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