net/ipv4/ipvs/ip_vs_proto_tcp.c

   1 /*
   2  * ip_vs_proto_tcp.c:   TCP load balancing support for IPVS
   3  *
   4  * Version:     $Id: ip_vs_proto_tcp.c,v 1.3 2002/11/30 01:50:35 wensong Exp $
   5  *
   6  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   7  *              Julian Anastasov <ja@ssi.bg>
   8  *
   9  *              This program is free software; you can redistribute it and/or
  10  *              modify it under the terms of the GNU General Public License
  11  *              as published by the Free Software Foundation; either version
  12  *              2 of the License, or (at your option) any later version.
  13  *
  14  * Changes:
  15  *
  16  */
  17
  18 #include <linux/config.h>
  19 #include <linux/compiler.h>
  20 #include <linux/ip.h>
  21 #include <linux/tcp.h>                  /* for tcphdr */
  22 #include <net/ip.h>
  23 #include <net/tcp.h>                    /* for csum_tcpudp_magic */
  24 #include <linux/netfilter_ipv4.h>
  25
  26 #include <net/ip_vs.h>
  27
  28
  29 static struct ip_vs_conn *
  30 tcp_conn_in_get(const struct sk_buff *skb, struct ip_vs_protocol *pp,
  31                 const struct iphdr *iph, unsigned int proto_off, int inverse)
  32 {
  33         __u16 ports[2];
  34
  35         if (skb_copy_bits(skb, proto_off, ports, sizeof(ports)) < 0)
  36                 return NULL;
  37
  38         if (likely(!inverse)) {
  39                 return ip_vs_conn_in_get(iph->protocol,
  40                                          iph->saddr, ports[0],
  41                                          iph->daddr, ports[1]);
  42         } else {
  43                 return ip_vs_conn_in_get(iph->protocol,
  44                                          iph->daddr, ports[1],
  45                                          iph->saddr, ports[0]);
  46         }
  47 }
  48
  49 static struct ip_vs_conn *
  50 tcp_conn_out_get(const struct sk_buff *skb, struct ip_vs_protocol *pp,
  51                  const struct iphdr *iph, unsigned int proto_off, int inverse)
  52 {
  53         __u16 ports[2];
  54
  55         if (skb_copy_bits(skb, proto_off, ports, sizeof(ports)) < 0)
  56                 return NULL;
  57
  58         if (likely(!inverse)) {
  59                 return ip_vs_conn_out_get(iph->protocol,
  60                                           iph->saddr, ports[0],
  61                                           iph->daddr, ports[1]);
  62         } else {
  63                 return ip_vs_conn_out_get(iph->protocol,
  64                                           iph->daddr, ports[1],
  65                                           iph->saddr, ports[0]);
  66         }
  67 }
  68
  69
  70 static int
  71 tcp_conn_schedule(struct sk_buff *skb,
  72                   struct ip_vs_protocol *pp,
  73                   int *verdict, struct ip_vs_conn **cpp)
  74 {
  75         struct ip_vs_service *svc;
  76         struct tcphdr tcph;
  77
  78         if (skb_copy_bits(skb, skb->nh.iph->ihl*4, &tcph, sizeof(tcph)) < 0) {
  79                 *verdict = NF_DROP;
  80                 return 0;
  81         }
  82
  83         if (tcph.syn &&
  84             (svc = ip_vs_service_get(skb->nfmark, skb->nh.iph->protocol,
  85                                      skb->nh.iph->daddr, tcph.dest))) {
  86                 if (ip_vs_todrop()) {
  87                         /*
  88                          * It seems that we are very loaded.
  89                          * We have to drop this packet :(
  90                          */
  91                         ip_vs_service_put(svc);
  92                         *verdict = NF_DROP;
  93                         return 0;
  94                 }
  95
  96                 /*
  97                  * Let the virtual server select a real server for the
  98                  * incoming connection, and create a connection entry.
  99                  */
 100                 *cpp = ip_vs_schedule(svc, skb);
 101                 if (!*cpp) {
 102                         *verdict = ip_vs_leave(svc, skb, pp);
 103                         return 0;
 104                 }
 105                 ip_vs_service_put(svc);
 106         }
 107         return 1;
 108 }
 109
 110
 111 static inline void
 112 tcp_fast_csum_update(struct tcphdr *tcph, u32 oldip, u32 newip,
 113                      u16 oldport, u16 newport)
 114 {
 115         tcph->check =
 116                 ip_vs_check_diff(~oldip, newip,
 117                                  ip_vs_check_diff(oldport ^ 0xFFFF,
 118                                                   newport, tcph->check));
 119 }
 120
 121
 122 static int
 123 tcp_snat_handler(struct sk_buff **pskb,
 124                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
 125 {
 126         struct tcphdr *tcph;
 127         unsigned int tcphoff = (*pskb)->nh.iph->ihl * 4;
 128
 129         /* csum_check requires unshared skb */
 130         if (!ip_vs_make_skb_writable(pskb, tcphoff+sizeof(*tcph)))
 131                 return 0;
 132
 133         if (unlikely(cp->app != NULL)) {
 134                 /* Some checks before mangling */
 135                 if (pp->csum_check && !pp->csum_check(*pskb, pp))
 136                         return 0;
 137
 138                 /* Call application helper if needed */
 139                 if (!ip_vs_app_pkt_out(cp, pskb))
 140                         return 0;
 141         }
 142
 143         tcph = (void *)(*pskb)->nh.iph + tcphoff;
 144         tcph->source = cp->vport;
 145
 146         /* Adjust TCP checksums */
 147         if (!cp->app) {
 148                 /* Only port and addr are changed, do fast csum update */
 149                 tcp_fast_csum_update(tcph, cp->daddr, cp->vaddr,
 150                                      cp->dport, cp->vport);
 151                 if ((*pskb)->ip_summed == CHECKSUM_HW)
 152                         (*pskb)->ip_summed = CHECKSUM_NONE;
 153         } else {
 154                 /* full checksum calculation */
 155                 tcph->check = 0;
 156                 (*pskb)->csum = skb_checksum(*pskb, tcphoff,
 157                                              (*pskb)->len - tcphoff, 0);
 158                 tcph->check = csum_tcpudp_magic(cp->vaddr, cp->caddr,
 159                                                 (*pskb)->len - tcphoff,
 160                                                 cp->protocol,
 161                                                 (*pskb)->csum);
 162                 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%d)\n",
 163                           pp->name, tcph->check,
 164                           (char*)&(tcph->check) - (char*)tcph);
 165         }
 166         return 1;
 167 }
 168
 169
 170 static int
 171 tcp_dnat_handler(struct sk_buff **pskb,
 172                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
 173 {
 174         struct tcphdr *tcph;
 175         unsigned int tcphoff = (*pskb)->nh.iph->ihl * 4;
 176
 177         /* csum_check requires unshared skb */
 178         if (!ip_vs_make_skb_writable(pskb, tcphoff+sizeof(*tcph)))
 179                 return 0;
 180
 181         if (unlikely(cp->app != NULL)) {
 182                 /* Some checks before mangling */
 183                 if (pp->csum_check && !pp->csum_check(*pskb, pp))
 184                         return 0;
 185
 186                 /*
 187                  *      Attempt ip_vs_app call.
 188                  *      It will fix ip_vs_conn and iph ack_seq stuff
 189                  */
 190                 if (!ip_vs_app_pkt_in(cp, pskb))
 191                         return 0;
 192         }
 193
 194         tcph = (void *)(*pskb)->nh.iph + tcphoff;
 195         tcph->dest = cp->dport;
 196
 197         /*
 198          *      Adjust TCP checksums
 199          */
 200         if (!cp->app) {
 201                 /* Only port and addr are changed, do fast csum update */
 202                 tcp_fast_csum_update(tcph, cp->vaddr, cp->daddr,
 203                                      cp->vport, cp->dport);
 204                 if ((*pskb)->ip_summed == CHECKSUM_HW)
 205                         (*pskb)->ip_summed = CHECKSUM_NONE;
 206         } else {
 207                 /* full checksum calculation */
 208                 tcph->check = 0;
 209                 (*pskb)->csum = skb_checksum(*pskb, tcphoff,
 210                                              (*pskb)->len - tcphoff, 0);
 211                 tcph->check = csum_tcpudp_magic(cp->caddr, cp->daddr,
 212                                                 (*pskb)->len - tcphoff,
 213                                                 cp->protocol,
 214                                                 (*pskb)->csum);
 215                 (*pskb)->ip_summed = CHECKSUM_UNNECESSARY;
 216         }
 217         return 1;
 218 }
 219
 220
 221 static int
 222 tcp_csum_check(struct sk_buff *skb, struct ip_vs_protocol *pp)
 223 {
 224         unsigned int tcphoff = skb->nh.iph->ihl*4;
 225
 226         switch (skb->ip_summed) {
 227         case CHECKSUM_NONE:
 228                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 229         case CHECKSUM_HW:
 230                 if (csum_tcpudp_magic(skb->nh.iph->saddr, skb->nh.iph->daddr,
 231                                       skb->len - tcphoff,
 232                                       skb->nh.iph->protocol, skb->csum)) {
 233                         IP_VS_DBG_RL_PKT(0, pp, skb, 0,
 234                                          "Failed checksum for");
 235                         return 0;
 236                 }
 237                 break;
 238         default:
 239                 /* CHECKSUM_UNNECESSARY */
 240                 break;
 241         }
 242
 243         return 1;
 244 }
 245
 246
 247 #define TCP_DIR_INPUT           0
 248 #define TCP_DIR_OUTPUT          4
 249 #define TCP_DIR_INPUT_ONLY      8
 250
 251 static int tcp_state_off[IP_VS_DIR_LAST] = {
 252         [IP_VS_DIR_INPUT]               =       TCP_DIR_INPUT,
 253         [IP_VS_DIR_OUTPUT]              =       TCP_DIR_OUTPUT,
 254         [IP_VS_DIR_INPUT_ONLY]          =       TCP_DIR_INPUT_ONLY,
 255 };
 256
 257 /*
 258  *      Timeout table[state]
 259  */
 260 static int tcp_timeouts[IP_VS_TCP_S_LAST+1] = {
 261         [IP_VS_TCP_S_NONE]              =       2*HZ,
 262         [IP_VS_TCP_S_ESTABLISHED]       =       15*60*HZ,
 263         [IP_VS_TCP_S_SYN_SENT]          =       2*60*HZ,
 264         [IP_VS_TCP_S_SYN_RECV]          =       1*60*HZ,
 265         [IP_VS_TCP_S_FIN_WAIT]          =       2*60*HZ,
 266         [IP_VS_TCP_S_TIME_WAIT]         =       2*60*HZ,
 267         [IP_VS_TCP_S_CLOSE]             =       10*HZ,
 268         [IP_VS_TCP_S_CLOSE_WAIT]        =       60*HZ,
 269         [IP_VS_TCP_S_LAST_ACK]          =       30*HZ,
 270         [IP_VS_TCP_S_LISTEN]            =       2*60*HZ,
 271         [IP_VS_TCP_S_SYNACK]            =       120*HZ,
 272         [IP_VS_TCP_S_LAST]              =       2*HZ,
 273 };
 274
 275
 276 #if 0
 277
 278 /* FIXME: This is going to die */
 279
 280 static int tcp_timeouts_dos[IP_VS_TCP_S_LAST+1] = {
 281         [IP_VS_TCP_S_NONE]              =       2*HZ,
 282         [IP_VS_TCP_S_ESTABLISHED]       =       8*60*HZ,
 283         [IP_VS_TCP_S_SYN_SENT]          =       60*HZ,
 284         [IP_VS_TCP_S_SYN_RECV]          =       10*HZ,
 285         [IP_VS_TCP_S_FIN_WAIT]          =       60*HZ,
 286         [IP_VS_TCP_S_TIME_WAIT]         =       60*HZ,
 287         [IP_VS_TCP_S_CLOSE]             =       10*HZ,
 288         [IP_VS_TCP_S_CLOSE_WAIT]        =       60*HZ,
 289         [IP_VS_TCP_S_LAST_ACK]          =       30*HZ,
 290         [IP_VS_TCP_S_LISTEN]            =       2*60*HZ,
 291         [IP_VS_TCP_S_SYNACK]            =       100*HZ,
 292         [IP_VS_TCP_S_LAST]              =       2*HZ,
 293 };
 294
 295 #endif
 296
 297 static char * tcp_state_name_table[IP_VS_TCP_S_LAST+1] = {
 298         [IP_VS_TCP_S_NONE]              =       "NONE",
 299         [IP_VS_TCP_S_ESTABLISHED]       =       "ESTABLISHED",
 300         [IP_VS_TCP_S_SYN_SENT]          =       "SYN_SENT",
 301         [IP_VS_TCP_S_SYN_RECV]          =       "SYN_RECV",
 302         [IP_VS_TCP_S_FIN_WAIT]          =       "FIN_WAIT",
 303         [IP_VS_TCP_S_TIME_WAIT]         =       "TIME_WAIT",
 304         [IP_VS_TCP_S_CLOSE]             =       "CLOSE",
 305         [IP_VS_TCP_S_CLOSE_WAIT]        =       "CLOSE_WAIT",
 306         [IP_VS_TCP_S_LAST_ACK]          =       "LAST_ACK",
 307         [IP_VS_TCP_S_LISTEN]            =       "LISTEN",
 308         [IP_VS_TCP_S_SYNACK]            =       "SYNACK",
 309         [IP_VS_TCP_S_LAST]              =       "BUG!",
 310 };
 311
 312 #define sNO IP_VS_TCP_S_NONE
 313 #define sES IP_VS_TCP_S_ESTABLISHED
 314 #define sSS IP_VS_TCP_S_SYN_SENT
 315 #define sSR IP_VS_TCP_S_SYN_RECV
 316 #define sFW IP_VS_TCP_S_FIN_WAIT
 317 #define sTW IP_VS_TCP_S_TIME_WAIT
 318 #define sCL IP_VS_TCP_S_CLOSE
 319 #define sCW IP_VS_TCP_S_CLOSE_WAIT
 320 #define sLA IP_VS_TCP_S_LAST_ACK
 321 #define sLI IP_VS_TCP_S_LISTEN
 322 #define sSA IP_VS_TCP_S_SYNACK
 323
 324 struct tcp_states_t {
 325         int next_state[IP_VS_TCP_S_LAST];
 326 };
 327
 328 static const char * tcp_state_name(int state)
 329 {
 330         if (state >= IP_VS_TCP_S_LAST)
 331                 return "ERR!";
 332         return tcp_state_name_table[state] ? tcp_state_name_table[state] : "?";
 333 }
 334
 335 static struct tcp_states_t tcp_states [] = {
 336 /*      INPUT */
 337 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 338 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }},
 339 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sTW }},
 340 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 341 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sSR }},
 342
 343 /*      OUTPUT */
 344 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 345 /*syn*/ {{sSS, sES, sSS, sSR, sSS, sSS, sSS, sSS, sSS, sLI, sSR }},
 346 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }},
 347 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }},
 348 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }},
 349
 350 /*      INPUT-ONLY */
 351 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 352 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }},
 353 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }},
 354 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 355 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 356 };
 357
 358 static struct tcp_states_t tcp_states_dos [] = {
 359 /*      INPUT */
 360 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 361 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSA }},
 362 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sSA }},
 363 /*ack*/ {{sCL, sES, sSS, sSR, sFW, sTW, sCL, sCW, sCL, sLI, sSA }},
 364 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 365
 366 /*      OUTPUT */
 367 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 368 /*syn*/ {{sSS, sES, sSS, sSA, sSS, sSS, sSS, sSS, sSS, sLI, sSA }},
 369 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }},
 370 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }},
 371 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }},
 372
 373 /*      INPUT-ONLY */
 374 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 375 /*syn*/ {{sSA, sES, sES, sSR, sSA, sSA, sSA, sSA, sSA, sSA, sSA }},
 376 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }},
 377 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 378 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 379 };
 380
 381 static struct tcp_states_t *tcp_state_table = tcp_states;
 382
 383
 384 static void tcp_timeout_change(struct ip_vs_protocol *pp, int flags)
 385 {
 386         int on = (flags & 1);           /* secure_tcp */
 387
 388         /*
 389         ** FIXME: change secure_tcp to independent sysctl var
 390         ** or make it per-service or per-app because it is valid
 391         ** for most if not for all of the applications. Something
 392         ** like "capabilities" (flags) for each object.
 393         */
 394         tcp_state_table = (on? tcp_states_dos : tcp_states);
 395 }
 396
 397 static int
 398 tcp_set_state_timeout(struct ip_vs_protocol *pp, char *sname, int to)
 399 {
 400         return ip_vs_set_state_timeout(pp->timeout_table, IP_VS_TCP_S_LAST,
 401                                        tcp_state_name_table, sname, to);
 402 }
 403
 404 static inline int tcp_state_idx(struct tcphdr *th)
 405 {
 406         if (th->rst)
 407                 return 3;
 408         if (th->syn)
 409                 return 0;
 410         if (th->fin)
 411                 return 1;
 412         if (th->ack)
 413                 return 2;
 414         return -1;
 415 }
 416
 417 static inline void
 418 set_tcp_state(struct ip_vs_protocol *pp, struct ip_vs_conn *cp,
 419               int direction, struct tcphdr *th)
 420 {
 421         int state_idx;
 422         int new_state = IP_VS_TCP_S_CLOSE;
 423         int state_off = tcp_state_off[direction];
 424
 425         /*
 426          *    Update state offset to INPUT_ONLY if necessary
 427          *    or delete NO_OUTPUT flag if output packet detected
 428          */
 429         if (cp->flags & IP_VS_CONN_F_NOOUTPUT) {
 430                 if (state_off == TCP_DIR_OUTPUT)
 431                         cp->flags &= ~IP_VS_CONN_F_NOOUTPUT;
 432                 else
 433                         state_off = TCP_DIR_INPUT_ONLY;
 434         }
 435
 436         if ((state_idx = tcp_state_idx(th)) < 0) {
 437                 IP_VS_DBG(8, "tcp_state_idx=%d!!!\n", state_idx);
 438                 goto tcp_state_out;
 439         }
 440
 441         new_state = tcp_state_table[state_off+state_idx].next_state[cp->state];
 442
 443   tcp_state_out:
 444         if (new_state != cp->state) {
 445                 struct ip_vs_dest *dest = cp->dest;
 446
 447                 IP_VS_DBG(8, "%s %s [%c%c%c%c] %u.%u.%u.%u:%d->"
 448                           "%u.%u.%u.%u:%d state: %s->%s cnt:%d\n",
 449                           pp->name,
 450                           (state_off==TCP_DIR_OUTPUT)?"output ":"input ",
 451                           th->syn? 'S' : '.',
 452                           th->fin? 'F' : '.',
 453                           th->ack? 'A' : '.',
 454                           th->rst? 'R' : '.',
 455                           NIPQUAD(cp->daddr), ntohs(cp->dport),
 456                           NIPQUAD(cp->caddr), ntohs(cp->cport),
 457                           tcp_state_name(cp->state),
 458                           tcp_state_name(new_state),
 459                           atomic_read(&cp->refcnt));
 460                 if (dest) {
 461                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 462                             (new_state != IP_VS_TCP_S_ESTABLISHED)) {
 463                                 atomic_dec(&dest->activeconns);
 464                                 atomic_inc(&dest->inactconns);
 465                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
 466                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
 467                                    (new_state == IP_VS_TCP_S_ESTABLISHED)) {
 468                                 atomic_inc(&dest->activeconns);
 469                                 atomic_dec(&dest->inactconns);
 470                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 471                         }
 472                 }
 473         }
 474
 475         cp->timeout = pp->timeout_table[cp->state = new_state];
 476 }
 477
 478
 479 /*
 480  *      Handle state transitions
 481  */
 482 static int
 483 tcp_state_transition(struct ip_vs_conn *cp, int direction,
 484                      const struct sk_buff *skb,
 485                      struct ip_vs_protocol *pp)
 486 {
 487         struct tcphdr tcph;
 488
 489         if (skb_copy_bits(skb, skb->nh.iph->ihl*4, &tcph, sizeof(tcph)) < 0)
 490                 return 0;
 491
 492         spin_lock(&cp->lock);
 493         set_tcp_state(pp, cp, direction, &tcph);
 494         spin_unlock(&cp->lock);
 495
 496         return 1;
 497 }
 498
 499
 500 /*
 501  *      Hash table for TCP application incarnations
 502  */
 503 #define TCP_APP_TAB_BITS        4
 504 #define TCP_APP_TAB_SIZE        (1 << TCP_APP_TAB_BITS)
 505 #define TCP_APP_TAB_MASK        (TCP_APP_TAB_SIZE - 1)
 506
 507 static struct list_head tcp_apps[TCP_APP_TAB_SIZE];
 508 static spinlock_t tcp_app_lock = SPIN_LOCK_UNLOCKED;
 509
 510 static inline __u16 tcp_app_hashkey(__u16 port)
 511 {
 512         return ((port >> TCP_APP_TAB_BITS) ^ port) & TCP_APP_TAB_MASK;
 513 }
 514
 515
 516 static int tcp_register_app(struct ip_vs_app *inc)
 517 {
 518         struct ip_vs_app *i;
 519         __u16 hash, port = inc->port;
 520         int ret = 0;
 521
 522         hash = tcp_app_hashkey(port);
 523
 524         spin_lock_bh(&tcp_app_lock);
 525         list_for_each_entry(i, &tcp_apps[hash], p_list) {
 526                 if (i->port == port) {
 527                         ret = -EEXIST;
 528                         goto out;
 529                 }
 530         }
 531         list_add(&inc->p_list, &tcp_apps[hash]);
 532         atomic_inc(&ip_vs_protocol_tcp.appcnt);
 533
 534   out:
 535         spin_unlock_bh(&tcp_app_lock);
 536         return ret;
 537 }
 538
 539
 540 static void
 541 tcp_unregister_app(struct ip_vs_app *inc)
 542 {
 543         spin_lock_bh(&tcp_app_lock);
 544         atomic_dec(&ip_vs_protocol_tcp.appcnt);
 545         list_del(&inc->p_list);
 546         spin_unlock_bh(&tcp_app_lock);
 547 }
 548
 549
 550 static int
 551 tcp_app_conn_bind(struct ip_vs_conn *cp)
 552 {
 553         int hash;
 554         struct ip_vs_app *inc;
 555         int result = 0;
 556
 557         /* Default binding: bind app only for NAT */
 558         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
 559                 return 0;
 560
 561         /* Lookup application incarnations and bind the right one */
 562         hash = tcp_app_hashkey(cp->vport);
 563
 564         spin_lock(&tcp_app_lock);
 565         list_for_each_entry(inc, &tcp_apps[hash], p_list) {
 566                 if (inc->port == cp->vport) {
 567                         if (unlikely(!ip_vs_app_inc_get(inc)))
 568                                 break;
 569                         spin_unlock(&tcp_app_lock);
 570
 571                         IP_VS_DBG(9, "%s: Binding conn %u.%u.%u.%u:%u->"
 572                                   "%u.%u.%u.%u:%u to app %s on port %u\n",
 573                                   __FUNCTION__,
 574                                   NIPQUAD(cp->caddr), ntohs(cp->cport),
 575                                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
 576                                   inc->name, ntohs(inc->port));
 577                         cp->app = inc;
 578                         if (inc->init_conn)
 579                                 result = inc->init_conn(inc, cp);
 580                         goto out;
 581                 }
 582         }
 583         spin_unlock(&tcp_app_lock);
 584
 585   out:
 586         return result;
 587 }
 588
 589
 590 /*
 591  *      Set LISTEN timeout. (ip_vs_conn_put will setup timer)
 592  */
 593 void ip_vs_tcp_conn_listen(struct ip_vs_conn *cp)
 594 {
 595         spin_lock(&cp->lock);
 596         cp->state = IP_VS_TCP_S_LISTEN;
 597         cp->timeout = ip_vs_protocol_tcp.timeout_table[IP_VS_TCP_S_LISTEN];
 598         spin_unlock(&cp->lock);
 599 }
 600
 601
 602 static void tcp_init(struct ip_vs_protocol *pp)
 603 {
 604         IP_VS_INIT_HASH_TABLE(tcp_apps);
 605         pp->timeout_table = tcp_timeouts;
 606 }
 607
 608
 609 static void tcp_exit(struct ip_vs_protocol *pp)
 610 {
 611 }
 612
 613
 614 struct ip_vs_protocol ip_vs_protocol_tcp = {
 615         .name =                 "TCP",
 616         .protocol =             IPPROTO_TCP,
 617         .dont_defrag =          0,
 618         .appcnt =               ATOMIC_INIT(0),
 619         .init =                 tcp_init,
 620         .exit =                 tcp_exit,
 621         .register_app =         tcp_register_app,
 622         .unregister_app =       tcp_unregister_app,
 623         .conn_schedule =        tcp_conn_schedule,
 624         .conn_in_get =          tcp_conn_in_get,
 625         .conn_out_get =         tcp_conn_out_get,
 626         .snat_handler =         tcp_snat_handler,
 627         .dnat_handler =         tcp_dnat_handler,
 628         .csum_check =           tcp_csum_check,
 629         .state_name =           tcp_state_name,
 630         .state_transition =     tcp_state_transition,
 631         .app_conn_bind =        tcp_app_conn_bind,
 632         .debug_packet =         ip_vs_tcpudp_debug_packet,
 633         .timeout_change =       tcp_timeout_change,
 634         .set_state_timeout =    tcp_set_state_timeout,
 635 };