net/ipv4/ipvs/ip_vs_proto_tcp.c

   1 /*
   2  * ip_vs_proto_tcp.c:   TCP load balancing support for IPVS
   3  *
   4  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   5  *              Julian Anastasov <ja@ssi.bg>
   6  *
   7  *              This program is free software; you can redistribute it and/or
   8  *              modify it under the terms of the GNU General Public License
   9  *              as published by the Free Software Foundation; either version
  10  *              2 of the License, or (at your option) any later version.
  11  *
  12  * Changes:
  13  *
  14  */
  15
  16 #include <linux/kernel.h>
  17 #include <linux/ip.h>
  18 #include <linux/tcp.h>                  /* for tcphdr */
  19 #include <net/ip.h>
  20 #include <net/tcp.h>                    /* for csum_tcpudp_magic */
  21 #include <linux/netfilter.h>
  22 #include <linux/netfilter_ipv4.h>
  23
  24 #include <net/ip_vs.h>
  25
  26
  27 static struct ip_vs_conn *
  28 tcp_conn_in_get(int af, const struct sk_buff *skb, struct ip_vs_protocol *pp,
  29                 const struct ip_vs_iphdr *iph, unsigned int proto_off,
  30                 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(af, iph->protocol,
  40                                          &iph->saddr, pptr[0],
  41                                          &iph->daddr, pptr[1]);
  42         } else {
  43                 return ip_vs_conn_in_get(af, 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(int af, const struct sk_buff *skb, struct ip_vs_protocol *pp,
  51                  const struct ip_vs_iphdr *iph, unsigned int proto_off,
  52                  int inverse)
  53 {
  54         __be16 _ports[2], *pptr;
  55
  56         pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports);
  57         if (pptr == NULL)
  58                 return NULL;
  59
  60         if (likely(!inverse)) {
  61                 return ip_vs_conn_out_get(af, iph->protocol,
  62                                           &iph->saddr, pptr[0],
  63                                           &iph->daddr, pptr[1]);
  64         } else {
  65                 return ip_vs_conn_out_get(af, iph->protocol,
  66                                           &iph->daddr, pptr[1],
  67                                           &iph->saddr, pptr[0]);
  68         }
  69 }
  70
  71
  72 static int
  73 tcp_conn_schedule(int af, struct sk_buff *skb, 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         struct ip_vs_iphdr iph;
  79
  80         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
  81
  82         th = skb_header_pointer(skb, iph.len, sizeof(_tcph), &_tcph);
  83         if (th == NULL) {
  84                 *verdict = NF_DROP;
  85                 return 0;
  86         }
  87
  88         if (th->syn &&
  89             (svc = ip_vs_service_get(af, skb->mark, iph.protocol, &iph.daddr,
  90                                      th->dest))) {
  91                 if (ip_vs_todrop()) {
  92                         /*
  93                          * It seems that we are very loaded.
  94                          * We have to drop this packet :(
  95                          */
  96                         ip_vs_service_put(svc);
  97                         *verdict = NF_DROP;
  98                         return 0;
  99                 }
 100
 101                 /*
 102                  * Let the virtual server select a real server for the
 103                  * incoming connection, and create a connection entry.
 104                  */
 105                 *cpp = ip_vs_schedule(svc, skb);
 106                 if (!*cpp) {
 107                         *verdict = ip_vs_leave(svc, skb, pp);
 108                         return 0;
 109                 }
 110                 ip_vs_service_put(svc);
 111         }
 112         return 1;
 113 }
 114
 115
 116 static inline void
 117 tcp_fast_csum_update(int af, struct tcphdr *tcph,
 118                      const union nf_inet_addr *oldip,
 119                      const union nf_inet_addr *newip,
 120                      __be16 oldport, __be16 newport)
 121 {
 122 #ifdef CONFIG_IP_VS_IPV6
 123         if (af == AF_INET6)
 124                 tcph->check =
 125                         csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
 126                                          ip_vs_check_diff2(oldport, newport,
 127                                                 ~csum_unfold(tcph->check))));
 128         else
 129 #endif
 130         tcph->check =
 131                 csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
 132                                  ip_vs_check_diff2(oldport, newport,
 133                                                 ~csum_unfold(tcph->check))));
 134 }
 135
 136
 137 static inline void
 138 tcp_partial_csum_update(int af, struct tcphdr *tcph,
 139                      const union nf_inet_addr *oldip,
 140                      const union nf_inet_addr *newip,
 141                      __be16 oldlen, __be16 newlen)
 142 {
 143 #ifdef CONFIG_IP_VS_IPV6
 144         if (af == AF_INET6)
 145                 tcph->check =
 146                         csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
 147                                          ip_vs_check_diff2(oldlen, newlen,
 148                                                 ~csum_unfold(tcph->check))));
 149         else
 150 #endif
 151         tcph->check =
 152                 csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
 153                                 ip_vs_check_diff2(oldlen, newlen,
 154                                                 ~csum_unfold(tcph->check))));
 155 }
 156
 157
 158 static int
 159 tcp_snat_handler(struct sk_buff *skb,
 160                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
 161 {
 162         struct tcphdr *tcph;
 163         unsigned int tcphoff;
 164         int oldlen;
 165
 166 #ifdef CONFIG_IP_VS_IPV6
 167         if (cp->af == AF_INET6)
 168                 tcphoff = sizeof(struct ipv6hdr);
 169         else
 170 #endif
 171                 tcphoff = ip_hdrlen(skb);
 172         oldlen = skb->len - tcphoff;
 173
 174         /* csum_check requires unshared skb */
 175         if (!skb_make_writable(skb, tcphoff+sizeof(*tcph)))
 176                 return 0;
 177
 178         if (unlikely(cp->app != NULL)) {
 179                 /* Some checks before mangling */
 180                 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
 181                         return 0;
 182
 183                 /* Call application helper if needed */
 184                 if (!ip_vs_app_pkt_out(cp, skb))
 185                         return 0;
 186         }
 187
 188         tcph = (void *)skb_network_header(skb) + tcphoff;
 189         tcph->source = cp->vport;
 190
 191         /* Adjust TCP checksums */
 192         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 193                 tcp_partial_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr,
 194                                         htonl(oldlen),
 195                                         htonl(skb->len - tcphoff));
 196         } else if (!cp->app) {
 197                 /* Only port and addr are changed, do fast csum update */
 198                 tcp_fast_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr,
 199                                      cp->dport, cp->vport);
 200                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 201                         skb->ip_summed = CHECKSUM_NONE;
 202         } else {
 203                 /* full checksum calculation */
 204                 tcph->check = 0;
 205                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 206 #ifdef CONFIG_IP_VS_IPV6
 207                 if (cp->af == AF_INET6)
 208                         tcph->check = csum_ipv6_magic(&cp->vaddr.in6,
 209                                                       &cp->caddr.in6,
 210                                                       skb->len - tcphoff,
 211                                                       cp->protocol, skb->csum);
 212                 else
 213 #endif
 214                         tcph->check = csum_tcpudp_magic(cp->vaddr.ip,
 215                                                         cp->caddr.ip,
 216                                                         skb->len - tcphoff,
 217                                                         cp->protocol,
 218                                                         skb->csum);
 219
 220                 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n",
 221                           pp->name, tcph->check,
 222                           (char*)&(tcph->check) - (char*)tcph);
 223         }
 224         return 1;
 225 }
 226
 227
 228 static int
 229 tcp_dnat_handler(struct sk_buff *skb,
 230                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
 231 {
 232         struct tcphdr *tcph;
 233         unsigned int tcphoff;
 234         int oldlen;
 235
 236 #ifdef CONFIG_IP_VS_IPV6
 237         if (cp->af == AF_INET6)
 238                 tcphoff = sizeof(struct ipv6hdr);
 239         else
 240 #endif
 241                 tcphoff = ip_hdrlen(skb);
 242         oldlen = skb->len - tcphoff;
 243
 244         /* csum_check requires unshared skb */
 245         if (!skb_make_writable(skb, tcphoff+sizeof(*tcph)))
 246                 return 0;
 247
 248         if (unlikely(cp->app != NULL)) {
 249                 /* Some checks before mangling */
 250                 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
 251                         return 0;
 252
 253                 /*
 254                  *      Attempt ip_vs_app call.
 255                  *      It will fix ip_vs_conn and iph ack_seq stuff
 256                  */
 257                 if (!ip_vs_app_pkt_in(cp, skb))
 258                         return 0;
 259         }
 260
 261         tcph = (void *)skb_network_header(skb) + tcphoff;
 262         tcph->dest = cp->dport;
 263
 264         /*
 265          *      Adjust TCP checksums
 266          */
 267         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 268                 tcp_partial_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr,
 269                                         htonl(oldlen),
 270                                         htonl(skb->len - tcphoff));
 271         } else if (!cp->app) {
 272                 /* Only port and addr are changed, do fast csum update */
 273                 tcp_fast_csum_update(cp->af, tcph, &cp->vaddr, &cp->daddr,
 274                                      cp->vport, cp->dport);
 275                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 276                         skb->ip_summed = CHECKSUM_NONE;
 277         } else {
 278                 /* full checksum calculation */
 279                 tcph->check = 0;
 280                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 281 #ifdef CONFIG_IP_VS_IPV6
 282                 if (cp->af == AF_INET6)
 283                         tcph->check = csum_ipv6_magic(&cp->caddr.in6,
 284                                                       &cp->daddr.in6,
 285                                                       skb->len - tcphoff,
 286                                                       cp->protocol, skb->csum);
 287                 else
 288 #endif
 289                         tcph->check = csum_tcpudp_magic(cp->caddr.ip,
 290                                                         cp->daddr.ip,
 291                                                         skb->len - tcphoff,
 292                                                         cp->protocol,
 293                                                         skb->csum);
 294                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 295         }
 296         return 1;
 297 }
 298
 299
 300 static int
 301 tcp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp)
 302 {
 303         unsigned int tcphoff;
 304
 305 #ifdef CONFIG_IP_VS_IPV6
 306         if (af == AF_INET6)
 307                 tcphoff = sizeof(struct ipv6hdr);
 308         else
 309 #endif
 310                 tcphoff = ip_hdrlen(skb);
 311
 312         switch (skb->ip_summed) {
 313         case CHECKSUM_NONE:
 314                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 315         case CHECKSUM_COMPLETE:
 316 #ifdef CONFIG_IP_VS_IPV6
 317                 if (af == AF_INET6) {
 318                         if (csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 319                                             &ipv6_hdr(skb)->daddr,
 320                                             skb->len - tcphoff,
 321                                             ipv6_hdr(skb)->nexthdr,
 322                                             skb->csum)) {
 323                                 IP_VS_DBG_RL_PKT(0, pp, skb, 0,
 324                                                  "Failed checksum for");
 325                                 return 0;
 326                         }
 327                 } else
 328 #endif
 329                         if (csum_tcpudp_magic(ip_hdr(skb)->saddr,
 330                                               ip_hdr(skb)->daddr,
 331                                               skb->len - tcphoff,
 332                                               ip_hdr(skb)->protocol,
 333                                               skb->csum)) {
 334                                 IP_VS_DBG_RL_PKT(0, pp, skb, 0,
 335                                                  "Failed checksum for");
 336                                 return 0;
 337                         }
 338                 break;
 339         default:
 340                 /* No need to checksum. */
 341                 break;
 342         }
 343
 344         return 1;
 345 }
 346
 347
 348 #define TCP_DIR_INPUT           0
 349 #define TCP_DIR_OUTPUT          4
 350 #define TCP_DIR_INPUT_ONLY      8
 351
 352 static const int tcp_state_off[IP_VS_DIR_LAST] = {
 353         [IP_VS_DIR_INPUT]               =       TCP_DIR_INPUT,
 354         [IP_VS_DIR_OUTPUT]              =       TCP_DIR_OUTPUT,
 355         [IP_VS_DIR_INPUT_ONLY]          =       TCP_DIR_INPUT_ONLY,
 356 };
 357
 358 /*
 359  *      Timeout table[state]
 360  */
 361 static int tcp_timeouts[IP_VS_TCP_S_LAST+1] = {
 362         [IP_VS_TCP_S_NONE]              =       2*HZ,
 363         [IP_VS_TCP_S_ESTABLISHED]       =       15*60*HZ,
 364         [IP_VS_TCP_S_SYN_SENT]          =       2*60*HZ,
 365         [IP_VS_TCP_S_SYN_RECV]          =       1*60*HZ,
 366         [IP_VS_TCP_S_FIN_WAIT]          =       2*60*HZ,
 367         [IP_VS_TCP_S_TIME_WAIT]         =       2*60*HZ,
 368         [IP_VS_TCP_S_CLOSE]             =       10*HZ,
 369         [IP_VS_TCP_S_CLOSE_WAIT]        =       60*HZ,
 370         [IP_VS_TCP_S_LAST_ACK]          =       30*HZ,
 371         [IP_VS_TCP_S_LISTEN]            =       2*60*HZ,
 372         [IP_VS_TCP_S_SYNACK]            =       120*HZ,
 373         [IP_VS_TCP_S_LAST]              =       2*HZ,
 374 };
 375
 376 static char * tcp_state_name_table[IP_VS_TCP_S_LAST+1] = {
 377         [IP_VS_TCP_S_NONE]              =       "NONE",
 378         [IP_VS_TCP_S_ESTABLISHED]       =       "ESTABLISHED",
 379         [IP_VS_TCP_S_SYN_SENT]          =       "SYN_SENT",
 380         [IP_VS_TCP_S_SYN_RECV]          =       "SYN_RECV",
 381         [IP_VS_TCP_S_FIN_WAIT]          =       "FIN_WAIT",
 382         [IP_VS_TCP_S_TIME_WAIT]         =       "TIME_WAIT",
 383         [IP_VS_TCP_S_CLOSE]             =       "CLOSE",
 384         [IP_VS_TCP_S_CLOSE_WAIT]        =       "CLOSE_WAIT",
 385         [IP_VS_TCP_S_LAST_ACK]          =       "LAST_ACK",
 386         [IP_VS_TCP_S_LISTEN]            =       "LISTEN",
 387         [IP_VS_TCP_S_SYNACK]            =       "SYNACK",
 388         [IP_VS_TCP_S_LAST]              =       "BUG!",
 389 };
 390
 391 #define sNO IP_VS_TCP_S_NONE
 392 #define sES IP_VS_TCP_S_ESTABLISHED
 393 #define sSS IP_VS_TCP_S_SYN_SENT
 394 #define sSR IP_VS_TCP_S_SYN_RECV
 395 #define sFW IP_VS_TCP_S_FIN_WAIT
 396 #define sTW IP_VS_TCP_S_TIME_WAIT
 397 #define sCL IP_VS_TCP_S_CLOSE
 398 #define sCW IP_VS_TCP_S_CLOSE_WAIT
 399 #define sLA IP_VS_TCP_S_LAST_ACK
 400 #define sLI IP_VS_TCP_S_LISTEN
 401 #define sSA IP_VS_TCP_S_SYNACK
 402
 403 struct tcp_states_t {
 404         int next_state[IP_VS_TCP_S_LAST];
 405 };
 406
 407 static const char * tcp_state_name(int state)
 408 {
 409         if (state >= IP_VS_TCP_S_LAST)
 410                 return "ERR!";
 411         return tcp_state_name_table[state] ? tcp_state_name_table[state] : "?";
 412 }
 413
 414 static struct tcp_states_t tcp_states [] = {
 415 /*      INPUT */
 416 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 417 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }},
 418 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sTW }},
 419 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 420 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sSR }},
 421
 422 /*      OUTPUT */
 423 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 424 /*syn*/ {{sSS, sES, sSS, sSR, sSS, sSS, sSS, sSS, sSS, sLI, sSR }},
 425 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }},
 426 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }},
 427 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }},
 428
 429 /*      INPUT-ONLY */
 430 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 431 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }},
 432 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }},
 433 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 434 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 435 };
 436
 437 static struct tcp_states_t tcp_states_dos [] = {
 438 /*      INPUT */
 439 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 440 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSA }},
 441 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sSA }},
 442 /*ack*/ {{sCL, sES, sSS, sSR, sFW, sTW, sCL, sCW, sCL, sLI, sSA }},
 443 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 444
 445 /*      OUTPUT */
 446 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 447 /*syn*/ {{sSS, sES, sSS, sSA, sSS, sSS, sSS, sSS, sSS, sLI, sSA }},
 448 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }},
 449 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }},
 450 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }},
 451
 452 /*      INPUT-ONLY */
 453 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 454 /*syn*/ {{sSA, sES, sES, sSR, sSA, sSA, sSA, sSA, sSA, sSA, sSA }},
 455 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }},
 456 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 457 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 458 };
 459
 460 static struct tcp_states_t *tcp_state_table = tcp_states;
 461
 462
 463 static void tcp_timeout_change(struct ip_vs_protocol *pp, int flags)
 464 {
 465         int on = (flags & 1);           /* secure_tcp */
 466
 467         /*
 468         ** FIXME: change secure_tcp to independent sysctl var
 469         ** or make it per-service or per-app because it is valid
 470         ** for most if not for all of the applications. Something
 471         ** like "capabilities" (flags) for each object.
 472         */
 473         tcp_state_table = (on? tcp_states_dos : tcp_states);
 474 }
 475
 476 static int
 477 tcp_set_state_timeout(struct ip_vs_protocol *pp, char *sname, int to)
 478 {
 479         return ip_vs_set_state_timeout(pp->timeout_table, IP_VS_TCP_S_LAST,
 480                                        tcp_state_name_table, sname, to);
 481 }
 482
 483 static inline int tcp_state_idx(struct tcphdr *th)
 484 {
 485         if (th->rst)
 486                 return 3;
 487         if (th->syn)
 488                 return 0;
 489         if (th->fin)
 490                 return 1;
 491         if (th->ack)
 492                 return 2;
 493         return -1;
 494 }
 495
 496 static inline void
 497 set_tcp_state(struct ip_vs_protocol *pp, struct ip_vs_conn *cp,
 498               int direction, struct tcphdr *th)
 499 {
 500         int state_idx;
 501         int new_state = IP_VS_TCP_S_CLOSE;
 502         int state_off = tcp_state_off[direction];
 503
 504         /*
 505          *    Update state offset to INPUT_ONLY if necessary
 506          *    or delete NO_OUTPUT flag if output packet detected
 507          */
 508         if (cp->flags & IP_VS_CONN_F_NOOUTPUT) {
 509                 if (state_off == TCP_DIR_OUTPUT)
 510                         cp->flags &= ~IP_VS_CONN_F_NOOUTPUT;
 511                 else
 512                         state_off = TCP_DIR_INPUT_ONLY;
 513         }
 514
 515         if ((state_idx = tcp_state_idx(th)) < 0) {
 516                 IP_VS_DBG(8, "tcp_state_idx=%d!!!\n", state_idx);
 517                 goto tcp_state_out;
 518         }
 519
 520         new_state = tcp_state_table[state_off+state_idx].next_state[cp->state];
 521
 522   tcp_state_out:
 523         if (new_state != cp->state) {
 524                 struct ip_vs_dest *dest = cp->dest;
 525
 526                 IP_VS_DBG_BUF(8, "%s %s [%c%c%c%c] %s:%d->"
 527                               "%s:%d state: %s->%s conn->refcnt:%d\n",
 528                               pp->name,
 529                               ((state_off == TCP_DIR_OUTPUT) ?
 530                                "output " : "input "),
 531                               th->syn ? 'S' : '.',
 532                               th->fin ? 'F' : '.',
 533                               th->ack ? 'A' : '.',
 534                               th->rst ? 'R' : '.',
 535                               IP_VS_DBG_ADDR(cp->af, &cp->daddr),
 536                               ntohs(cp->dport),
 537                               IP_VS_DBG_ADDR(cp->af, &cp->caddr),
 538                               ntohs(cp->cport),
 539                               tcp_state_name(cp->state),
 540                               tcp_state_name(new_state),
 541                               atomic_read(&cp->refcnt));
 542
 543                 if (dest) {
 544                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 545                             (new_state != IP_VS_TCP_S_ESTABLISHED)) {
 546                                 atomic_dec(&dest->activeconns);
 547                                 atomic_inc(&dest->inactconns);
 548                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
 549                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
 550                                    (new_state == IP_VS_TCP_S_ESTABLISHED)) {
 551                                 atomic_inc(&dest->activeconns);
 552                                 atomic_dec(&dest->inactconns);
 553                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 554                         }
 555                 }
 556         }
 557
 558         cp->timeout = pp->timeout_table[cp->state = new_state];
 559 }
 560
 561
 562 /*
 563  *      Handle state transitions
 564  */
 565 static int
 566 tcp_state_transition(struct ip_vs_conn *cp, int direction,
 567                      const struct sk_buff *skb,
 568                      struct ip_vs_protocol *pp)
 569 {
 570         struct tcphdr _tcph, *th;
 571
 572 #ifdef CONFIG_IP_VS_IPV6
 573         int ihl = cp->af == AF_INET ? ip_hdrlen(skb) : sizeof(struct ipv6hdr);
 574 #else
 575         int ihl = ip_hdrlen(skb);
 576 #endif
 577
 578         th = skb_header_pointer(skb, ihl, sizeof(_tcph), &_tcph);
 579         if (th == NULL)
 580                 return 0;
 581
 582         spin_lock(&cp->lock);
 583         set_tcp_state(pp, cp, direction, th);
 584         spin_unlock(&cp->lock);
 585
 586         return 1;
 587 }
 588
 589
 590 /*
 591  *      Hash table for TCP application incarnations
 592  */
 593 #define TCP_APP_TAB_BITS        4
 594 #define TCP_APP_TAB_SIZE        (1 << TCP_APP_TAB_BITS)
 595 #define TCP_APP_TAB_MASK        (TCP_APP_TAB_SIZE - 1)
 596
 597 static struct list_head tcp_apps[TCP_APP_TAB_SIZE];
 598 static DEFINE_SPINLOCK(tcp_app_lock);
 599
 600 static inline __u16 tcp_app_hashkey(__be16 port)
 601 {
 602         return (((__force u16)port >> TCP_APP_TAB_BITS) ^ (__force u16)port)
 603                 & TCP_APP_TAB_MASK;
 604 }
 605
 606
 607 static int tcp_register_app(struct ip_vs_app *inc)
 608 {
 609         struct ip_vs_app *i;
 610         __u16 hash;
 611         __be16 port = inc->port;
 612         int ret = 0;
 613
 614         hash = tcp_app_hashkey(port);
 615
 616         spin_lock_bh(&tcp_app_lock);
 617         list_for_each_entry(i, &tcp_apps[hash], p_list) {
 618                 if (i->port == port) {
 619                         ret = -EEXIST;
 620                         goto out;
 621                 }
 622         }
 623         list_add(&inc->p_list, &tcp_apps[hash]);
 624         atomic_inc(&ip_vs_protocol_tcp.appcnt);
 625
 626   out:
 627         spin_unlock_bh(&tcp_app_lock);
 628         return ret;
 629 }
 630
 631
 632 static void
 633 tcp_unregister_app(struct ip_vs_app *inc)
 634 {
 635         spin_lock_bh(&tcp_app_lock);
 636         atomic_dec(&ip_vs_protocol_tcp.appcnt);
 637         list_del(&inc->p_list);
 638         spin_unlock_bh(&tcp_app_lock);
 639 }
 640
 641
 642 static int
 643 tcp_app_conn_bind(struct ip_vs_conn *cp)
 644 {
 645         int hash;
 646         struct ip_vs_app *inc;
 647         int result = 0;
 648
 649         /* Default binding: bind app only for NAT */
 650         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
 651                 return 0;
 652
 653         /* Lookup application incarnations and bind the right one */
 654         hash = tcp_app_hashkey(cp->vport);
 655
 656         spin_lock(&tcp_app_lock);
 657         list_for_each_entry(inc, &tcp_apps[hash], p_list) {
 658                 if (inc->port == cp->vport) {
 659                         if (unlikely(!ip_vs_app_inc_get(inc)))
 660                                 break;
 661                         spin_unlock(&tcp_app_lock);
 662
 663                         IP_VS_DBG_BUF(9, "%s: Binding conn %s:%u->"
 664                                       "%s:%u to app %s on port %u\n",
 665                                       __func__,
 666                                       IP_VS_DBG_ADDR(cp->af, &cp->caddr),
 667                                       ntohs(cp->cport),
 668                                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
 669                                       ntohs(cp->vport),
 670                                       inc->name, ntohs(inc->port));
 671
 672                         cp->app = inc;
 673                         if (inc->init_conn)
 674                                 result = inc->init_conn(inc, cp);
 675                         goto out;
 676                 }
 677         }
 678         spin_unlock(&tcp_app_lock);
 679
 680   out:
 681         return result;
 682 }
 683
 684
 685 /*
 686  *      Set LISTEN timeout. (ip_vs_conn_put will setup timer)
 687  */
 688 void ip_vs_tcp_conn_listen(struct ip_vs_conn *cp)
 689 {
 690         spin_lock(&cp->lock);
 691         cp->state = IP_VS_TCP_S_LISTEN;
 692         cp->timeout = ip_vs_protocol_tcp.timeout_table[IP_VS_TCP_S_LISTEN];
 693         spin_unlock(&cp->lock);
 694 }
 695
 696
 697 static void ip_vs_tcp_init(struct ip_vs_protocol *pp)
 698 {
 699         IP_VS_INIT_HASH_TABLE(tcp_apps);
 700         pp->timeout_table = tcp_timeouts;
 701 }
 702
 703
 704 static void ip_vs_tcp_exit(struct ip_vs_protocol *pp)
 705 {
 706 }
 707
 708
 709 struct ip_vs_protocol ip_vs_protocol_tcp = {
 710         .name =                 "TCP",
 711         .protocol =             IPPROTO_TCP,
 712         .num_states =           IP_VS_TCP_S_LAST,
 713         .dont_defrag =          0,
 714         .appcnt =               ATOMIC_INIT(0),
 715         .init =                 ip_vs_tcp_init,
 716         .exit =                 ip_vs_tcp_exit,
 717         .register_app =         tcp_register_app,
 718         .unregister_app =       tcp_unregister_app,
 719         .conn_schedule =        tcp_conn_schedule,
 720         .conn_in_get =          tcp_conn_in_get,
 721         .conn_out_get =         tcp_conn_out_get,
 722         .snat_handler =         tcp_snat_handler,
 723         .dnat_handler =         tcp_dnat_handler,
 724         .csum_check =           tcp_csum_check,
 725         .state_name =           tcp_state_name,
 726         .state_transition =     tcp_state_transition,
 727         .app_conn_bind =        tcp_app_conn_bind,
 728         .debug_packet =         ip_vs_tcpudp_debug_packet,
 729         .timeout_change =       tcp_timeout_change,
 730         .set_state_timeout =    tcp_set_state_timeout,
 731 };