct dccp_hdr *dh;
 237         struct dccp_sock *dp;
 238         struct inet_sock *inet;
 239         const int type = icmp_hdr(skb)->type;
 240         const int code = icmp_hdr(skb)->code;
 241         struct sock *sk;
 242         __u64 seq;
 243         int err;
 244         struct net *net = dev_net(skb->dev);
 245
 246         /* For the first __dccp_basic_hdr_len() check, we only need dh->dccph_x,
 247          * which is in byte 7 of the dccp header.
 248          * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us.
 249          *
 250          * Later on, we want to access the sequence number fields, which are
 251          * beyond 8 bytes, so we have to pskb_may_pull() ourselves.
 252          */
 253         dh = (struct dccp_hdr *)(skb->data + offset);
 254         if (!pskb_may_pull(skb, offset + __dccp_basic_hdr_len(dh)))
 255                 return -EINVAL;
 256         iph = (struct iphdr *)skb->data;
 257         dh = (struct dccp_hdr *)(skb->data + offset);
 258
 259         sk = __inet_lookup_established(net, &dccp_hashinfo,
 260                                        iph->daddr, dh->dccph_dport,
 261                                        iph->saddr, ntohs(dh->dccph_sport),
 262                                        inet_iif(skb), 0);
 263         if (!sk) {
 264                 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 265                 return -ENOENT;
 266         }
 267
 268         if (sk->sk_state == DCCP_TIME_WAIT) {
 269                 inet_twsk_put(inet_twsk(sk));
 270                 return 0;
 271         }
 272         seq = dccp_hdr_seq(dh);
 273         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 274                 dccp_req_err(sk, seq);
 275                 return 0;
 276         }
 277
 278         bh_lock_sock(sk);
 279         /* If too many ICMPs get dropped on busy
 280          * servers this needs to be solved differently.
 281          */
 282         if (sock_owned_by_user(sk))
 283                 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
 284
 285         if (sk->sk_state == DCCP_CLOSED)
 286                 goto out;
 287
 288         dp = dccp_sk(sk);
 289         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 290             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 291                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 292                 goto out;
 293         }
 294
 295         switch (type) {
 296         case ICMP_REDIRECT:
 297                 if (!sock_owned_by_user(sk))
 298                         dccp_do_redirect(skb, sk);
 299                 goto out;
 300         case ICMP_SOURCE_QUENCH:
 301                 /* Just silently ignore these. */
 302                 goto out;
 303         case ICMP_PARAMETERPROB:
 304                 err = EPROTO;
 305                 break;
 306         case ICMP_DEST_UNREACH:
 307                 if (code > NR_ICMP_UNREACH)
 308                         goto out;
 309
 310                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 311                         if (!sock_owned_by_user(sk))
 312                                 dccp_do_pmtu_discovery(sk, iph, info);
 313                         goto out;
 314                 }
 315
 316                 err = icmp_err_convert[code].errno;
 317                 break;
 318         case ICMP_TIME_EXCEEDED:
 319                 err = EHOSTUNREACH;
 320                 break;
 321         default:
 322                 goto out;
 323         }
 324
 325         switch (sk->sk_state) {
 326         case DCCP_REQUESTING:
 327         case DCCP_RESPOND:
 328                 if (!sock_owned_by_user(sk)) {
 329                         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 330                         sk->sk_err = err;
 331
 332                         sk->sk_error_report(sk);
 333
 334                         dccp_done(sk);
 335                 } else
 336                         sk->sk_err_soft = err;
 337                 goto out;
 338         }
 339
 340         /* If we've already connected we will keep trying
 341          * until we time out, or the user gives up.
 342          *
 343          * rfc1122 4.2.3.9 allows to consider as hard errors
 344          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 345          * but it is obsoleted by pmtu discovery).
 346          *
 347          * Note, that in modern internet, where routing is unreliable
 348          * and in each dark corner broken firewalls sit, sending random
 349          * errors ordered by their masters even this two messages finally lose
 350          * their original sense (even Linux sends invalid PORT_UNREACHs)
 351          *
 352          * Now we are in compliance with RFCs.
 353          *                                                      --ANK (980905)
 354          */
 355
 356         inet = inet_sk(sk);
 357         if (!sock_owned_by_user(sk) && inet->recverr) {
 358                 sk->sk_err = err;
 359                 sk->sk_error_report(sk);
 360         } else /* Only an error on timeout */
 361                 sk->sk_err_soft = err;
 362 out:
 363         bh_unlock_sock(sk);
 364         sock_put(sk);
 365         return 0;
 366 }
 367
 368 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 369                                       __be32 src, __be32 dst)
 370 {
 371         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 372 }
 373
 374 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 375 {
 376         const struct inet_sock *inet = inet_sk(sk);
 377         struct dccp_hdr *dh = dccp_hdr(skb);
 378
 379         dccp_csum_outgoing(skb);
 380         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 381                                                  inet->inet_saddr,
 382                                                  inet->inet_daddr);
 383 }
 384 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 385
 386 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 387 {
 388         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 389                                            ip_hdr(skb)->saddr,
 390                                            dccp_hdr(skb)->dccph_dport,
 391                                            dccp_hdr(skb)->dccph_sport);
 392 }
 393
 394 /*
 395  * The three way handshake has completed - we got a valid ACK or DATAACK -
 396  * now create the new socket.
 397  *
 398  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 399  */
 400 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
 401                                        struct sk_buff *skb,
 402                                        struct request_sock *req,
 403                                        struct dst_entry *dst,
 404                                        struct request_sock *req_unhash,
 405                                        bool *own_req)
 406 {
 407         struct inet_request_sock *ireq;
 408         struct inet_sock *newinet;
 409         struct sock *newsk;
 410
 411         if (sk_acceptq_is_full(sk))
 412                 goto exit_overflow;
 413
 414         newsk = dccp_create_openreq_child(sk, req, skb);
 415         if (newsk == NULL)
 416                 goto exit_nonewsk;
 417
 418         newinet            = inet_sk(newsk);
 419         ireq               = inet_rsk(req);
 420         sk_daddr_set(newsk, ireq->ir_rmt_addr);
 421         sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
 422         newinet->inet_saddr     = ireq->ir_loc_addr;
 423         RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt));
 424         newinet->mc_index  = inet_iif(skb);
 425         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 426         newinet->inet_id   = prandom_u32();
 427
 428         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
 429                 goto put_and_exit;
 430
 431         sk_setup_caps(newsk, dst);
 432
 433         dccp_sync_mss(newsk, dst_mtu(dst));
 434
 435         if (__inet_inherit_port(sk, newsk) < 0)
 436                 goto put_and_exit;
 437         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), NULL);
 438         if (*own_req)
 439                 ireq->ireq_opt = NULL;
 440         else
 441                 newinet->inet_opt = NULL;
 442         return newsk;
 443
 444 exit_overflow:
 445         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 446 exit_nonewsk:
 447         dst_release(dst);
 448 exit:
 449         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
 450         return NULL;
 451 put_and_exit:
 452         newinet->inet_opt = NULL;
 453         inet_csk_prepare_forced_close(newsk);
 454         dccp_done(newsk);
 455         goto exit;
 456 }
 457 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 458
 459 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 460                                            struct sk_buff *skb)
 461 {
 462         struct rtable *rt;
 463         const struct iphdr *iph = ip_hdr(skb);
 464         struct flowi4 fl4 = {
 465                 .flowi4_oif = inet_iif(skb),
 466                 .daddr = iph->saddr,
 467                 .saddr = iph->daddr,
 468                 .flowi4_tos = RT_CONN_FLAGS(sk),
 469                 .flowi4_proto = sk->sk_protocol,
 470                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
 471                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
 472         };
 473
 474         security_skb_classify_flow(skb, flowi4_to_flowi(&fl4));
 475         rt = ip_route_output_flow(net, &fl4, sk);
 476         if (IS_ERR(rt)) {
 477                 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 478                 return NULL;
 479         }
 480
 481         return &rt->dst;
 482 }
 483
 484 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
 485 {
 486         int err = -1;
 487         struct sk_buff *skb;
 488         struct dst_entry *dst;
 489         struct flowi4 fl4;
 490
 491         dst = inet_csk_route_req(sk, &fl4, req);
 492         if (dst == NULL)
 493                 goto out;
 494
 495         skb = dccp_make_response(sk, dst, req);
 496         if (skb != NULL) {
 497                 const struct inet_request_sock *ireq = inet_rsk(req);
 498                 struct dccp_hdr *dh = dccp_hdr(skb);
 499
 500                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
 501                                                               ireq->ir_rmt_addr);
 502                 rcu_read_lock();
 503                 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 504                                             ireq->ir_rmt_addr,
 505                                             rcu_dereference(ireq->ireq_opt));
 506                 rcu_read_unlock();
 507                 err = net_xmit_eval(err);
 508         }
 509
 510 out:
 511         dst_release(dst);
 512         return err;
 513 }
 514
 515 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
 516 {
 517         int err;
 518         const struct iphdr *rxiph;
 519         struct sk_buff *skb;
 520         struct dst_entry *dst;
 521         struct net *net = dev_net(skb_dst(rxskb)->dev);
 522         struct sock *ctl_sk = net->dccp.v4_ctl_sk;
 523
 524         /* Never send a reset in response to a reset. */
 525         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 526                 return;
 527
 528         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 529                 return;
 530
 531         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 532         if (dst == NULL)
 533                 return;
 534
 535         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 536         if (skb == NULL)
 537                 goto out;
 538
 539         rxiph = ip_hdr(rxskb);
 540         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 541                                                                  rxiph->daddr);
 542         skb_dst_set(skb, dst_clone(dst));
 543
 544         local_bh_disable();
 545         bh_lock_sock(ctl_sk);
 546         err = ip_build_and_send_pkt(skb, ctl_sk,
 547                                     rxiph->daddr, rxiph->saddr, NULL);
 548         bh_unlock_sock(ctl_sk);
 549
 550         if (net_xmit_eval(err) == 0) {
 551                 __DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 552                 __DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
 553         }
 554         local_bh_enable();
 555 out:
 556         dst_release(dst);
 557 }
 558
 559 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 560 {
 561         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 562         kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
 563 }
 564
 565 void dccp_syn_ack_timeout(const struct request_sock *req)
 566 {
 567 }
 568 EXPORT_SYMBOL(dccp_syn_ack_timeout);
 569
 570 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 571         .family         = PF_INET,
 572         .obj_size       = sizeof(struct dccp_request_sock),
 573         .rtx_syn_ack    = dccp_v4_send_response,
 574         .send_ack       = dccp_reqsk_send_ack,
 575         .destructor     = dccp_v4_reqsk_destructor,
 576         .send_reset     = dccp_v4_ctl_send_reset,
 577         .syn_ack_timeout = dccp_syn_ack_timeout,
 578 };
 579
 580 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 581 {
 582         struct inet_request_sock *ireq;
 583         struct request_sock *req;
 584         struct dccp_request_sock *dreq;
 585         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 586         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 587
 588         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 589         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 590                 return 0;       /* discard, don't send a reset here */
 591
 592         if (dccp_bad_service_code(sk, service)) {
 593                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 594                 goto drop;
 595         }
 596         /*
 597          * TW buckets are converted to open requests without
 598          * limitations, they conserve resources and peer is
 599          * evidently real one.
 600          */
 601         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 602         if (inet_csk_reqsk_queue_is_full(sk))
 603                 goto drop;
 604
 605         if (sk_acceptq_is_full(sk))
 606                 goto drop;
 607
 608         req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
 609         if (req == NULL)
 610                 goto drop;
 611
 612         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 613                 goto drop_and_free;
 614
 615         dreq = dccp_rsk(req);
 616         if (dccp_parse_options(sk, dreq, skb))
 617                 goto drop_and_free;
 618
 619         if (security_inet_conn_request(sk, skb, req))
 620                 goto drop_and_free;
 621
 622         ireq = inet_rsk(req);
 623         sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
 624         sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
 625         ireq->ir_mark = inet_request_mark(sk, skb);
 626         ireq->ireq_family = AF_INET;
 627         ireq->ir_iif = sk->sk_bound_dev_if;
 628
 629         /*
 630          * Step 3: Process LISTEN state
 631          *
 632          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 633          *
 634          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 635          */
 636         dreq->dreq_isr     = dcb->dccpd_seq;
 637         dreq->dreq_gsr     = dreq->dreq_isr;
 638         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 639         dreq->dreq_gss     = dreq->dreq_iss;
 640         dreq->dreq_service = service;
 641
 642         if (dccp_v4_send_response(sk, req))
 643                 goto drop_and_free;
 644
 645         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 646         reqsk_put(req);
 647         return 0;
 648
 649 drop_and_free:
 650         reqsk_free(req);
 651 drop:
 652         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 653         return -1;
 654 }
 655 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 656
 657 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 658 {
 659         struct dccp_hdr *dh = dccp_hdr(skb);
 660
 661         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 662                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 663                         goto reset;
 664                 return 0;
 665         }
 666
 667         /*
 668          *  Step 3: Process LISTEN state
 669          *       If P.type == Request or P contains a valid Init Cookie option,
 670          *            (* Must scan the packet's options to check for Init
 671          *               Cookies.  Only Init Cookies are processed here,
 672          *               however; other options are processed in Step 8.  This
 673          *               scan need only be performed if the endpoint uses Init
 674          *               Cookies *)
 675          *            (* Generate a new socket and switch to that socket *)
 676          *            Set S := new socket for this port pair
 677          *            S.state = RESPOND
 678          *            Choose S.ISS (initial seqno) or set from Init Cookies
 679          *            Initialize S.GAR := S.ISS
 680          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 681          *            Continue with S.state == RESPOND
 682          *            (* A Response packet will be generated in Step 11 *)
 683          *       Otherwise,
 684          *            Generate Reset(No Connection) unless P.type == Reset
 685          *            Drop packet and return
 686          *
 687          * NOTE: the check for the packet types is done in
 688          *       dccp_rcv_state_process
 689          */
 690
 691         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 692                 goto reset;
 693         return 0;
 694
 695 reset:
 696         dccp_v4_ctl_send_reset(sk, skb);
 697         kfree_skb(skb);
 698         return 0;
 699 }
 700 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 701
 702 /**
 703  *      dccp_invalid_packet  -  check for malformed packets
 704  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 705  *      Packets that fail these checks are ignored and do not receive Resets.
 706  */
 707 int dccp_invalid_packet(struct sk_buff *skb)
 708 {
 709         const struct dccp_hdr *dh;
 710         unsigned int cscov;
 711         u8 dccph_doff;
 712
 713         if (skb->pkt_type != PACKET_HOST)
 714                 return 1;
 715
 716         /* If the packet is shorter than 12 bytes, drop packet and return */
 717         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 718                 DCCP_WARN("pskb_may_pull failed\n");
 719                 return 1;
 720         }
 721
 722         dh = dccp_hdr(skb);
 723
 724         /* If P.type is not understood, drop packet and return */
 725         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 726                 DCCP_WARN("invalid packet type\n");
 727                 return 1;
 728         }
 729
 730         /*
 731          * If P.Data Offset is too small for packet type, drop packet and return
 732          */
 733         dccph_doff = dh->dccph_doff;
 734         if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 735                 DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
 736                 return 1;
 737         }
 738         /*
 739          * If P.Data Offset is too too large for packet, drop packet and return
 740          */
 741         if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
 742                 DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
 743                 return 1;
 744         }
 745         dh = dccp_hdr(skb);
 746         /*
 747          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 748          * has short sequence numbers), drop packet and return
 749          */
 750         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 751             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 752                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 753                           dccp_packet_name(dh->dccph_type));
 754                 return 1;
 755         }
 756
 757         /*
 758          * If P.CsCov is too large for the packet size, drop packet and return.
 759          * This must come _before_ checksumming (not as RFC 4340 suggests).
 760          */
 761         cscov = dccp_csum_coverage(skb);
 762         if (cscov > skb->len) {
 763                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 764                           dh->dccph_cscov, skb->len);
 765                 return 1;
 766         }
 767
 768         /* If header checksum is incorrect, drop packet and return.
 769          * (This step is completed in the AF-dependent functions.) */
 770         skb->csum = skb_checksum(skb, 0, cscov, 0);
 771
 772         return 0;
 773 }
 774 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 775
 776 /* this is called when real data arrives */
 777 static int dccp_v4_rcv(struct sk_buff *skb)
 778 {
 779         const struct dccp_hdr *dh;
 780         const struct iphdr *iph;
 781         bool refcounted;
 782         struct sock *sk;
 783         int min_cov;
 784
 785         /* Step 1: Check header basics */
 786
 787         if (dccp_invalid_packet(skb))
 788                 goto discard_it;
 789
 790         iph = ip_hdr(skb);
 791         /* Step 1: If header checksum is incorrect, drop packet and return */
 792         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 793                 DCCP_WARN("dropped packet with invalid checksum\n");
 794                 goto discard_it;
 795         }
 796
 797         dh = dccp_hdr(skb);
 798
 799         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 800         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 801
 802         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 803                       dccp_packet_name(dh->dccph_type),
 804                       &iph->saddr, ntohs(dh->dccph_sport),
 805                       &iph->daddr, ntohs(dh->dccph_dport),
 806                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 807
 808         if (dccp_packet_without_ack(skb)) {
 809                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 810                 dccp_pr_debug_cat("\n");
 811         } else {
 812                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 813                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 814                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 815         }
 816
 817 lookup:
 818         sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
 819                                dh->dccph_sport, dh->dccph_dport, 0, &refcounted);
 820         if (!sk) {
 821                 dccp_pr_debug("failed to look up flow ID in table and "
 822                               "get corresponding socket\n");
 823                 goto no_dccp_socket;
 824         }
 825
 826         /*
 827          * Step 2:
 828          *      ... or S.state == TIMEWAIT,
 829          *              Generate Reset(No Connection) unless P.type == Reset
 830          *              Drop packet and return
 831          */
 832         if (sk->sk_state == DCCP_TIME_WAIT) {
 833                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 834                 inet_twsk_put(inet_twsk(sk));
 835                 goto no_dccp_socket;
 836         }
 837
 838         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 839                 struct request_sock *req = inet_reqsk(sk);
 840                 struct sock *nsk;
 841
 842                 sk = req->rsk_listener;
 843                 if (unlikely(sk->sk_state != DCCP_LISTEN)) {
 844                         inet_csk_reqsk_queue_drop_and_put(sk, req);
 845                         goto lookup;
 846                 }
 847                 sock_hold(sk);
 848                 refcounted = true;
 849                 nsk = dccp_check_req(sk, skb, req);
 850                 if (!nsk) {
 851                         reqsk_put(req);
 852                         goto discard_and_relse;
 853                 }
 854                 if (nsk == sk) {
 855                         reqsk_put(req);
 856                 } else if (dccp_child_process(sk, nsk, skb)) {
 857                         dccp_v4_ctl_send_reset(sk, skb);
 858                         goto discard_and_relse;
 859                 } else {
 860                         sock_put(sk);
 861                         return 0;
 862                 }
 863         }
 864         /*
 865          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 866          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 867          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 868          */
 869         min_cov = dccp_sk(sk)->dccps_pcrlen;
 870         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 871                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 872                               dh->dccph_cscov, min_cov);
 873                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 874                  *         options (Section 11.7) with Drop Code 0, Protocol
 875                  *         Constraints."                                     */
 876                 goto discard_and_relse;
 877         }
 878
 879         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 880                 goto discard_and_relse;
 881         nf_reset_ct(skb);
 882
 883         return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
 884
 885 no_dccp_socket:
 886         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 887                 goto discard_it;
 888         /*
 889          * Step 2:
 890          *      If no socket ...
 891          *              Generate Reset(No Connection) unless P.type == Reset
 892          *              Drop packet and return
 893          */
 894         if (dh->dccph_type != DCCP_PKT_RESET) {
 895                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 896                                         DCCP_RESET_CODE_NO_CONNECTION;
 897                 dccp_v4_ctl_send_reset(sk, skb);
 898         }
 899
 900 discard_it:
 901         kfree_skb(skb);
 902         return 0;
 903
 904 discard_and_relse:
 905         if (refcounted)
 906                 sock_put(sk);
 907         goto discard_it;
 908 }
 909
 910 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 911         .queue_xmit        = ip_queue_xmit,
 912         .send_check        = dccp_v4_send_check,
 913         .rebuild_header    = inet_sk_rebuild_header,
 914         .conn_request      = dccp_v4_conn_request,
 915         .syn_recv_sock     = dccp_v4_request_recv_sock,
 916         .net_header_len    = sizeof(struct iphdr),
 917         .setsockopt        = ip_setsockopt,
 918         .getsockopt        = ip_getsockopt,
 919         .addr2sockaddr     = inet_csk_addr2sockaddr,
 920         .sockaddr_len      = sizeof(struct sockaddr_in),
 921 #ifdef CONFIG_COMPAT
 922         .compat_setsockopt = compat_ip_setsockopt,
 923         .compat_getsockopt = compat_ip_getsockopt,
 924 #endif
 925 };
 926
 927 static int dccp_v4_init_sock(struct sock *sk)
 928 {
 929         static __u8 dccp_v4_ctl_sock_initialized;
 930         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 931
 932         if (err == 0) {
 933                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 934                         dccp_v4_ctl_sock_initialized = 1;
 935                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 936         }
 937
 938         return err;
 939 }
 940
 941 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 942         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 943 };
 944
 945 static struct proto dccp_v4_prot = {
 946         .name                   = "DCCP",
 947         .owner                  = THIS_MODULE,
 948         .close                  = dccp_close,
 949         .connect                = dccp_v4_connect,
 950         .disconnect             = dccp_disconnect,
 951         .ioctl                  = dccp_ioctl,
 952         .init                   = dccp_v4_init_sock,
 953         .setsockopt             = dccp_setsockopt,
 954         .getsockopt             = dccp_getsockopt,
 955         .sendmsg                = dccp_sendmsg,
 956         .recvmsg                = dccp_recvmsg,
 957         .backlog_rcv            = dccp_v4_do_rcv,
 958         .hash                   = inet_hash,
 959         .unhash                 = inet_unhash,
 960         .accept                 = inet_csk_accept,
 961         .get_port               = inet_csk_get_port,
 962         .shutdown               = dccp_shutdown,
 963         .destroy                = dccp_destroy_sock,
 964         .orphan_count           = &dccp_orphan_count,
 965         .max_header             = MAX_DCCP_HEADER,
 966         .obj_size               = sizeof(struct dccp_sock),
 967         .slab_flags             = SLAB_TYPESAFE_BY_RCU,
 968         .rsk_prot               = &dccp_request_sock_ops,
 969         .twsk_prot              = &dccp_timewait_sock_ops,
 970         .h.hashinfo             = &dccp_hashinfo,
 971 #ifdef CONFIG_COMPAT
 972         .compat_setsockopt      = compat_dccp_setsockopt,
 973         .compat_getsockopt      = compat_dccp_getsockopt,
 974 #endif
 975 };
 976
 977 static const struct net_protocol dccp_v4_protocol = {
 978         .handler        = dccp_v4_rcv,
 979         .err_handler    = dccp_v4_err,
 980         .no_policy      = 1,
 981         .netns_ok       = 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 #ifdef CONFIG_COMPAT
1008         .compat_setsockopt = compat_sock_common_setsockopt,
1009         .compat_getsockopt = compat_sock_common_getsockopt,
1010 #endif
1011 };
1012
1013 static struct inet_protosw dccp_v4_protosw = {
1014         .type           = SOCK_DCCP,
1015         .protocol       = IPPROTO_DCCP,
1016         .prot           = &dccp_v4_prot,
1017         .ops            = &inet_dccp_ops,
1018         .flags          = INET_PROTOSW_ICSK,
1019 };
1020
1021 static int __net_init dccp_v4_init_net(struct net *net)
1022 {
1023         if (dccp_hashinfo.bhash == NULL)
1024                 return -ESOCKTNOSUPPORT;
1025
1026         return inet_ctl_sock_create(&net->dccp.v4_ctl_sk, PF_INET,
1027                                     SOCK_DCCP, IPPROTO_DCCP, net);
1028 }
1029
1030 static void __net_exit dccp_v4_exit_net(struct net *net)
1031 {
1032         inet_ctl_sock_destroy(net->dccp.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 };
1045
1046 static int __init dccp_v4_init(void)
1047 {
1048         int err = proto_register(&dccp_v4_prot, 1);
1049
1050         if (err)
1051                 goto out;
1052
1053         inet_register_protosw(&dccp_v4_protosw);
1054
1055         err = register_pernet_subsys(&dccp_v4_ops);
1056         if (err)
1057                 goto out_destroy_ctl_sock;
1058
1059         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1060         if (err)
1061                 goto out_proto_unregister;
1062
1063 out:
1064         return err;
1065 out_proto_unregister:
1066         unregister_pernet_subsys(&dccp_v4_ops);
1067 out_destroy_ctl_sock:
1068         inet_unregister_protosw(&dccp_v4_protosw);
1069         proto_unregister(&dccp_v4_prot);
1070         goto out;
1071 }
1072
1073 static void __exit dccp_v4_exit(void)
1074 {
1075         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1076         unregister_pernet_subsys(&dccp_v4_ops);
1077         inet_unregister_protosw(&dccp_v4_protosw);
1078         proto_unregister(&dccp_v4_prot);
1079 }
1080
1081 module_init(dccp_v4_init);
1082 module_exit(dccp_v4_exit);
1083
1084 /*
1085  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1086  * values directly, Also cover the case where the protocol is not specified,
1087  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1088  */
1089 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1090 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1091 MODULE_LICENSE("GPL");
1092 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1093 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");