git.oblomov.eu Git - linux-2.6/blob - 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/kernel.h>
  19 #include <linux/ip.h>
  20 #include <linux/tcp.h>                  /* for tcphdr */
  21 #include <net/ip.h>
  22 #include <net/tcp.h>                    /* for csum_tcpudp_magic */
  23 #include <linux/netfilter_ipv4.h>
  24
  25 #include <net/ip_vs.h>
  26
  27
  28 static struct ip_vs_conn *
  29 tcp_conn_in_get(const struct sk_buff *skb, struct ip_vs_protocol *pp,
  30                 const struct iphdr *iph, unsigned int proto_off, int inverse)
  31 {
  32         __be16 _ports[2], *pptr;
  33
  34         pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports);
  35         if (pptr == NULL)
  36                 return NULL;
  37
  38         if (likely(!inverse)) {
  39                 return ip_vs_conn_in_get(iph->protocol,
  40                                          iph->saddr, pptr[0],
  41                                          iph->daddr, pptr[1]);
  42         } else {
  43                 return ip_vs_conn_in_get(iph->protocol,
  44                                          iph->daddr, pptr[1],
  45                                          iph->saddr, pptr[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         __be16 _ports[2], *pptr;
  54
  55         pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports);
  56         if (pptr == NULL)
  57                 return NULL;
  58
  59         if (likely(!inverse)) {
  60                 return ip_vs_conn_out_get(iph->protocol,
  61                                           iph->saddr, pptr[0],
  62                                           iph->daddr, pptr[1]);
  63         } else {
  64                 return ip_vs_conn_out_get(iph->protocol,
  65                                           iph->daddr, pptr[1],
  66                                           iph->saddr, pptr[0]);
  67         }
  68 }
  69
  70
  71 static int
  72 tcp_conn_schedule(struct sk_buff *skb,
  73                   struct ip_vs_protocol *pp,
  74                   int *verdict, struct ip_vs_conn **cpp)
  75 {
  76         struct ip_vs_service *svc;
  77         struct tcphdr _tcph, *th;
  78
  79         th = skb_header_pointer(skb, skb->nh.iph->ihl*4,
  80                                 sizeof(_tcph), &_tcph);
  81         if (th == NULL) {
  82                 *verdict = NF_DROP;
  83                 return 0;
  84         }
  85
  86         if (th->syn &&
  87             (svc = ip_vs_service_get(skb->nfmark, skb->nh.iph->protocol,
  88                                      skb->nh.iph->daddr, th->dest))) {
  89                 if (ip_vs_todrop()) {
  90                         /*
  91                          * It seems that we are very loaded.
  92                          * We have to drop this packet :(
  93                          */
  94                         ip_vs_service_put(svc);
  95                         *verdict = NF_DROP;
  96                         return 0;
  97                 }
  98
  99                 /*
 100                  * Let the virtual server select a real server for the
 101                  * incoming connection, and create a connection entry.
 102                  */
 103                 *cpp = ip_vs_schedule(svc, skb);
 104                 if (!*cpp) {
 105                         *verdict = ip_vs_leave(svc, skb, pp);
 106                         return 0;
 107                 }
 108                 ip_vs_service_put(svc);
 109         }
 110         return 1;
 111 }
 112
 113
 114 static inline void
 115 tcp_fast_csum_update(struct tcphdr *tcph, __be32 oldip, __be32 newip,
 116                      __be16 oldport, __be16 newport)
 117 {
 118         tcph->check =
 119                 ip_vs_check_diff(~oldip, newip,
 120                                  ip_vs_check_diff(oldport ^ htons(0xFFFF),
 121                                                   newport, tcph->check));
 122 }
 123
 124
 125 static int
 126 tcp_snat_handler(struct sk_buff **pskb,
 127                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
 128 {
 129         struct tcphdr *tcph;
 130         unsigned int tcphoff = (*pskb)->nh.iph->ihl * 4;
 131
 132         /* csum_check requires unshared skb */
 133         if (!ip_vs_make_skb_writable(pskb, tcphoff+sizeof(*tcph)))
 134                 return 0;
 135
 136         if (unlikely(cp->app != NULL)) {
 137                 /* Some checks before mangling */
 138                 if (pp->csum_check && !pp->csum_check(*pskb, pp))
 139                         return 0;
 140
 141                 /* Call application helper if needed */
 142                 if (!ip_vs_app_pkt_out(cp, pskb))
 143                         return 0;
 144         }
 145
 146         tcph = (void *)(*pskb)->nh.iph + tcphoff;
 147         tcph->source = cp->vport;
 148
 149         /* Adjust TCP checksums */
 150         if (!cp->app) {
 151                 /* Only port and addr are changed, do fast csum update */
 152                 tcp_fast_csum_update(tcph, cp->daddr, cp->vaddr,
 153                                      cp->dport, cp->vport);
 154                 if ((*pskb)->ip_summed == CHECKSUM_COMPLETE)
 155                         (*pskb)->ip_summed = CHECKSUM_NONE;
 156         } else {
 157                 /* full checksum calculation */
 158                 tcph->check = 0;
 159                 (*pskb)->csum = skb_checksum(*pskb, tcphoff,
 160                                              (*pskb)->len - tcphoff, 0);
 161                 tcph->check = csum_tcpudp_magic(cp->vaddr, cp->caddr,
 162                                                 (*pskb)->len - tcphoff,
 163                                                 cp->protocol,
 164                                                 (*pskb)->csum);
 165                 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n",
 166                           pp->name, tcph->check,
 167                           (char*)&(tcph->check) - (char*)tcph);
 168         }
 169         return 1;
 170 }
 171
 172
 173 static int
 174 tcp_dnat_handler(struct sk_buff **pskb,
 175                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
 176 {
 177         struct tcphdr *tcph;
 178         unsigned int tcphoff = (*pskb)->nh.iph->ihl * 4;
 179
 180         /* csum_check requires unshared skb */
 181         if (!ip_vs_make_skb_writable(pskb, tcphoff+sizeof(*tcph)))
 182                 return 0;
 183
 184         if (unlikely(cp->app != NULL)) {
 185                 /* Some checks before mangling */
 186                 if (pp->csum_check && !pp->csum_check(*pskb, pp))
 187                         return 0;
 188
 189                 /*
 190                  *      Attempt ip_vs_app call.
 191                  *      It will fix ip_vs_conn and iph ack_seq stuff
 192                  */
 193                 if (!ip_vs_app_pkt_in(cp, pskb))
 194                         return 0;
 195         }
 196
 197         tcph = (void *)(*pskb)->nh.iph + tcphoff;
 198         tcph->dest = cp->dport;
 199
 200         /*
 201          *      Adjust TCP checksums
 202          */
 203         if (!cp->app) {
 204                 /* Only port and addr are changed, do fast csum update */
 205                 tcp_fast_csum_update(tcph, cp->vaddr, cp->daddr,
 206                                      cp->vport, cp->dport);
 207                 if ((*pskb)->ip_summed == CHECKSUM_COMPLETE)
 208                         (*pskb)->ip_summed = CHECKSUM_NONE;
 209         } else {
 210                 /* full checksum calculation */
 211                 tcph->check = 0;
 212                 (*pskb)->csum = skb_checksum(*pskb, tcphoff,
 213                                              (*pskb)->len - tcphoff, 0);
 214                 tcph->check = csum_tcpudp_magic(cp->caddr, cp->daddr,
 215                                                 (*pskb)->len - tcphoff,
 216                                                 cp->protocol,
 217                                                 (*pskb)->csum);
 218                 (*pskb)->ip_summed = CHECKSUM_UNNECESSARY;
 219         }
 220         return 1;
 221 }
 222
 223
 224 static int
 225 tcp_csum_check(struct sk_buff *skb, struct ip_vs_protocol *pp)
 226 {
 227         unsigned int tcphoff = skb->nh.iph->ihl*4;
 228
 229         switch (skb->ip_summed) {
 230         case CHECKSUM_NONE:
 231                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 232         case CHECKSUM_COMPLETE:
 233                 if (csum_tcpudp_magic(skb->nh.iph->saddr, skb->nh.iph->daddr,
 234                                       skb->len - tcphoff,
 235                                       skb->nh.iph->protocol, skb->csum)) {
 236                         IP_VS_DBG_RL_PKT(0, pp, skb, 0,
 237                                          "Failed checksum for");
 238                         return 0;
 239                 }
 240                 break;
 241         default:
 242                 /* No need to checksum. */
 243                 break;
 244         }
 245
 246         return 1;
 247 }
 248
 249
 250 #define TCP_DIR_INPUT           0
 251 #define TCP_DIR_OUTPUT          4
 252 #define TCP_DIR_INPUT_ONLY      8
 253
 254 static const int tcp_state_off[IP_VS_DIR_LAST] = {
 255         [IP_VS_DIR_INPUT]               =       TCP_DIR_INPUT,
 256         [IP_VS_DIR_OUTPUT]              =       TCP_DIR_OUTPUT,
 257         [IP_VS_DIR_INPUT_ONLY]          =       TCP_DIR_INPUT_ONLY,
 258 };
 259
 260 /*
 261  *      Timeout table[state]
 262  */
 263 static int tcp_timeouts[IP_VS_TCP_S_LAST+1] = {
 264         [IP_VS_TCP_S_NONE]              =       2*HZ,
 265         [IP_VS_TCP_S_ESTABLISHED]       =       15*60*HZ,
 266         [IP_VS_TCP_S_SYN_SENT]          =       2*60*HZ,
 267         [IP_VS_TCP_S_SYN_RECV]          =       1*60*HZ,
 268         [IP_VS_TCP_S_FIN_WAIT]          =       2*60*HZ,
 269         [IP_VS_TCP_S_TIME_WAIT]         =       2*60*HZ,
 270         [IP_VS_TCP_S_CLOSE]             =       10*HZ,
 271         [IP_VS_TCP_S_CLOSE_WAIT]        =       60*HZ,
 272         [IP_VS_TCP_S_LAST_ACK]          =       30*HZ,
 273         [IP_VS_TCP_S_LISTEN]            =       2*60*HZ,
 274         [IP_VS_TCP_S_SYNACK]            =       120*HZ,
 275         [IP_VS_TCP_S_LAST]              =       2*HZ,
 276 };
 277
 278 static char * tcp_state_name_table[IP_VS_TCP_S_LAST+1] = {
 279         [IP_VS_TCP_S_NONE]              =       "NONE",
 280         [IP_VS_TCP_S_ESTABLISHED]       =       "ESTABLISHED",
 281         [IP_VS_TCP_S_SYN_SENT]          =       "SYN_SENT",
 282         [IP_VS_TCP_S_SYN_RECV]          =       "SYN_RECV",
 283         [IP_VS_TCP_S_FIN_WAIT]          =       "FIN_WAIT",
 284         [IP_VS_TCP_S_TIME_WAIT]         =       "TIME_WAIT",
 285         [IP_VS_TCP_S_CLOSE]             =       "CLOSE",
 286         [IP_VS_TCP_S_CLOSE_WAIT]        =       "CLOSE_WAIT",
 287         [IP_VS_TCP_S_LAST_ACK]          =       "LAST_ACK",
 288         [IP_VS_TCP_S_LISTEN]            =       "LISTEN",
 289         [IP_VS_TCP_S_SYNACK]            =       "SYNACK",
 290         [IP_VS_TCP_S_LAST]              =       "BUG!",
 291 };
 292
 293 #define sNO IP_VS_TCP_S_NONE
 294 #define sES IP_VS_TCP_S_ESTABLISHED
 295 #define sSS IP_VS_TCP_S_SYN_SENT
 296 #define sSR IP_VS_TCP_S_SYN_RECV
 297 #define sFW IP_VS_TCP_S_FIN_WAIT
 298 #define sTW IP_VS_TCP_S_TIME_WAIT
 299 #define sCL IP_VS_TCP_S_CLOSE
 300 #define sCW IP_VS_TCP_S_CLOSE_WAIT
 301 #define sLA IP_VS_TCP_S_LAST_ACK
 302 #define sLI IP_VS_TCP_S_LISTEN
 303 #define sSA IP_VS_TCP_S_SYNACK
 304
 305 struct tcp_states_t {
 306         int next_state[IP_VS_TCP_S_LAST];
 307 };
 308
 309 static const char * tcp_state_name(int state)
 310 {
 311         if (state >= IP_VS_TCP_S_LAST)
 312                 return "ERR!";
 313         return tcp_state_name_table[state] ? tcp_state_name_table[state] : "?";
 314 }
 315
 316 static struct tcp_states_t tcp_states [] = {
 317 /*      INPUT */
 318 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 319 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }},
 320 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sTW }},
 321 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 322 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sSR }},
 323
 324 /*      OUTPUT */
 325 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 326 /*syn*/ {{sSS, sES, sSS, sSR, sSS, sSS, sSS, sSS, sSS, sLI, sSR }},
 327 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }},
 328 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }},
 329 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }},
 330
 331 /*      INPUT-ONLY */
 332 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 333 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }},
 334 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }},
 335 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 336 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 337 };
 338
 339 static struct tcp_states_t tcp_states_dos [] = {
 340 /*      INPUT */
 341 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 342 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSA }},
 343 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sSA }},
 344 /*ack*/ {{sCL, sES, sSS, sSR, sFW, sTW, sCL, sCW, sCL, sLI, sSA }},
 345 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 346
 347 /*      OUTPUT */
 348 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 349 /*syn*/ {{sSS, sES, sSS, sSA, sSS, sSS, sSS, sSS, sSS, sLI, sSA }},
 350 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }},
 351 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }},
 352 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }},
 353
 354 /*      INPUT-ONLY */
 355 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 356 /*syn*/ {{sSA, sES, sES, sSR, sSA, sSA, sSA, sSA, sSA, sSA, sSA }},
 357 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }},
 358 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 359 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 360 };
 361
 362 static struct tcp_states_t *tcp_state_table = tcp_states;
 363
 364
 365 static void tcp_timeout_change(struct ip_vs_protocol *pp, int flags)
 366 {
 367         int on = (flags & 1);           /* secure_tcp */
 368
 369         /*
 370         ** FIXME: change secure_tcp to independent sysctl var
 371         ** or make it per-service or per-app because it is valid
 372         ** for most if not for all of the applications. Something
 373         ** like "capabilities" (flags) for each object.
 374         */
 375         tcp_state_table = (on? tcp_states_dos : tcp_states);
 376 }
 377
 378 static int
 379 tcp_set_state_timeout(struct ip_vs_protocol *pp, char *sname, int to)
 380 {
 381         return ip_vs_set_state_timeout(pp->timeout_table, IP_VS_TCP_S_LAST,
 382                                        tcp_state_name_table, sname, to);
 383 }
 384
 385 static inline int tcp_state_idx(struct tcphdr *th)
 386 {
 387         if (th->rst)
 388                 return 3;
 389         if (th->syn)
 390                 return 0;
 391         if (th->fin)
 392                 return 1;
 393         if (th->ack)
 394                 return 2;
 395         return -1;
 396 }
 397
 398 static inline void
 399 set_tcp_state(struct ip_vs_protocol *pp, struct ip_vs_conn *cp,
 400               int direction, struct tcphdr *th)
 401 {
 402         int state_idx;
 403         int new_state = IP_VS_TCP_S_CLOSE;
 404         int state_off = tcp_state_off[direction];
 405
 406         /*
 407          *    Update state offset to INPUT_ONLY if necessary
 408          *    or delete NO_OUTPUT flag if output packet detected
 409          */
 410         if (cp->flags & IP_VS_CONN_F_NOOUTPUT) {
 411                 if (state_off == TCP_DIR_OUTPUT)
 412                         cp->flags &= ~IP_VS_CONN_F_NOOUTPUT;
 413                 else
 414                         state_off = TCP_DIR_INPUT_ONLY;
 415         }
 416
 417         if ((state_idx = tcp_state_idx(th)) < 0) {
 418                 IP_VS_DBG(8, "tcp_state_idx=%d!!!\n", state_idx);
 419                 goto tcp_state_out;
 420         }
 421
 422         new_state = tcp_state_table[state_off+state_idx].next_state[cp->state];
 423
 424   tcp_state_out:
 425         if (new_state != cp->state) {
 426                 struct ip_vs_dest *dest = cp->dest;
 427
 428                 IP_VS_DBG(8, "%s %s [%c%c%c%c] %u.%u.%u.%u:%d->"
 429                           "%u.%u.%u.%u:%d state: %s->%s conn->refcnt:%d\n",
 430                           pp->name,
 431                           (state_off==TCP_DIR_OUTPUT)?"output ":"input ",
 432                           th->syn? 'S' : '.',
 433                           th->fin? 'F' : '.',
 434                           th->ack? 'A' : '.',
 435                           th->rst? 'R' : '.',
 436                           NIPQUAD(cp->daddr), ntohs(cp->dport),
 437                           NIPQUAD(cp->caddr), ntohs(cp->cport),
 438                           tcp_state_name(cp->state),
 439                           tcp_state_name(new_state),
 440                           atomic_read(&cp->refcnt));
 441                 if (dest) {
 442                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 443                             (new_state != IP_VS_TCP_S_ESTABLISHED)) {
 444                                 atomic_dec(&dest->activeconns);
 445                                 atomic_inc(&dest->inactconns);
 446                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
 447                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
 448                                    (new_state == IP_VS_TCP_S_ESTABLISHED)) {
 449                                 atomic_inc(&dest->activeconns);
 450                                 atomic_dec(&dest->inactconns);
 451                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 452                         }
 453                 }
 454         }
 455
 456         cp->timeout = pp->timeout_table[cp->state = new_state];
 457 }
 458
 459
 460 /*
 461  *      Handle state transitions
 462  */
 463 static int
 464 tcp_state_transition(struct ip_vs_conn *cp, int direction,
 465                      const struct sk_buff *skb,
 466                      struct ip_vs_protocol *pp)
 467 {
 468         struct tcphdr _tcph, *th;
 469
 470         th = skb_header_pointer(skb, skb->nh.iph->ihl*4,
 471                                 sizeof(_tcph), &_tcph);
 472         if (th == NULL)
 473                 return 0;
 474
 475         spin_lock(&cp->lock);
 476         set_tcp_state(pp, cp, direction, th);
 477         spin_unlock(&cp->lock);
 478
 479         return 1;
 480 }
 481
 482
 483 /*
 484  *      Hash table for TCP application incarnations
 485  */
 486 #define TCP_APP_TAB_BITS        4
 487 #define TCP_APP_TAB_SIZE        (1 << TCP_APP_TAB_BITS)
 488 #define TCP_APP_TAB_MASK        (TCP_APP_TAB_SIZE - 1)
 489
 490 static struct list_head tcp_apps[TCP_APP_TAB_SIZE];
 491 static DEFINE_SPINLOCK(tcp_app_lock);
 492
 493 static inline __u16 tcp_app_hashkey(__u16 port)
 494 {
 495         return ((port >> TCP_APP_TAB_BITS) ^ port) & TCP_APP_TAB_MASK;
 496 }
 497
 498
 499 static int tcp_register_app(struct ip_vs_app *inc)
 500 {
 501         struct ip_vs_app *i;
 502         __u16 hash, port = inc->port;
 503         int ret = 0;
 504
 505         hash = tcp_app_hashkey(port);
 506
 507         spin_lock_bh(&tcp_app_lock);
 508         list_for_each_entry(i, &tcp_apps[hash], p_list) {
 509                 if (i->port == port) {
 510                         ret = -EEXIST;
 511                         goto out;
 512                 }
 513         }
 514         list_add(&inc->p_list, &tcp_apps[hash]);
 515         atomic_inc(&ip_vs_protocol_tcp.appcnt);
 516
 517   out:
 518         spin_unlock_bh(&tcp_app_lock);
 519         return ret;
 520 }
 521
 522
 523 static void
 524 tcp_unregister_app(struct ip_vs_app *inc)
 525 {
 526         spin_lock_bh(&tcp_app_lock);
 527         atomic_dec(&ip_vs_protocol_tcp.appcnt);
 528         list_del(&inc->p_list);
 529         spin_unlock_bh(&tcp_app_lock);
 530 }
 531
 532
 533 static int
 534 tcp_app_conn_bind(struct ip_vs_conn *cp)
 535 {
 536         int hash;
 537         struct ip_vs_app *inc;
 538         int result = 0;
 539
 540         /* Default binding: bind app only for NAT */
 541         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
 542                 return 0;
 543
 544         /* Lookup application incarnations and bind the right one */
 545         hash = tcp_app_hashkey(cp->vport);
 546
 547         spin_lock(&tcp_app_lock);
 548         list_for_each_entry(inc, &tcp_apps[hash], p_list) {
 549                 if (inc->port == cp->vport) {
 550                         if (unlikely(!ip_vs_app_inc_get(inc)))
 551                                 break;
 552                         spin_unlock(&tcp_app_lock);
 553
 554                         IP_VS_DBG(9, "%s: Binding conn %u.%u.%u.%u:%u->"
 555                                   "%u.%u.%u.%u:%u to app %s on port %u\n",
 556                                   __FUNCTION__,
 557                                   NIPQUAD(cp->caddr), ntohs(cp->cport),
 558                                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
 559                                   inc->name, ntohs(inc->port));
 560                         cp->app = inc;
 561                         if (inc->init_conn)
 562                                 result = inc->init_conn(inc, cp);
 563                         goto out;
 564                 }
 565         }
 566         spin_unlock(&tcp_app_lock);
 567
 568   out:
 569         return result;
 570 }
 571
 572
 573 /*
 574  *      Set LISTEN timeout. (ip_vs_conn_put will setup timer)
 575  */
 576 void ip_vs_tcp_conn_listen(struct ip_vs_conn *cp)
 577 {
 578         spin_lock(&cp->lock);
 579         cp->state = IP_VS_TCP_S_LISTEN;
 580         cp->timeout = ip_vs_protocol_tcp.timeout_table[IP_VS_TCP_S_LISTEN];
 581         spin_unlock(&cp->lock);
 582 }
 583
 584
 585 static void ip_vs_tcp_init(struct ip_vs_protocol *pp)
 586 {
 587         IP_VS_INIT_HASH_TABLE(tcp_apps);
 588         pp->timeout_table = tcp_timeouts;
 589 }
 590
 591
 592 static void ip_vs_tcp_exit(struct ip_vs_protocol *pp)
 593 {
 594 }
 595
 596
 597 struct ip_vs_protocol ip_vs_protocol_tcp = {
 598         .name =                 "TCP",
 599         .protocol =             IPPROTO_TCP,
 600         .dont_defrag =          0,
 601         .appcnt =               ATOMIC_INIT(0),
 602         .init =                 ip_vs_tcp_init,
 603         .exit =                 ip_vs_tcp_exit,
 604         .register_app =         tcp_register_app,
 605         .unregister_app =       tcp_unregister_app,
 606         .conn_schedule =        tcp_conn_schedule,
 607         .conn_in_get =          tcp_conn_in_get,
 608         .conn_out_get =         tcp_conn_out_get,
 609         .snat_handler =         tcp_snat_handler,
 610         .dnat_handler =         tcp_dnat_handler,
 611         .csum_check =           tcp_csum_check,
 612         .state_name =           tcp_state_name,
 613         .state_transition =     tcp_state_transition,
 614         .app_conn_bind =        tcp_app_conn_bind,
 615         .debug_packet =         ip_vs_tcpudp_debug_packet,
 616         .timeout_change =       tcp_timeout_change,
 617         .set_state_timeout =    tcp_set_state_timeout,
 618 };