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 <net/ip6_checksum.h>
  22 #include <linux/netfilter.h>
  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(int af, const struct sk_buff *skb, struct ip_vs_protocol *pp,
  30                 const struct ip_vs_iphdr *iph, unsigned int proto_off,
  31                 int inverse)
  32 {
  33         __be16 _ports[2], *pptr;
  34
  35         pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports);
  36         if (pptr == NULL)
  37                 return NULL;
  38
  39         if (likely(!inverse)) {
  40                 return ip_vs_conn_in_get(af, iph->protocol,
  41                                          &iph->saddr, pptr[0],
  42                                          &iph->daddr, pptr[1]);
  43         } else {
  44                 return ip_vs_conn_in_get(af, iph->protocol,
  45                                          &iph->daddr, pptr[1],
  46                                          &iph->saddr, pptr[0]);
  47         }
  48 }
  49
  50 static struct ip_vs_conn *
  51 tcp_conn_out_get(int af, const struct sk_buff *skb, struct ip_vs_protocol *pp,
  52                  const struct ip_vs_iphdr *iph, unsigned int proto_off,
  53                  int inverse)
  54 {
  55         __be16 _ports[2], *pptr;
  56
  57         pptr = skb_header_pointer(skb, proto_off, sizeof(_ports), _ports);
  58         if (pptr == NULL)
  59                 return NULL;
  60
  61         if (likely(!inverse)) {
  62                 return ip_vs_conn_out_get(af, iph->protocol,
  63                                           &iph->saddr, pptr[0],
  64                                           &iph->daddr, pptr[1]);
  65         } else {
  66                 return ip_vs_conn_out_get(af, iph->protocol,
  67                                           &iph->daddr, pptr[1],
  68                                           &iph->saddr, pptr[0]);
  69         }
  70 }
  71
  72
  73 static int
  74 tcp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_protocol *pp,
  75                   int *verdict, struct ip_vs_conn **cpp)
  76 {
  77         struct ip_vs_service *svc;
  78         struct tcphdr _tcph, *th;
  79         struct ip_vs_iphdr iph;
  80
  81         ip_vs_fill_iphdr(af, skb_network_header(skb), &iph);
  82
  83         th = skb_header_pointer(skb, iph.len, sizeof(_tcph), &_tcph);
  84         if (th == NULL) {
  85                 *verdict = NF_DROP;
  86                 return 0;
  87         }
  88
  89         if (th->syn &&
  90             (svc = ip_vs_service_get(af, skb->mark, iph.protocol, &iph.daddr,
  91                                      th->dest))) {
  92                 if (ip_vs_todrop()) {
  93                         /*
  94                          * It seems that we are very loaded.
  95                          * We have to drop this packet :(
  96                          */
  97                         ip_vs_service_put(svc);
  98                         *verdict = NF_DROP;
  99                         return 0;
 100                 }
 101
 102                 /*
 103                  * Let the virtual server select a real server for the
 104                  * incoming connection, and create a connection entry.
 105                  */
 106                 *cpp = ip_vs_schedule(svc, skb);
 107                 if (!*cpp) {
 108                         *verdict = ip_vs_leave(svc, skb, pp);
 109                         return 0;
 110                 }
 111                 ip_vs_service_put(svc);
 112         }
 113         return 1;
 114 }
 115
 116
 117 static inline void
 118 tcp_fast_csum_update(int af, struct tcphdr *tcph,
 119                      const union nf_inet_addr *oldip,
 120                      const union nf_inet_addr *newip,
 121                      __be16 oldport, __be16 newport)
 122 {
 123 #ifdef CONFIG_IP_VS_IPV6
 124         if (af == AF_INET6)
 125                 tcph->check =
 126                         csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
 127                                          ip_vs_check_diff2(oldport, newport,
 128                                                 ~csum_unfold(tcph->check))));
 129         else
 130 #endif
 131         tcph->check =
 132                 csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
 133                                  ip_vs_check_diff2(oldport, newport,
 134                                                 ~csum_unfold(tcph->check))));
 135 }
 136
 137
 138 static inline void
 139 tcp_partial_csum_update(int af, struct tcphdr *tcph,
 140                      const union nf_inet_addr *oldip,
 141                      const union nf_inet_addr *newip,
 142                      __be16 oldlen, __be16 newlen)
 143 {
 144 #ifdef CONFIG_IP_VS_IPV6
 145         if (af == AF_INET6)
 146                 tcph->check =
 147                         csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
 148                                          ip_vs_check_diff2(oldlen, newlen,
 149                                                 ~csum_unfold(tcph->check))));
 150         else
 151 #endif
 152         tcph->check =
 153                 csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
 154                                 ip_vs_check_diff2(oldlen, newlen,
 155                                                 ~csum_unfold(tcph->check))));
 156 }
 157
 158
 159 static int
 160 tcp_snat_handler(struct sk_buff *skb,
 161                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
 162 {
 163         struct tcphdr *tcph;
 164         unsigned int tcphoff;
 165         int oldlen;
 166
 167 #ifdef CONFIG_IP_VS_IPV6
 168         if (cp->af == AF_INET6)
 169                 tcphoff = sizeof(struct ipv6hdr);
 170         else
 171 #endif
 172                 tcphoff = ip_hdrlen(skb);
 173         oldlen = skb->len - tcphoff;
 174
 175         /* csum_check requires unshared skb */
 176         if (!skb_make_writable(skb, tcphoff+sizeof(*tcph)))
 177                 return 0;
 178
 179         if (unlikely(cp->app != NULL)) {
 180                 /* Some checks before mangling */
 181                 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
 182                         return 0;
 183
 184                 /* Call application helper if needed */
 185                 if (!ip_vs_app_pkt_out(cp, skb))
 186                         return 0;
 187         }
 188
 189         tcph = (void *)skb_network_header(skb) + tcphoff;
 190         tcph->source = cp->vport;
 191
 192         /* Adjust TCP checksums */
 193         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 194                 tcp_partial_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr,
 195                                         htonl(oldlen),
 196                                         htonl(skb->len - tcphoff));
 197         } else if (!cp->app) {
 198                 /* Only port and addr are changed, do fast csum update */
 199                 tcp_fast_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr,
 200                                      cp->dport, cp->vport);
 201                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 202                         skb->ip_summed = CHECKSUM_NONE;
 203         } else {
 204                 /* full checksum calculation */
 205                 tcph->check = 0;
 206                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 207 #ifdef CONFIG_IP_VS_IPV6
 208                 if (cp->af == AF_INET6)
 209                         tcph->check = csum_ipv6_magic(&cp->vaddr.in6,
 210                                                       &cp->caddr.in6,
 211                                                       skb->len - tcphoff,
 212                                                       cp->protocol, skb->csum);
 213                 else
 214 #endif
 215                         tcph->check = csum_tcpudp_magic(cp->vaddr.ip,
 216                                                         cp->caddr.ip,
 217                                                         skb->len - tcphoff,
 218                                                         cp->protocol,
 219                                                         skb->csum);
 220
 221                 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n",
 222                           pp->name, tcph->check,
 223                           (char*)&(tcph->check) - (char*)tcph);
 224         }
 225         return 1;
 226 }
 227
 228
 229 static int
 230 tcp_dnat_handler(struct sk_buff *skb,
 231                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
 232 {
 233         struct tcphdr *tcph;
 234         unsigned int tcphoff;
 235         int oldlen;
 236
 237 #ifdef CONFIG_IP_VS_IPV6
 238         if (cp->af == AF_INET6)
 239                 tcphoff = sizeof(struct ipv6hdr);
 240         else
 241 #endif
 242                 tcphoff = ip_hdrlen(skb);
 243         oldlen = skb->len - tcphoff;
 244
 245         /* csum_check requires unshared skb */
 246         if (!skb_make_writable(skb, tcphoff+sizeof(*tcph)))
 247                 return 0;
 248
 249         if (unlikely(cp->app != NULL)) {
 250                 /* Some checks before mangling */
 251                 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
 252                         return 0;
 253
 254                 /*
 255                  *      Attempt ip_vs_app call.
 256                  *      It will fix ip_vs_conn and iph ack_seq stuff
 257                  */
 258                 if (!ip_vs_app_pkt_in(cp, skb))
 259                         return 0;
 260         }
 261
 262         tcph = (void *)skb_network_header(skb) + tcphoff;
 263         tcph->dest = cp->dport;
 264
 265         /*
 266          *      Adjust TCP checksums
 267          */
 268         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 269                 tcp_partial_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr,
 270                                         htonl(oldlen),
 271                                         htonl(skb->len - tcphoff));
 272         } else if (!cp->app) {
 273                 /* Only port and addr are changed, do fast csum update */
 274                 tcp_fast_csum_update(cp->af, tcph, &cp->vaddr, &cp->daddr,
 275                                      cp->vport, cp->dport);
 276                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 277                         skb->ip_summed = CHECKSUM_NONE;
 278         } else {
 279                 /* full checksum calculation */
 280                 tcph->check = 0;
 281                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 282 #ifdef CONFIG_IP_VS_IPV6
 283                 if (cp->af == AF_INET6)
 284                         tcph->check = csum_ipv6_magic(&cp->caddr.in6,
 285                                                       &cp->daddr.in6,
 286                                                       skb->len - tcphoff,
 287                                                       cp->protocol, skb->csum);
 288                 else
 289 #endif
 290                         tcph->check = csum_tcpudp_magic(cp->caddr.ip,
 291                                                         cp->daddr.ip,
 292                                                         skb->len - tcphoff,
 293                                                         cp->protocol,
 294                                                         skb->csum);
 295                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 296         }
 297         return 1;
 298 }
 299
 300
 301 static int
 302 tcp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp)
 303 {
 304         unsigned int tcphoff;
 305
 306 #ifdef CONFIG_IP_VS_IPV6
 307         if (af == AF_INET6)
 308                 tcphoff = sizeof(struct ipv6hdr);
 309         else
 310 #endif
 311                 tcphoff = ip_hdrlen(skb);
 312
 313         switch (skb->ip_summed) {
 314         case CHECKSUM_NONE:
 315                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 316         case CHECKSUM_COMPLETE:
 317 #ifdef CONFIG_IP_VS_IPV6
 318                 if (af == AF_INET6) {
 319                         if (csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 320                                             &ipv6_hdr(skb)->daddr,
 321                                             skb->len - tcphoff,
 322                                             ipv6_hdr(skb)->nexthdr,
 323                                             skb->csum)) {
 324                                 IP_VS_DBG_RL_PKT(0, pp, skb, 0,
 325                                                  "Failed checksum for");
 326                                 return 0;
 327                         }
 328                 } else
 329 #endif
 330                         if (csum_tcpudp_magic(ip_hdr(skb)->saddr,
 331                                               ip_hdr(skb)->daddr,
 332                                               skb->len - tcphoff,
 333                                               ip_hdr(skb)->protocol,
 334                                               skb->csum)) {
 335                                 IP_VS_DBG_RL_PKT(0, pp, skb, 0,
 336                                                  "Failed checksum for");
 337                                 return 0;
 338                         }
 339                 break;
 340         default:
 341                 /* No need to checksum. */
 342                 break;
 343         }
 344
 345         return 1;
 346 }
 347
 348
 349 #define TCP_DIR_INPUT           0
 350 #define TCP_DIR_OUTPUT          4
 351 #define TCP_DIR_INPUT_ONLY      8
 352
 353 static const int tcp_state_off[IP_VS_DIR_LAST] = {
 354         [IP_VS_DIR_INPUT]               =       TCP_DIR_INPUT,
 355         [IP_VS_DIR_OUTPUT]              =       TCP_DIR_OUTPUT,
 356         [IP_VS_DIR_INPUT_ONLY]          =       TCP_DIR_INPUT_ONLY,
 357 };
 358
 359 /*
 360  *      Timeout table[state]
 361  */
 362 static int tcp_timeouts[IP_VS_TCP_S_LAST+1] = {
 363         [IP_VS_TCP_S_NONE]              =       2*HZ,
 364         [IP_VS_TCP_S_ESTABLISHED]       =       15*60*HZ,
 365         [IP_VS_TCP_S_SYN_SENT]          =       2*60*HZ,
 366         [IP_VS_TCP_S_SYN_RECV]          =       1*60*HZ,
 367         [IP_VS_TCP_S_FIN_WAIT]          =       2*60*HZ,
 368         [IP_VS_TCP_S_TIME_WAIT]         =       2*60*HZ,
 369         [IP_VS_TCP_S_CLOSE]             =       10*HZ,
 370         [IP_VS_TCP_S_CLOSE_WAIT]        =       60*HZ,
 371         [IP_VS_TCP_S_LAST_ACK]          =       30*HZ,
 372         [IP_VS_TCP_S_LISTEN]            =       2*60*HZ,
 373         [IP_VS_TCP_S_SYNACK]            =       120*HZ,
 374         [IP_VS_TCP_S_LAST]              =       2*HZ,
 375 };
 376
 377 static char * tcp_state_name_table[IP_VS_TCP_S_LAST+1] = {
 378         [IP_VS_TCP_S_NONE]              =       "NONE",
 379         [IP_VS_TCP_S_ESTABLISHED]       =       "ESTABLISHED",
 380         [IP_VS_TCP_S_SYN_SENT]          =       "SYN_SENT",
 381         [IP_VS_TCP_S_SYN_RECV]          =       "SYN_RECV",
 382         [IP_VS_TCP_S_FIN_WAIT]          =       "FIN_WAIT",
 383         [IP_VS_TCP_S_TIME_WAIT]         =       "TIME_WAIT",
 384         [IP_VS_TCP_S_CLOSE]             =       "CLOSE",
 385         [IP_VS_TCP_S_CLOSE_WAIT]        =       "CLOSE_WAIT",
 386         [IP_VS_TCP_S_LAST_ACK]          =       "LAST_ACK",
 387         [IP_VS_TCP_S_LISTEN]            =       "LISTEN",
 388         [IP_VS_TCP_S_SYNACK]            =       "SYNACK",
 389         [IP_VS_TCP_S_LAST]              =       "BUG!",
 390 };
 391
 392 #define sNO IP_VS_TCP_S_NONE
 393 #define sES IP_VS_TCP_S_ESTABLISHED
 394 #define sSS IP_VS_TCP_S_SYN_SENT
 395 #define sSR IP_VS_TCP_S_SYN_RECV
 396 #define sFW IP_VS_TCP_S_FIN_WAIT
 397 #define sTW IP_VS_TCP_S_TIME_WAIT
 398 #define sCL IP_VS_TCP_S_CLOSE
 399 #define sCW IP_VS_TCP_S_CLOSE_WAIT
 400 #define sLA IP_VS_TCP_S_LAST_ACK
 401 #define sLI IP_VS_TCP_S_LISTEN
 402 #define sSA IP_VS_TCP_S_SYNACK
 403
 404 struct tcp_states_t {
 405         int next_state[IP_VS_TCP_S_LAST];
 406 };
 407
 408 static const char * tcp_state_name(int state)
 409 {
 410         if (state >= IP_VS_TCP_S_LAST)
 411                 return "ERR!";
 412         return tcp_state_name_table[state] ? tcp_state_name_table[state] : "?";
 413 }
 414
 415 static struct tcp_states_t tcp_states [] = {
 416 /*      INPUT */
 417 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 418 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }},
 419 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sTW }},
 420 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 421 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sSR }},
 422
 423 /*      OUTPUT */
 424 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 425 /*syn*/ {{sSS, sES, sSS, sSR, sSS, sSS, sSS, sSS, sSS, sLI, sSR }},
 426 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }},
 427 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }},
 428 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }},
 429
 430 /*      INPUT-ONLY */
 431 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 432 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }},
 433 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }},
 434 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 435 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 436 };
 437
 438 static struct tcp_states_t tcp_states_dos [] = {
 439 /*      INPUT */
 440 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 441 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSA }},
 442 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sSA }},
 443 /*ack*/ {{sCL, sES, sSS, sSR, sFW, sTW, sCL, sCW, sCL, sLI, sSA }},
 444 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 445
 446 /*      OUTPUT */
 447 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 448 /*syn*/ {{sSS, sES, sSS, sSA, sSS, sSS, sSS, sSS, sSS, sLI, sSA }},
 449 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }},
 450 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }},
 451 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }},
 452
 453 /*      INPUT-ONLY */
 454 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 455 /*syn*/ {{sSA, sES, sES, sSR, sSA, sSA, sSA, sSA, sSA, sSA, sSA }},
 456 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }},
 457 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 458 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 459 };
 460
 461 static struct tcp_states_t *tcp_state_table = tcp_states;
 462
 463
 464 static void tcp_timeout_change(struct ip_vs_protocol *pp, int flags)
 465 {
 466         int on = (flags & 1);           /* secure_tcp */
 467
 468         /*
 469         ** FIXME: change secure_tcp to independent sysctl var
 470         ** or make it per-service or per-app because it is valid
 471         ** for most if not for all of the applications. Something
 472         ** like "capabilities" (flags) for each object.
 473         */
 474         tcp_state_table = (on? tcp_states_dos : tcp_states);
 475 }
 476
 477 static int
 478 tcp_set_state_timeout(struct ip_vs_protocol *pp, char *sname, int to)
 479 {
 480         return ip_vs_set_state_timeout(pp->timeout_table, IP_VS_TCP_S_LAST,
 481                                        tcp_state_name_table, sname, to);
 482 }
 483
 484 static inline int tcp_state_idx(struct tcphdr *th)
 485 {
 486         if (th->rst)
 487                 return 3;
 488         if (th->syn)
 489                 return 0;
 490         if (th->fin)
 491                 return 1;
 492         if (th->ack)
 493                 return 2;
 494         return -1;
 495 }
 496
 497 static inline void
 498 set_tcp_state(struct ip_vs_protocol *pp, struct ip_vs_conn *cp,
 499               int direction, struct tcphdr *th)
 500 {
 501         int state_idx;
 502         int new_state = IP_VS_TCP_S_CLOSE;
 503         int state_off = tcp_state_off[direction];
 504
 505         /*
 506          *    Update state offset to INPUT_ONLY if necessary
 507          *    or delete NO_OUTPUT flag if output packet detected
 508          */
 509         if (cp->flags & IP_VS_CONN_F_NOOUTPUT) {
 510                 if (state_off == TCP_DIR_OUTPUT)
 511                         cp->flags &= ~IP_VS_CONN_F_NOOUTPUT;
 512                 else
 513                         state_off = TCP_DIR_INPUT_ONLY;
 514         }
 515
 516         if ((state_idx = tcp_state_idx(th)) < 0) {
 517                 IP_VS_DBG(8, "tcp_state_idx=%d!!!\n", state_idx);
 518                 goto tcp_state_out;
 519         }
 520
 521         new_state = tcp_state_table[state_off+state_idx].next_state[cp->state];
 522
 523   tcp_state_out:
 524         if (new_state != cp->state) {
 525                 struct ip_vs_dest *dest = cp->dest;
 526
 527                 IP_VS_DBG_BUF(8, "%s %s [%c%c%c%c] %s:%d->"
 528                               "%s:%d state: %s->%s conn->refcnt:%d\n",
 529                               pp->name,
 530                               ((state_off == TCP_DIR_OUTPUT) ?
 531                                "output " : "input "),
 532                               th->syn ? 'S' : '.',
 533                               th->fin ? 'F' : '.',
 534                               th->ack ? 'A' : '.',
 535                               th->rst ? 'R' : '.',
 536                               IP_VS_DBG_ADDR(cp->af, &cp->daddr),
 537                               ntohs(cp->dport),
 538                               IP_VS_DBG_ADDR(cp->af, &cp->caddr),
 539                               ntohs(cp->cport),
 540                               tcp_state_name(cp->state),
 541                               tcp_state_name(new_state),
 542                               atomic_read(&cp->refcnt));
 543
 544                 if (dest) {
 545                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 546                             (new_state != IP_VS_TCP_S_ESTABLISHED)) {
 547                                 atomic_dec(&dest->activeconns);
 548                                 atomic_inc(&dest->inactconns);
 549                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
 550                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
 551                                    (new_state == IP_VS_TCP_S_ESTABLISHED)) {
 552                                 atomic_inc(&dest->activeconns);
 553                                 atomic_dec(&dest->inactconns);
 554                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 555                         }
 556                 }
 557         }
 558
 559         cp->timeout = pp->timeout_table[cp->state = new_state];
 560 }
 561
 562
 563 /*
 564  *      Handle state transitions
 565  */
 566 static int
 567 tcp_state_transition(struct ip_vs_conn *cp, int direction,
 568                      const struct sk_buff *skb,
 569                      struct ip_vs_protocol *pp)
 570 {
 571         struct tcphdr _tcph, *th;
 572
 573 #ifdef CONFIG_IP_VS_IPV6
 574         int ihl = cp->af == AF_INET ? ip_hdrlen(skb) : sizeof(struct ipv6hdr);
 575 #else
 576         int ihl = ip_hdrlen(skb);
 577 #endif
 578
 579         th = skb_header_pointer(skb, ihl, sizeof(_tcph), &_tcph);
 580         if (th == NULL)
 581                 return 0;
 582
 583         spin_lock(&cp->lock);
 584         set_tcp_state(pp, cp, direction, th);
 585         spin_unlock(&cp->lock);
 586
 587         return 1;
 588 }
 589
 590
 591 /*
 592  *      Hash table for TCP application incarnations
 593  */
 594 #define TCP_APP_TAB_BITS        4
 595 #define TCP_APP_TAB_SIZE        (1 << TCP_APP_TAB_BITS)
 596 #define TCP_APP_TAB_MASK        (TCP_APP_TAB_SIZE - 1)
 597
 598 static struct list_head tcp_apps[TCP_APP_TAB_SIZE];
 599 static DEFINE_SPINLOCK(tcp_app_lock);
 600
 601 static inline __u16 tcp_app_hashkey(__be16 port)
 602 {
 603         return (((__force u16)port >> TCP_APP_TAB_BITS) ^ (__force u16)port)
 604                 & TCP_APP_TAB_MASK;
 605 }
 606
 607
 608 static int tcp_register_app(struct ip_vs_app *inc)
 609 {
 610         struct ip_vs_app *i;
 611         __u16 hash;
 612         __be16 port = inc->port;
 613         int ret = 0;
 614
 615         hash = tcp_app_hashkey(port);
 616
 617         spin_lock_bh(&tcp_app_lock);
 618         list_for_each_entry(i, &tcp_apps[hash], p_list) {
 619                 if (i->port == port) {
 620                         ret = -EEXIST;
 621                         goto out;
 622                 }
 623         }
 624         list_add(&inc->p_list, &tcp_apps[hash]);
 625         atomic_inc(&ip_vs_protocol_tcp.appcnt);
 626
 627   out:
 628         spin_unlock_bh(&tcp_app_lock);
 629         return ret;
 630 }
 631
 632
 633 static void
 634 tcp_unregister_app(struct ip_vs_app *inc)
 635 {
 636         spin_lock_bh(&tcp_app_lock);
 637         atomic_dec(&ip_vs_protocol_tcp.appcnt);
 638         list_del(&inc->p_list);
 639         spin_unlock_bh(&tcp_app_lock);
 640 }
 641
 642
 643 static int
 644 tcp_app_conn_bind(struct ip_vs_conn *cp)
 645 {
 646         int hash;
 647         struct ip_vs_app *inc;
 648         int result = 0;
 649
 650         /* Default binding: bind app only for NAT */
 651         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
 652                 return 0;
 653
 654         /* Lookup application incarnations and bind the right one */
 655         hash = tcp_app_hashkey(cp->vport);
 656
 657         spin_lock(&tcp_app_lock);
 658         list_for_each_entry(inc, &tcp_apps[hash], p_list) {
 659                 if (inc->port == cp->vport) {
 660                         if (unlikely(!ip_vs_app_inc_get(inc)))
 661                                 break;
 662                         spin_unlock(&tcp_app_lock);
 663
 664                         IP_VS_DBG_BUF(9, "%s: Binding conn %s:%u->"
 665                                       "%s:%u to app %s on port %u\n",
 666                                       __func__,
 667                                       IP_VS_DBG_ADDR(cp->af, &cp->caddr),
 668                                       ntohs(cp->cport),
 669                                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
 670                                       ntohs(cp->vport),
 671                                       inc->name, ntohs(inc->port));
 672
 673                         cp->app = inc;
 674                         if (inc->init_conn)
 675                                 result = inc->init_conn(inc, cp);
 676                         goto out;
 677                 }
 678         }
 679         spin_unlock(&tcp_app_lock);
 680
 681   out:
 682         return result;
 683 }
 684
 685
 686 /*
 687  *      Set LISTEN timeout. (ip_vs_conn_put will setup timer)
 688  */
 689 void ip_vs_tcp_conn_listen(struct ip_vs_conn *cp)
 690 {
 691         spin_lock(&cp->lock);
 692         cp->state = IP_VS_TCP_S_LISTEN;
 693         cp->timeout = ip_vs_protocol_tcp.timeout_table[IP_VS_TCP_S_LISTEN];
 694         spin_unlock(&cp->lock);
 695 }
 696
 697
 698 static void ip_vs_tcp_init(struct ip_vs_protocol *pp)
 699 {
 700         IP_VS_INIT_HASH_TABLE(tcp_apps);
 701         pp->timeout_table = tcp_timeouts;
 702 }
 703
 704
 705 static void ip_vs_tcp_exit(struct ip_vs_protocol *pp)
 706 {
 707 }
 708
 709
 710 struct ip_vs_protocol ip_vs_protocol_tcp = {
 711         .name =                 "TCP",
 712         .protocol =             IPPROTO_TCP,
 713         .num_states =           IP_VS_TCP_S_LAST,
 714         .dont_defrag =          0,
 715         .appcnt =               ATOMIC_INIT(0),
 716         .init =                 ip_vs_tcp_init,
 717         .exit =                 ip_vs_tcp_exit,
 718         .register_app =         tcp_register_app,
 719         .unregister_app =       tcp_unregister_app,
 720         .conn_schedule =        tcp_conn_schedule,
 721         .conn_in_get =          tcp_conn_in_get,
 722         .conn_out_get =         tcp_conn_out_get,
 723         .snat_handler =         tcp_snat_handler,
 724         .dnat_handler =         tcp_dnat_handler,
 725         .csum_check =           tcp_csum_check,
 726         .state_name =           tcp_state_name,
 727         .state_transition =     tcp_state_transition,
 728         .app_conn_bind =        tcp_app_conn_bind,
 729         .debug_packet =         ip_vs_tcpudp_debug_packet,
 730         .timeout_change =       tcp_timeout_change,
 731         .set_state_timeout =    tcp_set_state_timeout,
 732 };