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                               sk->sk_bound_dev_if, IPPROTO_DCCP, orig_sport,
  80                               orig_dport, sk);
  81         if (IS_ERR(rt))
  82                 return PTR_ERR(rt);
  83
  84         if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
  85                 ip_rt_put(rt);
  86                 return -ENETUNREACH;
  87         }
  88
  89         if (inet_opt == NULL || !inet_opt->opt.srr)
  90                 daddr = fl4->daddr;
  91
  92         if (inet->inet_saddr == 0) {
  93                 err = inet_bhash2_update_saddr(sk,  &fl4->saddr, AF_INET);
  94                 if (err) {
  95                         ip_rt_put(rt);
  96                         return err;
  97                 }
  98         } else {
  99                 sk_rcv_saddr_set(sk, inet->inet_saddr);
 100         }
 101
 102         inet->inet_dport = usin->sin_port;
 103         sk_daddr_set(sk, daddr);
 104
 105         inet_csk(sk)->icsk_ext_hdr_len = 0;
 106         if (inet_opt)
 107                 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
 108         /*
 109          * Socket identity is still unknown (sport may be zero).
 110          * However we set state to DCCP_REQUESTING and not releasing socket
 111          * lock select source port, enter ourselves into the hash tables and
 112          * complete initialization after this.
 113          */
 114         dccp_set_state(sk, DCCP_REQUESTING);
 115         err = inet_hash_connect(&dccp_death_row, sk);
 116         if (err != 0)
 117                 goto failure;
 118
 119         rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 120                                inet->inet_sport, inet->inet_dport, sk);
 121         if (IS_ERR(rt)) {
 122                 err = PTR_ERR(rt);
 123                 rt = NULL;
 124                 goto failure;
 125         }
 126         /* OK, now commit destination to socket.  */
 127         sk_setup_caps(sk, &rt->dst);
 128
 129         dp->dccps_iss = secure_dccp_sequence_number(inet->inet_saddr,
 130                                                     inet->inet_daddr,
 131                                                     inet->inet_sport,
 132                                                     inet->inet_dport);
 133         atomic_set(&inet->inet_id, get_random_u16());
 134
 135         err = dccp_connect(sk);
 136         rt = NULL;
 137         if (err != 0)
 138                 goto failure;
 139 out:
 140         return err;
 141 failure:
 142         /*
 143          * This unhashes the socket and releases the local port, if necessary.
 144          */
 145         dccp_set_state(sk, DCCP_CLOSED);
 146         inet_bhash2_reset_saddr(sk);
 147         ip_rt_put(rt);
 148         sk->sk_route_caps = 0;
 149         inet->inet_dport = 0;
 150         goto out;
 151 }
 152 EXPORT_SYMBOL_GPL(dccp_v4_connect);
 153
 154 /*
 155  * This routine does path mtu discovery as defined in RFC1191.
 156  */
 157 static inline void dccp_do_pmtu_discovery(struct sock *sk,
 158                                           const struct iphdr *iph,
 159                                           u32 mtu)
 160 {
 161         struct dst_entry *dst;
 162         const struct inet_sock *inet = inet_sk(sk);
 163         const struct dccp_sock *dp = dccp_sk(sk);
 164
 165         /* We are not interested in DCCP_LISTEN and request_socks (RESPONSEs
 166          * send out by Linux are always < 576bytes so they should go through
 167          * unfragmented).
 168          */
 169         if (sk->sk_state == DCCP_LISTEN)
 170                 return;
 171
 172         dst = inet_csk_update_pmtu(sk, mtu);
 173         if (!dst)
 174                 return;
 175
 176         /* Something is about to be wrong... Remember soft error
 177          * for the case, if this connection will not able to recover.
 178          */
 179         if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 180                 WRITE_ONCE(sk->sk_err_soft, EMSGSIZE);
 181
 182         mtu = dst_mtu(dst);
 183
 184         if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 185             ip_sk_accept_pmtu(sk) &&
 186             inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 187                 dccp_sync_mss(sk, mtu);
 188
 189                 /*
 190                  * From RFC 4340, sec. 14.1:
 191                  *
 192                  *      DCCP-Sync packets are the best choice for upward
 193                  *      probing, since DCCP-Sync probes do not risk application
 194                  *      data loss.
 195                  */
 196                 dccp_send_sync(sk, dp->dccps_gsr, DCCP_PKT_SYNC);
 197         } /* else let the usual retransmit timer handle it */
 198 }
 199
 200 static void dccp_do_redirect(struct sk_buff *skb, struct sock *sk)
 201 {
 202         struct dst_entry *dst = __sk_dst_check(sk, 0);
 203
 204         if (dst)
 205                 dst->ops->redirect(dst, sk, skb);
 206 }
 207
 208 void dccp_req_err(struct sock *sk, u64 seq)
 209         {
 210         struct request_sock *req = inet_reqsk(sk);
 211         struct net *net = sock_net(sk);
 212
 213         /*
 214          * ICMPs are not backlogged, hence we cannot get an established
 215          * socket here.
 216          */
 217         if (!between48(seq, dccp_rsk(req)->dreq_iss, dccp_rsk(req)->dreq_gss)) {
 218                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 219         } else {
 220                 /*
 221                  * Still in RESPOND, just remove it silently.
 222                  * There is no good way to pass the error to the newly
 223                  * created socket, and POSIX does not want network
 224                  * errors returned from accept().
 225                  */
 226                 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
 227         }
 228         reqsk_put(req);
 229 }
 230 EXPORT_SYMBOL(dccp_req_err);
 231
 232 /*
 233  * This routine is called by the ICMP module when it gets some sort of error
 234  * condition. If err < 0 then the socket should be closed and the error
 235  * returned to the user. If err > 0 it's just the icmp type << 8 | icmp code.
 236  * After adjustment header points to the first 8 bytes of the tcp header. We
 237  * need to find the appropriate port.
 238  *
 239  * The locking strategy used here is very "optimistic". When someone else
 240  * accesses the socket the ICMP is just dropped and for some paths there is no
 241  * check at all. A more general error queue to queue errors for later handling
 242  * is probably better.
 243  */
 244 static int dccp_v4_err(struct sk_buff *skb, u32 info)
 245 {
 246         const struct iphdr *iph = (struct iphdr *)skb->data;
 247         const u8 offset = iph->ihl << 2;
 248         const struct dccp_hdr *dh;
 249         struct dccp_sock *dp;
 250         struct inet_sock *inet;
 251         const int type = icmp_hdr(skb)->type;
 252         const int code = icmp_hdr(skb)->code;
 253         struct sock *sk;
 254         __u64 seq;
 255         int err;
 256         struct net *net = dev_net(skb->dev);
 257
 258         /* Only need dccph_dport & dccph_sport which are the first
 259          * 4 bytes in dccp header.
 260          * Our caller (icmp_socket_deliver()) already pulled 8 bytes for us.
 261          */
 262         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_sport) > 8);
 263         BUILD_BUG_ON(offsetofend(struct dccp_hdr, dccph_dport) > 8);
 264         dh = (struct dccp_hdr *)(skb->data + offset);
 265
 266         sk = __inet_lookup_established(net, &dccp_hashinfo,
 267                                        iph->daddr, dh->dccph_dport,
 268                                        iph->saddr, ntohs(dh->dccph_sport),
 269                                        inet_iif(skb), 0);
 270         if (!sk) {
 271                 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 272                 return -ENOENT;
 273         }
 274
 275         if (sk->sk_state == DCCP_TIME_WAIT) {
 276                 inet_twsk_put(inet_twsk(sk));
 277                 return 0;
 278         }
 279         seq = dccp_hdr_seq(dh);
 280         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 281                 dccp_req_err(sk, seq);
 282                 return 0;
 283         }
 284
 285         bh_lock_sock(sk);
 286         /* If too many ICMPs get dropped on busy
 287          * servers this needs to be solved differently.
 288          */
 289         if (sock_owned_by_user(sk))
 290                 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
 291
 292         if (sk->sk_state == DCCP_CLOSED)
 293                 goto out;
 294
 295         dp = dccp_sk(sk);
 296         if ((1 << sk->sk_state) & ~(DCCPF_REQUESTING | DCCPF_LISTEN) &&
 297             !between48(seq, dp->dccps_awl, dp->dccps_awh)) {
 298                 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 299                 goto out;
 300         }
 301
 302         switch (type) {
 303         case ICMP_REDIRECT:
 304                 if (!sock_owned_by_user(sk))
 305                         dccp_do_redirect(skb, sk);
 306                 goto out;
 307         case ICMP_SOURCE_QUENCH:
 308                 /* Just silently ignore these. */
 309                 goto out;
 310         case ICMP_PARAMETERPROB:
 311                 err = EPROTO;
 312                 break;
 313         case ICMP_DEST_UNREACH:
 314                 if (code > NR_ICMP_UNREACH)
 315                         goto out;
 316
 317                 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 318                         if (!sock_owned_by_user(sk))
 319                                 dccp_do_pmtu_discovery(sk, iph, info);
 320                         goto out;
 321                 }
 322
 323                 err = icmp_err_convert[code].errno;
 324                 break;
 325         case ICMP_TIME_EXCEEDED:
 326                 err = EHOSTUNREACH;
 327                 break;
 328         default:
 329                 goto out;
 330         }
 331
 332         switch (sk->sk_state) {
 333         case DCCP_REQUESTING:
 334         case DCCP_RESPOND:
 335                 if (!sock_owned_by_user(sk)) {
 336                         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 337                         sk->sk_err = err;
 338
 339                         sk_error_report(sk);
 340
 341                         dccp_done(sk);
 342                 } else {
 343                         WRITE_ONCE(sk->sk_err_soft, err);
 344                 }
 345                 goto out;
 346         }
 347
 348         /* If we've already connected we will keep trying
 349          * until we time out, or the user gives up.
 350          *
 351          * rfc1122 4.2.3.9 allows to consider as hard errors
 352          * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 353          * but it is obsoleted by pmtu discovery).
 354          *
 355          * Note, that in modern internet, where routing is unreliable
 356          * and in each dark corner broken firewalls sit, sending random
 357          * errors ordered by their masters even this two messages finally lose
 358          * their original sense (even Linux sends invalid PORT_UNREACHs)
 359          *
 360          * Now we are in compliance with RFCs.
 361          *                                                      --ANK (980905)
 362          */
 363
 364         inet = inet_sk(sk);
 365         if (!sock_owned_by_user(sk) && inet->recverr) {
 366                 sk->sk_err = err;
 367                 sk_error_report(sk);
 368         } else { /* Only an error on timeout */
 369                 WRITE_ONCE(sk->sk_err_soft, err);
 370         }
 371 out:
 372         bh_unlock_sock(sk);
 373         sock_put(sk);
 374         return 0;
 375 }
 376
 377 static inline __sum16 dccp_v4_csum_finish(struct sk_buff *skb,
 378                                       __be32 src, __be32 dst)
 379 {
 380         return csum_tcpudp_magic(src, dst, skb->len, IPPROTO_DCCP, skb->csum);
 381 }
 382
 383 void dccp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 384 {
 385         const struct inet_sock *inet = inet_sk(sk);
 386         struct dccp_hdr *dh = dccp_hdr(skb);
 387
 388         dccp_csum_outgoing(skb);
 389         dh->dccph_checksum = dccp_v4_csum_finish(skb,
 390                                                  inet->inet_saddr,
 391                                                  inet->inet_daddr);
 392 }
 393 EXPORT_SYMBOL_GPL(dccp_v4_send_check);
 394
 395 static inline u64 dccp_v4_init_sequence(const struct sk_buff *skb)
 396 {
 397         return secure_dccp_sequence_number(ip_hdr(skb)->daddr,
 398                                            ip_hdr(skb)->saddr,
 399                                            dccp_hdr(skb)->dccph_dport,
 400                                            dccp_hdr(skb)->dccph_sport);
 401 }
 402
 403 /*
 404  * The three way handshake has completed - we got a valid ACK or DATAACK -
 405  * now create the new socket.
 406  *
 407  * This is the equivalent of TCP's tcp_v4_syn_recv_sock
 408  */
 409 struct sock *dccp_v4_request_recv_sock(const struct sock *sk,
 410                                        struct sk_buff *skb,
 411                                        struct request_sock *req,
 412                                        struct dst_entry *dst,
 413                                        struct request_sock *req_unhash,
 414                                        bool *own_req)
 415 {
 416         struct inet_request_sock *ireq;
 417         struct inet_sock *newinet;
 418         struct sock *newsk;
 419
 420         if (sk_acceptq_is_full(sk))
 421                 goto exit_overflow;
 422
 423         newsk = dccp_create_openreq_child(sk, req, skb);
 424         if (newsk == NULL)
 425                 goto exit_nonewsk;
 426
 427         newinet            = inet_sk(newsk);
 428         ireq               = inet_rsk(req);
 429         sk_daddr_set(newsk, ireq->ir_rmt_addr);
 430         sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
 431         newinet->inet_saddr     = ireq->ir_loc_addr;
 432         RCU_INIT_POINTER(newinet->inet_opt, rcu_dereference(ireq->ireq_opt));
 433         newinet->mc_index  = inet_iif(skb);
 434         newinet->mc_ttl    = ip_hdr(skb)->ttl;
 435         atomic_set(&newinet->inet_id, get_random_u16());
 436
 437         if (dst == NULL && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
 438                 goto put_and_exit;
 439
 440         sk_setup_caps(newsk, dst);
 441
 442         dccp_sync_mss(newsk, dst_mtu(dst));
 443
 444         if (__inet_inherit_port(sk, newsk) < 0)
 445                 goto put_and_exit;
 446         *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash), NULL);
 447         if (*own_req)
 448                 ireq->ireq_opt = NULL;
 449         else
 450                 newinet->inet_opt = NULL;
 451         return newsk;
 452
 453 exit_overflow:
 454         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 455 exit_nonewsk:
 456         dst_release(dst);
 457 exit:
 458         __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
 459         return NULL;
 460 put_and_exit:
 461         newinet->inet_opt = NULL;
 462         inet_csk_prepare_forced_close(newsk);
 463         dccp_done(newsk);
 464         goto exit;
 465 }
 466 EXPORT_SYMBOL_GPL(dccp_v4_request_recv_sock);
 467
 468 static struct dst_entry* dccp_v4_route_skb(struct net *net, struct sock *sk,
 469                                            struct sk_buff *skb)
 470 {
 471         struct rtable *rt;
 472         const struct iphdr *iph = ip_hdr(skb);
 473         struct flowi4 fl4 = {
 474                 .flowi4_oif = inet_iif(skb),
 475                 .daddr = iph->saddr,
 476                 .saddr = iph->daddr,
 477                 .flowi4_tos = RT_CONN_FLAGS(sk),
 478                 .flowi4_proto = sk->sk_protocol,
 479                 .fl4_sport = dccp_hdr(skb)->dccph_dport,
 480                 .fl4_dport = dccp_hdr(skb)->dccph_sport,
 481         };
 482
 483         security_skb_classify_flow(skb, flowi4_to_flowi_common(&fl4));
 484         rt = ip_route_output_flow(net, &fl4, sk);
 485         if (IS_ERR(rt)) {
 486                 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 487                 return NULL;
 488         }
 489
 490         return &rt->dst;
 491 }
 492
 493 static int dccp_v4_send_response(const struct sock *sk, struct request_sock *req)
 494 {
 495         int err = -1;
 496         struct sk_buff *skb;
 497         struct dst_entry *dst;
 498         struct flowi4 fl4;
 499
 500         dst = inet_csk_route_req(sk, &fl4, req);
 501         if (dst == NULL)
 502                 goto out;
 503
 504         skb = dccp_make_response(sk, dst, req);
 505         if (skb != NULL) {
 506                 const struct inet_request_sock *ireq = inet_rsk(req);
 507                 struct dccp_hdr *dh = dccp_hdr(skb);
 508
 509                 dh->dccph_checksum = dccp_v4_csum_finish(skb, ireq->ir_loc_addr,
 510                                                               ireq->ir_rmt_addr);
 511                 rcu_read_lock();
 512                 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 513                                             ireq->ir_rmt_addr,
 514                                             rcu_dereference(ireq->ireq_opt),
 515                                             inet_sk(sk)->tos);
 516                 rcu_read_unlock();
 517                 err = net_xmit_eval(err);
 518         }
 519
 520 out:
 521         dst_release(dst);
 522         return err;
 523 }
 524
 525 static void dccp_v4_ctl_send_reset(const struct sock *sk, struct sk_buff *rxskb)
 526 {
 527         int err;
 528         const struct iphdr *rxiph;
 529         struct sk_buff *skb;
 530         struct dst_entry *dst;
 531         struct net *net = dev_net(skb_dst(rxskb)->dev);
 532         struct dccp_v4_pernet *pn;
 533         struct sock *ctl_sk;
 534
 535         /* Never send a reset in response to a reset. */
 536         if (dccp_hdr(rxskb)->dccph_type == DCCP_PKT_RESET)
 537                 return;
 538
 539         if (skb_rtable(rxskb)->rt_type != RTN_LOCAL)
 540                 return;
 541
 542         pn = net_generic(net, dccp_v4_pernet_id);
 543         ctl_sk = pn->v4_ctl_sk;
 544         dst = dccp_v4_route_skb(net, ctl_sk, rxskb);
 545         if (dst == NULL)
 546                 return;
 547
 548         skb = dccp_ctl_make_reset(ctl_sk, rxskb);
 549         if (skb == NULL)
 550                 goto out;
 551
 552         rxiph = ip_hdr(rxskb);
 553         dccp_hdr(skb)->dccph_checksum = dccp_v4_csum_finish(skb, rxiph->saddr,
 554                                                                  rxiph->daddr);
 555         skb_dst_set(skb, dst_clone(dst));
 556
 557         local_bh_disable();
 558         bh_lock_sock(ctl_sk);
 559         err = ip_build_and_send_pkt(skb, ctl_sk,
 560                                     rxiph->daddr, rxiph->saddr, NULL,
 561                                     inet_sk(ctl_sk)->tos);
 562         bh_unlock_sock(ctl_sk);
 563
 564         if (net_xmit_eval(err) == 0) {
 565                 __DCCP_INC_STATS(DCCP_MIB_OUTSEGS);
 566                 __DCCP_INC_STATS(DCCP_MIB_OUTRSTS);
 567         }
 568         local_bh_enable();
 569 out:
 570         dst_release(dst);
 571 }
 572
 573 static void dccp_v4_reqsk_destructor(struct request_sock *req)
 574 {
 575         dccp_feat_list_purge(&dccp_rsk(req)->dreq_featneg);
 576         kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
 577 }
 578
 579 void dccp_syn_ack_timeout(const struct request_sock *req)
 580 {
 581 }
 582 EXPORT_SYMBOL(dccp_syn_ack_timeout);
 583
 584 static struct request_sock_ops dccp_request_sock_ops __read_mostly = {
 585         .family         = PF_INET,
 586         .obj_size       = sizeof(struct dccp_request_sock),
 587         .rtx_syn_ack    = dccp_v4_send_response,
 588         .send_ack       = dccp_reqsk_send_ack,
 589         .destructor     = dccp_v4_reqsk_destructor,
 590         .send_reset     = dccp_v4_ctl_send_reset,
 591         .syn_ack_timeout = dccp_syn_ack_timeout,
 592 };
 593
 594 int dccp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
 595 {
 596         struct inet_request_sock *ireq;
 597         struct request_sock *req;
 598         struct dccp_request_sock *dreq;
 599         const __be32 service = dccp_hdr_request(skb)->dccph_req_service;
 600         struct dccp_skb_cb *dcb = DCCP_SKB_CB(skb);
 601
 602         /* Never answer to DCCP_PKT_REQUESTs send to broadcast or multicast */
 603         if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
 604                 return 0;       /* discard, don't send a reset here */
 605
 606         if (dccp_bad_service_code(sk, service)) {
 607                 dcb->dccpd_reset_code = DCCP_RESET_CODE_BAD_SERVICE_CODE;
 608                 goto drop;
 609         }
 610         /*
 611          * TW buckets are converted to open requests without
 612          * limitations, they conserve resources and peer is
 613          * evidently real one.
 614          */
 615         dcb->dccpd_reset_code = DCCP_RESET_CODE_TOO_BUSY;
 616         if (inet_csk_reqsk_queue_is_full(sk))
 617                 goto drop;
 618
 619         if (sk_acceptq_is_full(sk))
 620                 goto drop;
 621
 622         req = inet_reqsk_alloc(&dccp_request_sock_ops, sk, true);
 623         if (req == NULL)
 624                 goto drop;
 625
 626         if (dccp_reqsk_init(req, dccp_sk(sk), skb))
 627                 goto drop_and_free;
 628
 629         dreq = dccp_rsk(req);
 630         if (dccp_parse_options(sk, dreq, skb))
 631                 goto drop_and_free;
 632
 633         if (security_inet_conn_request(sk, skb, req))
 634                 goto drop_and_free;
 635
 636         ireq = inet_rsk(req);
 637         sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
 638         sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
 639         ireq->ir_mark = inet_request_mark(sk, skb);
 640         ireq->ireq_family = AF_INET;
 641         ireq->ir_iif = READ_ONCE(sk->sk_bound_dev_if);
 642
 643         /*
 644          * Step 3: Process LISTEN state
 645          *
 646          * Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookie
 647          *
 648          * Setting S.SWL/S.SWH to is deferred to dccp_create_openreq_child().
 649          */
 650         dreq->dreq_isr     = dcb->dccpd_seq;
 651         dreq->dreq_gsr     = dreq->dreq_isr;
 652         dreq->dreq_iss     = dccp_v4_init_sequence(skb);
 653         dreq->dreq_gss     = dreq->dreq_iss;
 654         dreq->dreq_service = service;
 655
 656         if (dccp_v4_send_response(sk, req))
 657                 goto drop_and_free;
 658
 659         inet_csk_reqsk_queue_hash_add(sk, req, DCCP_TIMEOUT_INIT);
 660         reqsk_put(req);
 661         return 0;
 662
 663 drop_and_free:
 664         reqsk_free(req);
 665 drop:
 666         __DCCP_INC_STATS(DCCP_MIB_ATTEMPTFAILS);
 667         return -1;
 668 }
 669 EXPORT_SYMBOL_GPL(dccp_v4_conn_request);
 670
 671 int dccp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 672 {
 673         struct dccp_hdr *dh = dccp_hdr(skb);
 674
 675         if (sk->sk_state == DCCP_OPEN) { /* Fast path */
 676                 if (dccp_rcv_established(sk, skb, dh, skb->len))
 677                         goto reset;
 678                 return 0;
 679         }
 680
 681         /*
 682          *  Step 3: Process LISTEN state
 683          *       If P.type == Request or P contains a valid Init Cookie option,
 684          *            (* Must scan the packet's options to check for Init
 685          *               Cookies.  Only Init Cookies are processed here,
 686          *               however; other options are processed in Step 8.  This
 687          *               scan need only be performed if the endpoint uses Init
 688          *               Cookies *)
 689          *            (* Generate a new socket and switch to that socket *)
 690          *            Set S := new socket for this port pair
 691          *            S.state = RESPOND
 692          *            Choose S.ISS (initial seqno) or set from Init Cookies
 693          *            Initialize S.GAR := S.ISS
 694          *            Set S.ISR, S.GSR, S.SWL, S.SWH from packet or Init Cookies
 695          *            Continue with S.state == RESPOND
 696          *            (* A Response packet will be generated in Step 11 *)
 697          *       Otherwise,
 698          *            Generate Reset(No Connection) unless P.type == Reset
 699          *            Drop packet and return
 700          *
 701          * NOTE: the check for the packet types is done in
 702          *       dccp_rcv_state_process
 703          */
 704
 705         if (dccp_rcv_state_process(sk, skb, dh, skb->len))
 706                 goto reset;
 707         return 0;
 708
 709 reset:
 710         dccp_v4_ctl_send_reset(sk, skb);
 711         kfree_skb(skb);
 712         return 0;
 713 }
 714 EXPORT_SYMBOL_GPL(dccp_v4_do_rcv);
 715
 716 /**
 717  *      dccp_invalid_packet  -  check for malformed packets
 718  *      @skb: Packet to validate
 719  *
 720  *      Implements RFC 4340, 8.5:  Step 1: Check header basics
 721  *      Packets that fail these checks are ignored and do not receive Resets.
 722  */
 723 int dccp_invalid_packet(struct sk_buff *skb)
 724 {
 725         const struct dccp_hdr *dh;
 726         unsigned int cscov;
 727         u8 dccph_doff;
 728
 729         if (skb->pkt_type != PACKET_HOST)
 730                 return 1;
 731
 732         /* If the packet is shorter than 12 bytes, drop packet and return */
 733         if (!pskb_may_pull(skb, sizeof(struct dccp_hdr))) {
 734                 DCCP_WARN("pskb_may_pull failed\n");
 735                 return 1;
 736         }
 737
 738         dh = dccp_hdr(skb);
 739
 740         /* If P.type is not understood, drop packet and return */
 741         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 742                 DCCP_WARN("invalid packet type\n");
 743                 return 1;
 744         }
 745
 746         /*
 747          * If P.Data Offset is too small for packet type, drop packet and return
 748          */
 749         dccph_doff = dh->dccph_doff;
 750         if (dccph_doff < dccp_hdr_len(skb) / sizeof(u32)) {
 751                 DCCP_WARN("P.Data Offset(%u) too small\n", dccph_doff);
 752                 return 1;
 753         }
 754         /*
 755          * If P.Data Offset is too large for packet, drop packet and return
 756          */
 757         if (!pskb_may_pull(skb, dccph_doff * sizeof(u32))) {
 758                 DCCP_WARN("P.Data Offset(%u) too large\n", dccph_doff);
 759                 return 1;
 760         }
 761         dh = dccp_hdr(skb);
 762         /*
 763          * If P.type is not Data, Ack, or DataAck and P.X == 0 (the packet
 764          * has short sequence numbers), drop packet and return
 765          */
 766         if ((dh->dccph_type < DCCP_PKT_DATA    ||
 767             dh->dccph_type > DCCP_PKT_DATAACK) && dh->dccph_x == 0)  {
 768                 DCCP_WARN("P.type (%s) not Data || [Data]Ack, while P.X == 0\n",
 769                           dccp_packet_name(dh->dccph_type));
 770                 return 1;
 771         }
 772
 773         /*
 774          * If P.CsCov is too large for the packet size, drop packet and return.
 775          * This must come _before_ checksumming (not as RFC 4340 suggests).
 776          */
 777         cscov = dccp_csum_coverage(skb);
 778         if (cscov > skb->len) {
 779                 DCCP_WARN("P.CsCov %u exceeds packet length %d\n",
 780                           dh->dccph_cscov, skb->len);
 781                 return 1;
 782         }
 783
 784         /* If header checksum is incorrect, drop packet and return.
 785          * (This step is completed in the AF-dependent functions.) */
 786         skb->csum = skb_checksum(skb, 0, cscov, 0);
 787
 788         return 0;
 789 }
 790 EXPORT_SYMBOL_GPL(dccp_invalid_packet);
 791
 792 /* this is called when real data arrives */
 793 static int dccp_v4_rcv(struct sk_buff *skb)
 794 {
 795         const struct dccp_hdr *dh;
 796         const struct iphdr *iph;
 797         bool refcounted;
 798         struct sock *sk;
 799         int min_cov;
 800
 801         /* Step 1: Check header basics */
 802
 803         if (dccp_invalid_packet(skb))
 804                 goto discard_it;
 805
 806         iph = ip_hdr(skb);
 807         /* Step 1: If header checksum is incorrect, drop packet and return */
 808         if (dccp_v4_csum_finish(skb, iph->saddr, iph->daddr)) {
 809                 DCCP_WARN("dropped packet with invalid checksum\n");
 810                 goto discard_it;
 811         }
 812
 813         dh = dccp_hdr(skb);
 814
 815         DCCP_SKB_CB(skb)->dccpd_seq  = dccp_hdr_seq(dh);
 816         DCCP_SKB_CB(skb)->dccpd_type = dh->dccph_type;
 817
 818         dccp_pr_debug("%8.8s src=%pI4@%-5d dst=%pI4@%-5d seq=%llu",
 819                       dccp_packet_name(dh->dccph_type),
 820                       &iph->saddr, ntohs(dh->dccph_sport),
 821                       &iph->daddr, ntohs(dh->dccph_dport),
 822                       (unsigned long long) DCCP_SKB_CB(skb)->dccpd_seq);
 823
 824         if (dccp_packet_without_ack(skb)) {
 825                 DCCP_SKB_CB(skb)->dccpd_ack_seq = DCCP_PKT_WITHOUT_ACK_SEQ;
 826                 dccp_pr_debug_cat("\n");
 827         } else {
 828                 DCCP_SKB_CB(skb)->dccpd_ack_seq = dccp_hdr_ack_seq(skb);
 829                 dccp_pr_debug_cat(", ack=%llu\n", (unsigned long long)
 830                                   DCCP_SKB_CB(skb)->dccpd_ack_seq);
 831         }
 832
 833 lookup:
 834         sk = __inet_lookup_skb(&dccp_hashinfo, skb, __dccp_hdr_len(dh),
 835                                dh->dccph_sport, dh->dccph_dport, 0, &refcounted);
 836         if (!sk) {
 837                 dccp_pr_debug("failed to look up flow ID in table and "
 838                               "get corresponding socket\n");
 839                 goto no_dccp_socket;
 840         }
 841
 842         /*
 843          * Step 2:
 844          *      ... or S.state == TIMEWAIT,
 845          *              Generate Reset(No Connection) unless P.type == Reset
 846          *              Drop packet and return
 847          */
 848         if (sk->sk_state == DCCP_TIME_WAIT) {
 849                 dccp_pr_debug("sk->sk_state == DCCP_TIME_WAIT: do_time_wait\n");
 850                 inet_twsk_put(inet_twsk(sk));
 851                 goto no_dccp_socket;
 852         }
 853
 854         if (sk->sk_state == DCCP_NEW_SYN_RECV) {
 855                 struct request_sock *req = inet_reqsk(sk);
 856                 struct sock *nsk;
 857
 858                 sk = req->rsk_listener;
 859                 if (unlikely(sk->sk_state != DCCP_LISTEN)) {
 860                         inet_csk_reqsk_queue_drop_and_put(sk, req);
 861                         goto lookup;
 862                 }
 863                 sock_hold(sk);
 864                 refcounted = true;
 865                 nsk = dccp_check_req(sk, skb, req);
 866                 if (!nsk) {
 867                         reqsk_put(req);
 868                         goto discard_and_relse;
 869                 }
 870                 if (nsk == sk) {
 871                         reqsk_put(req);
 872                 } else if (dccp_child_process(sk, nsk, skb)) {
 873                         dccp_v4_ctl_send_reset(sk, skb);
 874                         goto discard_and_relse;
 875                 } else {
 876                         sock_put(sk);
 877                         return 0;
 878                 }
 879         }
 880         /*
 881          * RFC 4340, sec. 9.2.1: Minimum Checksum Coverage
 882          *      o if MinCsCov = 0, only packets with CsCov = 0 are accepted
 883          *      o if MinCsCov > 0, also accept packets with CsCov >= MinCsCov
 884          */
 885         min_cov = dccp_sk(sk)->dccps_pcrlen;
 886         if (dh->dccph_cscov && (min_cov == 0 || dh->dccph_cscov < min_cov))  {
 887                 dccp_pr_debug("Packet CsCov %d does not satisfy MinCsCov %d\n",
 888                               dh->dccph_cscov, min_cov);
 889                 /* FIXME: "Such packets SHOULD be reported using Data Dropped
 890                  *         options (Section 11.7) with Drop Code 0, Protocol
 891                  *         Constraints."                                     */
 892                 goto discard_and_relse;
 893         }
 894
 895         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
 896                 goto discard_and_relse;
 897         nf_reset_ct(skb);
 898
 899         return __sk_receive_skb(sk, skb, 1, dh->dccph_doff * 4, refcounted);
 900
 901 no_dccp_socket:
 902         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
 903                 goto discard_it;
 904         /*
 905          * Step 2:
 906          *      If no socket ...
 907          *              Generate Reset(No Connection) unless P.type == Reset
 908          *              Drop packet and return
 909          */
 910         if (dh->dccph_type != DCCP_PKT_RESET) {
 911                 DCCP_SKB_CB(skb)->dccpd_reset_code =
 912                                         DCCP_RESET_CODE_NO_CONNECTION;
 913                 dccp_v4_ctl_send_reset(sk, skb);
 914         }
 915
 916 discard_it:
 917         kfree_skb(skb);
 918         return 0;
 919
 920 discard_and_relse:
 921         if (refcounted)
 922                 sock_put(sk);
 923         goto discard_it;
 924 }
 925
 926 static const struct inet_connection_sock_af_ops dccp_ipv4_af_ops = {
 927         .queue_xmit        = ip_queue_xmit,
 928         .send_check        = dccp_v4_send_check,
 929         .rebuild_header    = inet_sk_rebuild_header,
 930         .conn_request      = dccp_v4_conn_request,
 931         .syn_recv_sock     = dccp_v4_request_recv_sock,
 932         .net_header_len    = sizeof(struct iphdr),
 933         .setsockopt        = ip_setsockopt,
 934         .getsockopt        = ip_getsockopt,
 935         .addr2sockaddr     = inet_csk_addr2sockaddr,
 936         .sockaddr_len      = sizeof(struct sockaddr_in),
 937 };
 938
 939 static int dccp_v4_init_sock(struct sock *sk)
 940 {
 941         static __u8 dccp_v4_ctl_sock_initialized;
 942         int err = dccp_init_sock(sk, dccp_v4_ctl_sock_initialized);
 943
 944         if (err == 0) {
 945                 if (unlikely(!dccp_v4_ctl_sock_initialized))
 946                         dccp_v4_ctl_sock_initialized = 1;
 947                 inet_csk(sk)->icsk_af_ops = &dccp_ipv4_af_ops;
 948         }
 949
 950         return err;
 951 }
 952
 953 static struct timewait_sock_ops dccp_timewait_sock_ops = {
 954         .twsk_obj_size  = sizeof(struct inet_timewait_sock),
 955 };
 956
 957 static struct proto dccp_v4_prot = {
 958         .name                   = "DCCP",
 959         .owner                  = THIS_MODULE,
 960         .close                  = dccp_close,
 961         .connect                = dccp_v4_connect,
 962         .disconnect             = dccp_disconnect,
 963         .ioctl                  = dccp_ioctl,
 964         .init                   = dccp_v4_init_sock,
 965         .setsockopt             = dccp_setsockopt,
 966         .getsockopt             = dccp_getsockopt,
 967         .sendmsg                = dccp_sendmsg,
 968         .recvmsg                = dccp_recvmsg,
 969         .backlog_rcv            = dccp_v4_do_rcv,
 970         .hash                   = inet_hash,
 971         .unhash                 = inet_unhash,
 972         .accept                 = inet_csk_accept,
 973         .get_port               = inet_csk_get_port,
 974         .shutdown               = dccp_shutdown,
 975         .destroy                = dccp_destroy_sock,
 976         .orphan_count           = &dccp_orphan_count,
 977         .max_header             = MAX_DCCP_HEADER,
 978         .obj_size               = sizeof(struct dccp_sock),
 979         .slab_flags             = SLAB_TYPESAFE_BY_RCU,
 980         .rsk_prot               = &dccp_request_sock_ops,
 981         .twsk_prot              = &dccp_timewait_sock_ops,
 982         .h.hashinfo             = &dccp_hashinfo,
 983 };
 984
 985 static const struct net_protocol dccp_v4_protocol = {
 986         .handler        = dccp_v4_rcv,
 987         .err_handler    = dccp_v4_err,
 988         .no_policy      = 1,
 989         .icmp_strict_tag_validation = 1,
 990 };
 991
 992 static const struct proto_ops inet_dccp_ops = {
 993         .family            = PF_INET,
 994         .owner             = THIS_MODULE,
 995         .release           = inet_release,
 996         .bind              = inet_bind,
 997         .connect           = inet_stream_connect,
 998         .socketpair        = sock_no_socketpair,
 999         .accept            = inet_accept,
1000         .getname           = inet_getname,
1001         /* FIXME: work on tcp_poll to rename it to inet_csk_poll */
1002         .poll              = dccp_poll,
1003         .ioctl             = inet_ioctl,
1004         .gettstamp         = sock_gettstamp,
1005         /* FIXME: work on inet_listen to rename it to sock_common_listen */
1006         .listen            = inet_dccp_listen,
1007         .shutdown          = inet_shutdown,
1008         .setsockopt        = sock_common_setsockopt,
1009         .getsockopt        = sock_common_getsockopt,
1010         .sendmsg           = inet_sendmsg,
1011         .recvmsg           = sock_common_recvmsg,
1012         .mmap              = sock_no_mmap,
1013 };
1014
1015 static struct inet_protosw dccp_v4_protosw = {
1016         .type           = SOCK_DCCP,
1017         .protocol       = IPPROTO_DCCP,
1018         .prot           = &dccp_v4_prot,
1019         .ops            = &inet_dccp_ops,
1020         .flags          = INET_PROTOSW_ICSK,
1021 };
1022
1023 static int __net_init dccp_v4_init_net(struct net *net)
1024 {
1025         struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1026
1027         if (dccp_hashinfo.bhash == NULL)
1028                 return -ESOCKTNOSUPPORT;
1029
1030         return inet_ctl_sock_create(&pn->v4_ctl_sk, PF_INET,
1031                                     SOCK_DCCP, IPPROTO_DCCP, net);
1032 }
1033
1034 static void __net_exit dccp_v4_exit_net(struct net *net)
1035 {
1036         struct dccp_v4_pernet *pn = net_generic(net, dccp_v4_pernet_id);
1037
1038         inet_ctl_sock_destroy(pn->v4_ctl_sk);
1039 }
1040
1041 static void __net_exit dccp_v4_exit_batch(struct list_head *net_exit_list)
1042 {
1043         inet_twsk_purge(&dccp_hashinfo, AF_INET);
1044 }
1045
1046 static struct pernet_operations dccp_v4_ops = {
1047         .init   = dccp_v4_init_net,
1048         .exit   = dccp_v4_exit_net,
1049         .exit_batch = dccp_v4_exit_batch,
1050         .id     = &dccp_v4_pernet_id,
1051         .size   = sizeof(struct dccp_v4_pernet),
1052 };
1053
1054 static int __init dccp_v4_init(void)
1055 {
1056         int err = proto_register(&dccp_v4_prot, 1);
1057
1058         if (err)
1059                 goto out;
1060
1061         inet_register_protosw(&dccp_v4_protosw);
1062
1063         err = register_pernet_subsys(&dccp_v4_ops);
1064         if (err)
1065                 goto out_destroy_ctl_sock;
1066
1067         err = inet_add_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1068         if (err)
1069                 goto out_proto_unregister;
1070
1071 out:
1072         return err;
1073 out_proto_unregister:
1074         unregister_pernet_subsys(&dccp_v4_ops);
1075 out_destroy_ctl_sock:
1076         inet_unregister_protosw(&dccp_v4_protosw);
1077         proto_unregister(&dccp_v4_prot);
1078         goto out;
1079 }
1080
1081 static void __exit dccp_v4_exit(void)
1082 {
1083         inet_del_protocol(&dccp_v4_protocol, IPPROTO_DCCP);
1084         unregister_pernet_subsys(&dccp_v4_ops);
1085         inet_unregister_protosw(&dccp_v4_protosw);
1086         proto_unregister(&dccp_v4_prot);
1087 }
1088
1089 module_init(dccp_v4_init);
1090 module_exit(dccp_v4_exit);
1091
1092 /*
1093  * __stringify doesn't likes enums, so use SOCK_DCCP (6) and IPPROTO_DCCP (33)
1094  * values directly, Also cover the case where the protocol is not specified,
1095  * i.e. net-pf-PF_INET-proto-0-type-SOCK_DCCP
1096  */
1097 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 33, 6);
1098 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 0, 6);
1099 MODULE_LICENSE("GPL");
1100 MODULE_AUTHOR("Arnaldo Carvalho de Melo <acme@mandriva.com>");
1101 MODULE_DESCRIPTION("DCCP - Datagram Congestion Controlled Protocol");