net/netfilter/nf_conntrack_proto_dccp.c

   1 /*
   2  * DCCP connection tracking protocol helper
   3  *
   4  * Copyright (c) 2005, 2006, 2008 Patrick McHardy <kaber@trash.net>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  *
  10  */
  11 #include <linux/kernel.h>
  12 #include <linux/module.h>
  13 #include <linux/init.h>
  14 #include <linux/sysctl.h>
  15 #include <linux/spinlock.h>
  16 #include <linux/skbuff.h>
  17 #include <linux/dccp.h>
  18
  19 #include <net/net_namespace.h>
  20 #include <net/netns/generic.h>
  21
  22 #include <linux/netfilter/nfnetlink_conntrack.h>
  23 #include <net/netfilter/nf_conntrack.h>
  24 #include <net/netfilter/nf_conntrack_l4proto.h>
  25 #include <net/netfilter/nf_conntrack_ecache.h>
  26 #include <net/netfilter/nf_log.h>
  27
  28 /* Timeouts are based on values from RFC4340:
  29  *
  30  * - REQUEST:
  31  *
  32  *   8.1.2. Client Request
  33  *
  34  *   A client MAY give up on its DCCP-Requests after some time
  35  *   (3 minutes, for example).
  36  *
  37  * - RESPOND:
  38  *
  39  *   8.1.3. Server Response
  40  *
  41  *   It MAY also leave the RESPOND state for CLOSED after a timeout of
  42  *   not less than 4MSL (8 minutes);
  43  *
  44  * - PARTOPEN:
  45  *
  46  *   8.1.5. Handshake Completion
  47  *
  48  *   If the client remains in PARTOPEN for more than 4MSL (8 minutes),
  49  *   it SHOULD reset the connection with Reset Code 2, "Aborted".
  50  *
  51  * - OPEN:
  52  *
  53  *   The DCCP timestamp overflows after 11.9 hours. If the connection
  54  *   stays idle this long the sequence number won't be recognized
  55  *   as valid anymore.
  56  *
  57  * - CLOSEREQ/CLOSING:
  58  *
  59  *   8.3. Termination
  60  *
  61  *   The retransmission timer should initially be set to go off in two
  62  *   round-trip times and should back off to not less than once every
  63  *   64 seconds ...
  64  *
  65  * - TIMEWAIT:
  66  *
  67  *   4.3. States
  68  *
  69  *   A server or client socket remains in this state for 2MSL (4 minutes)
  70  *   after the connection has been town down, ...
  71  */
  72
  73 #define DCCP_MSL (2 * 60 * HZ)
  74
  75 static const char * const dccp_state_names[] = {
  76         [CT_DCCP_NONE]          = "NONE",
  77         [CT_DCCP_REQUEST]       = "REQUEST",
  78         [CT_DCCP_RESPOND]       = "RESPOND",
  79         [CT_DCCP_PARTOPEN]      = "PARTOPEN",
  80         [CT_DCCP_OPEN]          = "OPEN",
  81         [CT_DCCP_CLOSEREQ]      = "CLOSEREQ",
  82         [CT_DCCP_CLOSING]       = "CLOSING",
  83         [CT_DCCP_TIMEWAIT]      = "TIMEWAIT",
  84         [CT_DCCP_IGNORE]        = "IGNORE",
  85         [CT_DCCP_INVALID]       = "INVALID",
  86 };
  87
  88 #define sNO     CT_DCCP_NONE
  89 #define sRQ     CT_DCCP_REQUEST
  90 #define sRS     CT_DCCP_RESPOND
  91 #define sPO     CT_DCCP_PARTOPEN
  92 #define sOP     CT_DCCP_OPEN
  93 #define sCR     CT_DCCP_CLOSEREQ
  94 #define sCG     CT_DCCP_CLOSING
  95 #define sTW     CT_DCCP_TIMEWAIT
  96 #define sIG     CT_DCCP_IGNORE
  97 #define sIV     CT_DCCP_INVALID
  98
  99 /*
 100  * DCCP state transistion table
 101  *
 102  * The assumption is the same as for TCP tracking:
 103  *
 104  * We are the man in the middle. All the packets go through us but might
 105  * get lost in transit to the destination. It is assumed that the destination
 106  * can't receive segments we haven't seen.
 107  *
 108  * The following states exist:
 109  *
 110  * NONE:        Initial state, expecting Request
 111  * REQUEST:     Request seen, waiting for Response from server
 112  * RESPOND:     Response from server seen, waiting for Ack from client
 113  * PARTOPEN:    Ack after Response seen, waiting for packet other than Response,
 114  *              Reset or Sync from server
 115  * OPEN:        Packet other than Response, Reset or Sync seen
 116  * CLOSEREQ:    CloseReq from server seen, expecting Close from client
 117  * CLOSING:     Close seen, expecting Reset
 118  * TIMEWAIT:    Reset seen
 119  * IGNORE:      Not determinable whether packet is valid
 120  *
 121  * Some states exist only on one side of the connection: REQUEST, RESPOND,
 122  * PARTOPEN, CLOSEREQ. For the other side these states are equivalent to
 123  * the one it was in before.
 124  *
 125  * Packets are marked as ignored (sIG) if we don't know if they're valid
 126  * (for example a reincarnation of a connection we didn't notice is dead
 127  * already) and the server may send back a connection closing Reset or a
 128  * Response. They're also used for Sync/SyncAck packets, which we don't
 129  * care about.
 130  */
 131 static const u_int8_t
 132 dccp_state_table[CT_DCCP_ROLE_MAX + 1][DCCP_PKT_SYNCACK + 1][CT_DCCP_MAX + 1] = {
 133         [CT_DCCP_ROLE_CLIENT] = {
 134                 [DCCP_PKT_REQUEST] = {
 135                 /*
 136                  * sNO -> sRQ           Regular Request
 137                  * sRQ -> sRQ           Retransmitted Request or reincarnation
 138                  * sRS -> sRS           Retransmitted Request (apparently Response
 139                  *                      got lost after we saw it) or reincarnation
 140                  * sPO -> sIG           Ignore, conntrack might be out of sync
 141                  * sOP -> sIG           Ignore, conntrack might be out of sync
 142                  * sCR -> sIG           Ignore, conntrack might be out of sync
 143                  * sCG -> sIG           Ignore, conntrack might be out of sync
 144                  * sTW -> sRQ           Reincarnation
 145                  *
 146                  *      sNO, sRQ, sRS, sPO. sOP, sCR, sCG, sTW, */
 147                         sRQ, sRQ, sRS, sIG, sIG, sIG, sIG, sRQ,
 148                 },
 149                 [DCCP_PKT_RESPONSE] = {
 150                 /*
 151                  * sNO -> sIV           Invalid
 152                  * sRQ -> sIG           Ignore, might be response to ignored Request
 153                  * sRS -> sIG           Ignore, might be response to ignored Request
 154                  * sPO -> sIG           Ignore, might be response to ignored Request
 155                  * sOP -> sIG           Ignore, might be response to ignored Request
 156                  * sCR -> sIG           Ignore, might be response to ignored Request
 157                  * sCG -> sIG           Ignore, might be response to ignored Request
 158                  * sTW -> sIV           Invalid, reincarnation in reverse direction
 159                  *                      goes through sRQ
 160                  *
 161                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 162                         sIV, sIG, sIG, sIG, sIG, sIG, sIG, sIV,
 163                 },
 164                 [DCCP_PKT_ACK] = {
 165                 /*
 166                  * sNO -> sIV           No connection
 167                  * sRQ -> sIV           No connection
 168                  * sRS -> sPO           Ack for Response, move to PARTOPEN (8.1.5.)
 169                  * sPO -> sPO           Retransmitted Ack for Response, remain in PARTOPEN
 170                  * sOP -> sOP           Regular ACK, remain in OPEN
 171                  * sCR -> sCR           Ack in CLOSEREQ MAY be processed (8.3.)
 172                  * sCG -> sCG           Ack in CLOSING MAY be processed (8.3.)
 173                  * sTW -> sIV
 174                  *
 175                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 176                         sIV, sIV, sPO, sPO, sOP, sCR, sCG, sIV
 177                 },
 178                 [DCCP_PKT_DATA] = {
 179                 /*
 180                  * sNO -> sIV           No connection
 181                  * sRQ -> sIV           No connection
 182                  * sRS -> sIV           No connection
 183                  * sPO -> sIV           MUST use DataAck in PARTOPEN state (8.1.5.)
 184                  * sOP -> sOP           Regular Data packet
 185                  * sCR -> sCR           Data in CLOSEREQ MAY be processed (8.3.)
 186                  * sCG -> sCG           Data in CLOSING MAY be processed (8.3.)
 187                  * sTW -> sIV
 188                  *
 189                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 190                         sIV, sIV, sIV, sIV, sOP, sCR, sCG, sIV,
 191                 },
 192                 [DCCP_PKT_DATAACK] = {
 193                 /*
 194                  * sNO -> sIV           No connection
 195                  * sRQ -> sIV           No connection
 196                  * sRS -> sPO           Ack for Response, move to PARTOPEN (8.1.5.)
 197                  * sPO -> sPO           Remain in PARTOPEN state
 198                  * sOP -> sOP           Regular DataAck packet in OPEN state
 199                  * sCR -> sCR           DataAck in CLOSEREQ MAY be processed (8.3.)
 200                  * sCG -> sCG           DataAck in CLOSING MAY be processed (8.3.)
 201                  * sTW -> sIV
 202                  *
 203                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 204                         sIV, sIV, sPO, sPO, sOP, sCR, sCG, sIV
 205                 },
 206                 [DCCP_PKT_CLOSEREQ] = {
 207                 /*
 208                  * CLOSEREQ may only be sent by the server.
 209                  *
 210                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 211                         sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV
 212                 },
 213                 [DCCP_PKT_CLOSE] = {
 214                 /*
 215                  * sNO -> sIV           No connection
 216                  * sRQ -> sIV           No connection
 217                  * sRS -> sIV           No connection
 218                  * sPO -> sCG           Client-initiated close
 219                  * sOP -> sCG           Client-initiated close
 220                  * sCR -> sCG           Close in response to CloseReq (8.3.)
 221                  * sCG -> sCG           Retransmit
 222                  * sTW -> sIV           Late retransmit, already in TIME_WAIT
 223                  *
 224                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 225                         sIV, sIV, sIV, sCG, sCG, sCG, sIV, sIV
 226                 },
 227                 [DCCP_PKT_RESET] = {
 228                 /*
 229                  * sNO -> sIV           No connection
 230                  * sRQ -> sTW           Sync received or timeout, SHOULD send Reset (8.1.1.)
 231                  * sRS -> sTW           Response received without Request
 232                  * sPO -> sTW           Timeout, SHOULD send Reset (8.1.5.)
 233                  * sOP -> sTW           Connection reset
 234                  * sCR -> sTW           Connection reset
 235                  * sCG -> sTW           Connection reset
 236                  * sTW -> sIG           Ignore (don't refresh timer)
 237                  *
 238                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 239                         sIV, sTW, sTW, sTW, sTW, sTW, sTW, sIG
 240                 },
 241                 [DCCP_PKT_SYNC] = {
 242                 /*
 243                  * We currently ignore Sync packets
 244                  *
 245                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 246                         sIG, sIG, sIG, sIG, sIG, sIG, sIG, sIG,
 247                 },
 248                 [DCCP_PKT_SYNCACK] = {
 249                 /*
 250                  * We currently ignore SyncAck packets
 251                  *
 252                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 253                         sIG, sIG, sIG, sIG, sIG, sIG, sIG, sIG,
 254                 },
 255         },
 256         [CT_DCCP_ROLE_SERVER] = {
 257                 [DCCP_PKT_REQUEST] = {
 258                 /*
 259                  * sNO -> sIV           Invalid
 260                  * sRQ -> sIG           Ignore, conntrack might be out of sync
 261                  * sRS -> sIG           Ignore, conntrack might be out of sync
 262                  * sPO -> sIG           Ignore, conntrack might be out of sync
 263                  * sOP -> sIG           Ignore, conntrack might be out of sync
 264                  * sCR -> sIG           Ignore, conntrack might be out of sync
 265                  * sCG -> sIG           Ignore, conntrack might be out of sync
 266                  * sTW -> sRQ           Reincarnation, must reverse roles
 267                  *
 268                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 269                         sIV, sIG, sIG, sIG, sIG, sIG, sIG, sRQ
 270                 },
 271                 [DCCP_PKT_RESPONSE] = {
 272                 /*
 273                  * sNO -> sIV           Response without Request
 274                  * sRQ -> sRS           Response to clients Request
 275                  * sRS -> sRS           Retransmitted Response (8.1.3. SHOULD NOT)
 276                  * sPO -> sIG           Response to an ignored Request or late retransmit
 277                  * sOP -> sIG           Ignore, might be response to ignored Request
 278                  * sCR -> sIG           Ignore, might be response to ignored Request
 279                  * sCG -> sIG           Ignore, might be response to ignored Request
 280                  * sTW -> sIV           Invalid, Request from client in sTW moves to sRQ
 281                  *
 282                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 283                         sIV, sRS, sRS, sIG, sIG, sIG, sIG, sIV
 284                 },
 285                 [DCCP_PKT_ACK] = {
 286                 /*
 287                  * sNO -> sIV           No connection
 288                  * sRQ -> sIV           No connection
 289                  * sRS -> sIV           No connection
 290                  * sPO -> sOP           Enter OPEN state (8.1.5.)
 291                  * sOP -> sOP           Regular Ack in OPEN state
 292                  * sCR -> sIV           Waiting for Close from client
 293                  * sCG -> sCG           Ack in CLOSING MAY be processed (8.3.)
 294                  * sTW -> sIV
 295                  *
 296                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 297                         sIV, sIV, sIV, sOP, sOP, sIV, sCG, sIV
 298                 },
 299                 [DCCP_PKT_DATA] = {
 300                 /*
 301                  * sNO -> sIV           No connection
 302                  * sRQ -> sIV           No connection
 303                  * sRS -> sIV           No connection
 304                  * sPO -> sOP           Enter OPEN state (8.1.5.)
 305                  * sOP -> sOP           Regular Data packet in OPEN state
 306                  * sCR -> sIV           Waiting for Close from client
 307                  * sCG -> sCG           Data in CLOSING MAY be processed (8.3.)
 308                  * sTW -> sIV
 309                  *
 310                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 311                         sIV, sIV, sIV, sOP, sOP, sIV, sCG, sIV
 312                 },
 313                 [DCCP_PKT_DATAACK] = {
 314                 /*
 315                  * sNO -> sIV           No connection
 316                  * sRQ -> sIV           No connection
 317                  * sRS -> sIV           No connection
 318                  * sPO -> sOP           Enter OPEN state (8.1.5.)
 319                  * sOP -> sOP           Regular DataAck in OPEN state
 320                  * sCR -> sIV           Waiting for Close from client
 321                  * sCG -> sCG           Data in CLOSING MAY be processed (8.3.)
 322                  * sTW -> sIV
 323                  *
 324                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 325                         sIV, sIV, sIV, sOP, sOP, sIV, sCG, sIV
 326                 },
 327                 [DCCP_PKT_CLOSEREQ] = {
 328                 /*
 329                  * sNO -> sIV           No connection
 330                  * sRQ -> sIV           No connection
 331                  * sRS -> sIV           No connection
 332                  * sPO -> sOP -> sCR    Move directly to CLOSEREQ (8.1.5.)
 333                  * sOP -> sCR           CloseReq in OPEN state
 334                  * sCR -> sCR           Retransmit
 335                  * sCG -> sCR           Simultaneous close, client sends another Close
 336                  * sTW -> sIV           Already closed
 337                  *
 338                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 339                         sIV, sIV, sIV, sCR, sCR, sCR, sCR, sIV
 340                 },
 341                 [DCCP_PKT_CLOSE] = {
 342                 /*
 343                  * sNO -> sIV           No connection
 344                  * sRQ -> sIV           No connection
 345                  * sRS -> sIV           No connection
 346                  * sPO -> sOP -> sCG    Move direcly to CLOSING
 347                  * sOP -> sCG           Move to CLOSING
 348                  * sCR -> sIV           Close after CloseReq is invalid
 349                  * sCG -> sCG           Retransmit
 350                  * sTW -> sIV           Already closed
 351                  *
 352                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 353                         sIV, sIV, sIV, sCG, sCG, sIV, sCG, sIV
 354                 },
 355                 [DCCP_PKT_RESET] = {
 356                 /*
 357                  * sNO -> sIV           No connection
 358                  * sRQ -> sTW           Reset in response to Request
 359                  * sRS -> sTW           Timeout, SHOULD send Reset (8.1.3.)
 360                  * sPO -> sTW           Timeout, SHOULD send Reset (8.1.3.)
 361                  * sOP -> sTW
 362                  * sCR -> sTW
 363                  * sCG -> sTW
 364                  * sTW -> sIG           Ignore (don't refresh timer)
 365                  *
 366                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW, sTW */
 367                         sIV, sTW, sTW, sTW, sTW, sTW, sTW, sTW, sIG
 368                 },
 369                 [DCCP_PKT_SYNC] = {
 370                 /*
 371                  * We currently ignore Sync packets
 372                  *
 373                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 374                         sIG, sIG, sIG, sIG, sIG, sIG, sIG, sIG,
 375                 },
 376                 [DCCP_PKT_SYNCACK] = {
 377                 /*
 378                  * We currently ignore SyncAck packets
 379                  *
 380                  *      sNO, sRQ, sRS, sPO, sOP, sCR, sCG, sTW */
 381                         sIG, sIG, sIG, sIG, sIG, sIG, sIG, sIG,
 382                 },
 383         },
 384 };
 385
 386 /* this module per-net specifics */
 387 static int dccp_net_id;
 388 struct dccp_net {
 389         int dccp_loose;
 390         unsigned int dccp_timeout[CT_DCCP_MAX + 1];
 391 #ifdef CONFIG_SYSCTL
 392         struct ctl_table_header *sysctl_header;
 393         struct ctl_table *sysctl_table;
 394 #endif
 395 };
 396
 397 static inline struct dccp_net *dccp_pernet(struct net *net)
 398 {
 399         return net_generic(net, dccp_net_id);
 400 }
 401
 402 static bool dccp_pkt_to_tuple(const struct sk_buff *skb, unsigned int dataoff,
 403                               struct nf_conntrack_tuple *tuple)
 404 {
 405         struct dccp_hdr _hdr, *dh;
 406
 407         dh = skb_header_pointer(skb, dataoff, sizeof(_hdr), &_hdr);
 408         if (dh == NULL)
 409                 return false;
 410
 411         tuple->src.u.dccp.port = dh->dccph_sport;
 412         tuple->dst.u.dccp.port = dh->dccph_dport;
 413         return true;
 414 }
 415
 416 static bool dccp_invert_tuple(struct nf_conntrack_tuple *inv,
 417                               const struct nf_conntrack_tuple *tuple)
 418 {
 419         inv->src.u.dccp.port = tuple->dst.u.dccp.port;
 420         inv->dst.u.dccp.port = tuple->src.u.dccp.port;
 421         return true;
 422 }
 423
 424 static bool dccp_new(struct nf_conn *ct, const struct sk_buff *skb,
 425                      unsigned int dataoff)
 426 {
 427         struct net *net = nf_ct_net(ct);
 428         struct dccp_net *dn;
 429         struct dccp_hdr _dh, *dh;
 430         const char *msg;
 431         u_int8_t state;
 432
 433         dh = skb_header_pointer(skb, dataoff, sizeof(_dh), &dh);
 434         BUG_ON(dh == NULL);
 435
 436         state = dccp_state_table[CT_DCCP_ROLE_CLIENT][dh->dccph_type][CT_DCCP_NONE];
 437         switch (state) {
 438         default:
 439                 dn = dccp_pernet(net);
 440                 if (dn->dccp_loose == 0) {
 441                         msg = "nf_ct_dccp: not picking up existing connection ";
 442                         goto out_invalid;
 443                 }
 444         case CT_DCCP_REQUEST:
 445                 break;
 446         case CT_DCCP_INVALID:
 447                 msg = "nf_ct_dccp: invalid state transition ";
 448                 goto out_invalid;
 449         }
 450
 451         ct->proto.dccp.role[IP_CT_DIR_ORIGINAL] = CT_DCCP_ROLE_CLIENT;
 452         ct->proto.dccp.role[IP_CT_DIR_REPLY] = CT_DCCP_ROLE_SERVER;
 453         ct->proto.dccp.state = CT_DCCP_NONE;
 454         return true;
 455
 456 out_invalid:
 457         if (LOG_INVALID(net, IPPROTO_DCCP))
 458                 nf_log_packet(nf_ct_l3num(ct), 0, skb, NULL, NULL, NULL, msg);
 459         return false;
 460 }
 461
 462 static u64 dccp_ack_seq(const struct dccp_hdr *dh)
 463 {
 464         const struct dccp_hdr_ack_bits *dhack;
 465
 466         dhack = (void *)dh + __dccp_basic_hdr_len(dh);
 467         return ((u64)ntohs(dhack->dccph_ack_nr_high) << 32) +
 468                      ntohl(dhack->dccph_ack_nr_low);
 469 }
 470
 471 static int dccp_packet(struct nf_conn *ct, const struct sk_buff *skb,
 472                        unsigned int dataoff, enum ip_conntrack_info ctinfo,
 473                        u_int8_t pf, unsigned int hooknum)
 474 {
 475         struct net *net = nf_ct_net(ct);
 476         struct dccp_net *dn;
 477         enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
 478         struct dccp_hdr _dh, *dh;
 479         u_int8_t type, old_state, new_state;
 480         enum ct_dccp_roles role;
 481
 482         dh = skb_header_pointer(skb, dataoff, sizeof(_dh), &dh);
 483         BUG_ON(dh == NULL);
 484         type = dh->dccph_type;
 485
 486         if (type == DCCP_PKT_RESET &&
 487             !test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
 488                 /* Tear down connection immediately if only reply is a RESET */
 489                 nf_ct_kill_acct(ct, ctinfo, skb);
 490                 return NF_ACCEPT;
 491         }
 492
 493         spin_lock_bh(&ct->lock);
 494
 495         role = ct->proto.dccp.role[dir];
 496         old_state = ct->proto.dccp.state;
 497         new_state = dccp_state_table[role][type][old_state];
 498
 499         switch (new_state) {
 500         case CT_DCCP_REQUEST:
 501                 if (old_state == CT_DCCP_TIMEWAIT &&
 502                     role == CT_DCCP_ROLE_SERVER) {
 503                         /* Reincarnation in the reverse direction: reopen and
 504                          * reverse client/server roles. */
 505                         ct->proto.dccp.role[dir] = CT_DCCP_ROLE_CLIENT;
 506                         ct->proto.dccp.role[!dir] = CT_DCCP_ROLE_SERVER;
 507                 }
 508                 break;
 509         case CT_DCCP_RESPOND:
 510                 if (old_state == CT_DCCP_REQUEST)
 511                         ct->proto.dccp.handshake_seq = dccp_hdr_seq(dh);
 512                 break;
 513         case CT_DCCP_PARTOPEN:
 514                 if (old_state == CT_DCCP_RESPOND &&
 515                     type == DCCP_PKT_ACK &&
 516                     dccp_ack_seq(dh) == ct->proto.dccp.handshake_seq)
 517                         set_bit(IPS_ASSURED_BIT, &ct->status);
 518                 break;
 519         case CT_DCCP_IGNORE:
 520                 /*
 521                  * Connection tracking might be out of sync, so we ignore
 522                  * packets that might establish a new connection and resync
 523                  * if the server responds with a valid Response.
 524                  */
 525                 if (ct->proto.dccp.last_dir == !dir &&
 526                     ct->proto.dccp.last_pkt == DCCP_PKT_REQUEST &&
 527                     type == DCCP_PKT_RESPONSE) {
 528                         ct->proto.dccp.role[!dir] = CT_DCCP_ROLE_CLIENT;
 529                         ct->proto.dccp.role[dir] = CT_DCCP_ROLE_SERVER;
 530                         ct->proto.dccp.handshake_seq = dccp_hdr_seq(dh);
 531                         new_state = CT_DCCP_RESPOND;
 532                         break;
 533                 }
 534                 ct->proto.dccp.last_dir = dir;
 535                 ct->proto.dccp.last_pkt = type;
 536
 537                 spin_unlock_bh(&ct->lock);
 538                 if (LOG_INVALID(net, IPPROTO_DCCP))
 539                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 540                                       "nf_ct_dccp: invalid packet ignored ");
 541                 return NF_ACCEPT;
 542         case CT_DCCP_INVALID:
 543                 spin_unlock_bh(&ct->lock);
 544                 if (LOG_INVALID(net, IPPROTO_DCCP))
 545                         nf_log_packet(pf, 0, skb, NULL, NULL, NULL,
 546                                       "nf_ct_dccp: invalid state transition ");
 547                 return -NF_ACCEPT;
 548         }
 549
 550         ct->proto.dccp.last_dir = dir;
 551         ct->proto.dccp.last_pkt = type;
 552         ct->proto.dccp.state = new_state;
 553         spin_unlock_bh(&ct->lock);
 554
 555         if (new_state != old_state)
 556                 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
 557
 558         dn = dccp_pernet(net);
 559         nf_ct_refresh_acct(ct, ctinfo, skb, dn->dccp_timeout[new_state]);
 560
 561         return NF_ACCEPT;
 562 }
 563
 564 static int dccp_error(struct net *net, struct sk_buff *skb,
 565                       unsigned int dataoff, enum ip_conntrack_info *ctinfo,
 566                       u_int8_t pf, unsigned int hooknum)
 567 {
 568         struct dccp_hdr _dh, *dh;
 569         unsigned int dccp_len = skb->len - dataoff;
 570         unsigned int cscov;
 571         const char *msg;
 572
 573         dh = skb_header_pointer(skb, dataoff, sizeof(_dh), &dh);
 574         if (dh == NULL) {
 575                 msg = "nf_ct_dccp: short packet ";
 576                 goto out_invalid;
 577         }
 578
 579         if (dh->dccph_doff * 4 < sizeof(struct dccp_hdr) ||
 580             dh->dccph_doff * 4 > dccp_len) {
 581                 msg = "nf_ct_dccp: truncated/malformed packet ";
 582                 goto out_invalid;
 583         }
 584
 585         cscov = dccp_len;
 586         if (dh->dccph_cscov) {
 587                 cscov = (dh->dccph_cscov - 1) * 4;
 588                 if (cscov > dccp_len) {
 589                         msg = "nf_ct_dccp: bad checksum coverage ";
 590                         goto out_invalid;
 591                 }
 592         }
 593
 594         if (net->ct.sysctl_checksum && hooknum == NF_INET_PRE_ROUTING &&
 595             nf_checksum_partial(skb, hooknum, dataoff, cscov, IPPROTO_DCCP,
 596                                 pf)) {
 597                 msg = "nf_ct_dccp: bad checksum ";
 598                 goto out_invalid;
 599         }
 600
 601         if (dh->dccph_type >= DCCP_PKT_INVALID) {
 602                 msg = "nf_ct_dccp: reserved packet type ";
 603                 goto out_invalid;
 604         }
 605
 606         return NF_ACCEPT;
 607
 608 out_invalid:
 609         if (LOG_INVALID(net, IPPROTO_DCCP))
 610                 nf_log_packet(pf, 0, skb, NULL, NULL, NULL, msg);
 611         return -NF_ACCEPT;
 612 }
 613
 614 static int dccp_print_tuple(struct seq_file *s,
 615                             const struct nf_conntrack_tuple *tuple)
 616 {
 617         return seq_printf(s, "sport=%hu dport=%hu ",
 618                           ntohs(tuple->src.u.dccp.port),
 619                           ntohs(tuple->dst.u.dccp.port));
 620 }
 621
 622 static int dccp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
 623 {
 624         return seq_printf(s, "%s ", dccp_state_names[ct->proto.dccp.state]);
 625 }
 626
 627 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 628 static int dccp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
 629                           struct nf_conn *ct)
 630 {
 631         struct nlattr *nest_parms;
 632
 633         spin_lock_bh(&ct->lock);
 634         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_DCCP | NLA_F_NESTED);
 635         if (!nest_parms)
 636                 goto nla_put_failure;
 637         NLA_PUT_U8(skb, CTA_PROTOINFO_DCCP_STATE, ct->proto.dccp.state);
 638         NLA_PUT_U8(skb, CTA_PROTOINFO_DCCP_ROLE,
 639                    ct->proto.dccp.role[IP_CT_DIR_ORIGINAL]);
 640         NLA_PUT_BE64(skb, CTA_PROTOINFO_DCCP_HANDSHAKE_SEQ,
 641                      cpu_to_be64(ct->proto.dccp.handshake_seq));
 642         nla_nest_end(skb, nest_parms);
 643         spin_unlock_bh(&ct->lock);
 644         return 0;
 645
 646 nla_put_failure:
 647         spin_unlock_bh(&ct->lock);
 648         return -1;
 649 }
 650
 651 static const struct nla_policy dccp_nla_policy[CTA_PROTOINFO_DCCP_MAX + 1] = {
 652         [CTA_PROTOINFO_DCCP_STATE]      = { .type = NLA_U8 },
 653         [CTA_PROTOINFO_DCCP_ROLE]       = { .type = NLA_U8 },
 654         [CTA_PROTOINFO_DCCP_HANDSHAKE_SEQ] = { .type = NLA_U64 },
 655 };
 656
 657 static int nlattr_to_dccp(struct nlattr *cda[], struct nf_conn *ct)
 658 {
 659         struct nlattr *attr = cda[CTA_PROTOINFO_DCCP];
 660         struct nlattr *tb[CTA_PROTOINFO_DCCP_MAX + 1];
 661         int err;
 662
 663         if (!attr)
 664                 return 0;
 665
 666         err = nla_parse_nested(tb, CTA_PROTOINFO_DCCP_MAX, attr,
 667                                dccp_nla_policy);
 668         if (err < 0)
 669                 return err;
 670
 671         if (!tb[CTA_PROTOINFO_DCCP_STATE] ||
 672             !tb[CTA_PROTOINFO_DCCP_ROLE] ||
 673             nla_get_u8(tb[CTA_PROTOINFO_DCCP_ROLE]) > CT_DCCP_ROLE_MAX ||
 674             nla_get_u8(tb[CTA_PROTOINFO_DCCP_STATE]) >= CT_DCCP_IGNORE) {
 675                 return -EINVAL;
 676         }
 677
 678         spin_lock_bh(&ct->lock);
 679         ct->proto.dccp.state = nla_get_u8(tb[CTA_PROTOINFO_DCCP_STATE]);
 680         if (nla_get_u8(tb[CTA_PROTOINFO_DCCP_ROLE]) == CT_DCCP_ROLE_CLIENT) {
 681                 ct->proto.dccp.role[IP_CT_DIR_ORIGINAL] = CT_DCCP_ROLE_CLIENT;
 682                 ct->proto.dccp.role[IP_CT_DIR_REPLY] = CT_DCCP_ROLE_SERVER;
 683         } else {
 684                 ct->proto.dccp.role[IP_CT_DIR_ORIGINAL] = CT_DCCP_ROLE_SERVER;
 685                 ct->proto.dccp.role[IP_CT_DIR_REPLY] = CT_DCCP_ROLE_CLIENT;
 686         }
 687         if (tb[CTA_PROTOINFO_DCCP_HANDSHAKE_SEQ]) {
 688                 ct->proto.dccp.handshake_seq =
 689                 be64_to_cpu(nla_get_be64(tb[CTA_PROTOINFO_DCCP_HANDSHAKE_SEQ]));
 690         }
 691         spin_unlock_bh(&ct->lock);
 692         return 0;
 693 }
 694
 695 static int dccp_nlattr_size(void)
 696 {
 697         return nla_total_size(0)        /* CTA_PROTOINFO_DCCP */
 698                 + nla_policy_len(dccp_nla_policy, CTA_PROTOINFO_DCCP_MAX + 1);
 699 }
 700 #endif
 701
 702 #ifdef CONFIG_SYSCTL
 703 /* template, data assigned later */
 704 static struct ctl_table dccp_sysctl_table[] = {
 705         {
 706                 .ctl_name       = CTL_UNNUMBERED,
 707                 .procname       = "nf_conntrack_dccp_timeout_request",
 708                 .maxlen         = sizeof(unsigned int),
 709                 .mode           = 0644,
 710                 .proc_handler   = proc_dointvec_jiffies,
 711         },
 712         {
 713                 .ctl_name       = CTL_UNNUMBERED,
 714                 .procname       = "nf_conntrack_dccp_timeout_respond",
 715                 .maxlen         = sizeof(unsigned int),
 716                 .mode           = 0644,
 717                 .proc_handler   = proc_dointvec_jiffies,
 718         },
 719         {
 720                 .ctl_name       = CTL_UNNUMBERED,
 721                 .procname       = "nf_conntrack_dccp_timeout_partopen",
 722                 .maxlen         = sizeof(unsigned int),
 723                 .mode           = 0644,
 724                 .proc_handler   = proc_dointvec_jiffies,
 725         },
 726         {
 727                 .ctl_name       = CTL_UNNUMBERED,
 728                 .procname       = "nf_conntrack_dccp_timeout_open",
 729                 .maxlen         = sizeof(unsigned int),
 730                 .mode           = 0644,
 731                 .proc_handler   = proc_dointvec_jiffies,
 732         },
 733         {
 734                 .ctl_name       = CTL_UNNUMBERED,
 735                 .procname       = "nf_conntrack_dccp_timeout_closereq",
 736                 .maxlen         = sizeof(unsigned int),
 737                 .mode           = 0644,
 738                 .proc_handler   = proc_dointvec_jiffies,
 739         },
 740         {
 741                 .ctl_name       = CTL_UNNUMBERED,
 742                 .procname       = "nf_conntrack_dccp_timeout_closing",
 743                 .maxlen         = sizeof(unsigned int),
 744                 .mode           = 0644,
 745                 .proc_handler   = proc_dointvec_jiffies,
 746         },
 747         {
 748                 .ctl_name       = CTL_UNNUMBERED,
 749                 .procname       = "nf_conntrack_dccp_timeout_timewait",
 750                 .maxlen         = sizeof(unsigned int),
 751                 .mode           = 0644,
 752                 .proc_handler   = proc_dointvec_jiffies,
 753         },
 754         {
 755                 .ctl_name       = CTL_UNNUMBERED,
 756                 .procname       = "nf_conntrack_dccp_loose",
 757                 .maxlen         = sizeof(int),
 758                 .mode           = 0644,
 759                 .proc_handler   = proc_dointvec,
 760         },
 761         {
 762                 .ctl_name       = 0,
 763         }
 764 };
 765 #endif /* CONFIG_SYSCTL */
 766
 767 static struct nf_conntrack_l4proto dccp_proto4 __read_mostly = {
 768         .l3proto                = AF_INET,
 769         .l4proto                = IPPROTO_DCCP,
 770         .name                   = "dccp",
 771         .pkt_to_tuple           = dccp_pkt_to_tuple,
 772         .invert_tuple           = dccp_invert_tuple,
 773         .new                    = dccp_new,
 774         .packet                 = dccp_packet,
 775         .error                  = dccp_error,
 776         .print_tuple            = dccp_print_tuple,
 777         .print_conntrack        = dccp_print_conntrack,
 778 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 779         .to_nlattr              = dccp_to_nlattr,
 780         .nlattr_size            = dccp_nlattr_size,
 781         .from_nlattr            = nlattr_to_dccp,
 782         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
 783         .nlattr_tuple_size      = nf_ct_port_nlattr_tuple_size,
 784         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
 785         .nla_policy             = nf_ct_port_nla_policy,
 786 #endif
 787 };
 788
 789 static struct nf_conntrack_l4proto dccp_proto6 __read_mostly = {
 790         .l3proto                = AF_INET6,
 791         .l4proto                = IPPROTO_DCCP,
 792         .name                   = "dccp",
 793         .pkt_to_tuple           = dccp_pkt_to_tuple,
 794         .invert_tuple           = dccp_invert_tuple,
 795         .new                    = dccp_new,
 796         .packet                 = dccp_packet,
 797         .error                  = dccp_error,
 798         .print_tuple            = dccp_print_tuple,
 799         .print_conntrack        = dccp_print_conntrack,
 800 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 801         .to_nlattr              = dccp_to_nlattr,
 802         .nlattr_size            = dccp_nlattr_size,
 803         .from_nlattr            = nlattr_to_dccp,
 804         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
 805         .nlattr_tuple_size      = nf_ct_port_nlattr_tuple_size,
 806         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
 807         .nla_policy             = nf_ct_port_nla_policy,
 808 #endif
 809 };
 810
 811 static __net_init int dccp_net_init(struct net *net)
 812 {
 813         struct dccp_net *dn;
 814         int err;
 815
 816         dn = kmalloc(sizeof(*dn), GFP_KERNEL);
 817         if (!dn)
 818                 return -ENOMEM;
 819
 820         /* default values */
 821         dn->dccp_loose = 1;
 822         dn->dccp_timeout[CT_DCCP_REQUEST]       = 2 * DCCP_MSL;
 823         dn->dccp_timeout[CT_DCCP_RESPOND]       = 4 * DCCP_MSL;
 824         dn->dccp_timeout[CT_DCCP_PARTOPEN]      = 4 * DCCP_MSL;
 825         dn->dccp_timeout[CT_DCCP_OPEN]          = 12 * 3600 * HZ;
 826         dn->dccp_timeout[CT_DCCP_CLOSEREQ]      = 64 * HZ;
 827         dn->dccp_timeout[CT_DCCP_CLOSING]       = 64 * HZ;
 828         dn->dccp_timeout[CT_DCCP_TIMEWAIT]      = 2 * DCCP_MSL;
 829
 830         err = net_assign_generic(net, dccp_net_id, dn);
 831         if (err)
 832                 goto out;
 833
 834 #ifdef CONFIG_SYSCTL
 835         err = -ENOMEM;
 836         dn->sysctl_table = kmemdup(dccp_sysctl_table,
 837                         sizeof(dccp_sysctl_table), GFP_KERNEL);
 838         if (!dn->sysctl_table)
 839                 goto out;
 840
 841         dn->sysctl_table[0].data = &dn->dccp_timeout[CT_DCCP_REQUEST];
 842         dn->sysctl_table[1].data = &dn->dccp_timeout[CT_DCCP_RESPOND];
 843         dn->sysctl_table[2].data = &dn->dccp_timeout[CT_DCCP_PARTOPEN];
 844         dn->sysctl_table[3].data = &dn->dccp_timeout[CT_DCCP_OPEN];
 845         dn->sysctl_table[4].data = &dn->dccp_timeout[CT_DCCP_CLOSEREQ];
 846         dn->sysctl_table[5].data = &dn->dccp_timeout[CT_DCCP_CLOSING];
 847         dn->sysctl_table[6].data = &dn->dccp_timeout[CT_DCCP_TIMEWAIT];
 848         dn->sysctl_table[7].data = &dn->dccp_loose;
 849
 850         dn->sysctl_header = register_net_sysctl_table(net,
 851                         nf_net_netfilter_sysctl_path, dn->sysctl_table);
 852         if (!dn->sysctl_header) {
 853                 kfree(dn->sysctl_table);
 854                 goto out;
 855         }
 856 #endif
 857
 858         return 0;
 859
 860 out:
 861         kfree(dn);
 862         return err;
 863 }
 864
 865 static __net_exit void dccp_net_exit(struct net *net)
 866 {
 867         struct dccp_net *dn = dccp_pernet(net);
 868 #ifdef CONFIG_SYSCTL
 869         unregister_net_sysctl_table(dn->sysctl_header);
 870         kfree(dn->sysctl_table);
 871 #endif
 872         kfree(dn);
 873
 874         net_assign_generic(net, dccp_net_id, NULL);
 875 }
 876
 877 static struct pernet_operations dccp_net_ops = {
 878         .init = dccp_net_init,
 879         .exit = dccp_net_exit,
 880 };
 881
 882 static int __init nf_conntrack_proto_dccp_init(void)
 883 {
 884         int err;
 885
 886         err = register_pernet_gen_subsys(&dccp_net_id, &dccp_net_ops);
 887         if (err < 0)
 888                 goto err1;
 889
 890         err = nf_conntrack_l4proto_register(&dccp_proto4);
 891         if (err < 0)
 892                 goto err2;
 893
 894         err = nf_conntrack_l4proto_register(&dccp_proto6);
 895         if (err < 0)
 896                 goto err3;
 897         return 0;
 898
 899 err3:
 900         nf_conntrack_l4proto_unregister(&dccp_proto4);
 901 err2:
 902         unregister_pernet_gen_subsys(dccp_net_id, &dccp_net_ops);
 903 err1:
 904         return err;
 905 }
 906
 907 static void __exit nf_conntrack_proto_dccp_fini(void)
 908 {
 909         unregister_pernet_gen_subsys(dccp_net_id, &dccp_net_ops);
 910         nf_conntrack_l4proto_unregister(&dccp_proto6);
 911         nf_conntrack_l4proto_unregister(&dccp_proto4);
 912 }
 913
 914 module_init(nf_conntrack_proto_dccp_init);
 915 module_exit(nf_conntrack_proto_dccp_fini);
 916
 917 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
 918 MODULE_DESCRIPTION("DCCP connection tracking protocol helper");
 919 MODULE_LICENSE("GPL");