net/netfilter/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:     Hans Schillstrom <hans.schillstrom@ericsson.com>
  13  *
  14  *              Network name space (netns) aware.
  15  *              Global data moved to netns i.e struct netns_ipvs
  16  *              tcp_timeouts table has copy per netns in a hash table per
  17  *              protocol ip_vs_proto_data and is handled by netns
  18  */
  19
  20 #define KMSG_COMPONENT "IPVS"
  21 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  22
  23 #include <linux/kernel.h>
  24 #include <linux/ip.h>
  25 #include <linux/tcp.h>                  /* for tcphdr */
  26 #include <net/ip.h>
  27 #include <net/tcp.h>                    /* for csum_tcpudp_magic */
  28 #include <net/ip6_checksum.h>
  29 #include <linux/netfilter.h>
  30 #include <linux/netfilter_ipv4.h>
  31
  32 #include <net/ip_vs.h>
  33
  34 static int
  35 tcp_conn_schedule(int af, struct sk_buff *skb, struct ip_vs_proto_data *pd,
  36                   int *verdict, struct ip_vs_conn **cpp)
  37 {
  38         struct net *net;
  39         struct ip_vs_service *svc;
  40         struct tcphdr _tcph, *th;
  41         struct ip_vs_iphdr iph;
  42
  43         ip_vs_fill_iph_skb(af, skb, &iph);
  44
  45         th = skb_header_pointer(skb, iph.len, sizeof(_tcph), &_tcph);
  46         if (th == NULL) {
  47                 *verdict = NF_DROP;
  48                 return 0;
  49         }
  50         net = skb_net(skb);
  51         /* No !th->ack check to allow scheduling on SYN+ACK for Active FTP */
  52         if (th->syn &&
  53             (svc = ip_vs_service_get(net, af, skb->mark, iph.protocol,
  54                                      &iph.daddr, th->dest))) {
  55                 int ignored;
  56
  57                 if (ip_vs_todrop(net_ipvs(net))) {
  58                         /*
  59                          * It seems that we are very loaded.
  60                          * We have to drop this packet :(
  61                          */
  62                         ip_vs_service_put(svc);
  63                         *verdict = NF_DROP;
  64                         return 0;
  65                 }
  66
  67                 /*
  68                  * Let the virtual server select a real server for the
  69                  * incoming connection, and create a connection entry.
  70                  */
  71                 *cpp = ip_vs_schedule(svc, skb, pd, &ignored);
  72                 if (!*cpp && ignored <= 0) {
  73                         if (!ignored)
  74                                 *verdict = ip_vs_leave(svc, skb, pd);
  75                         else {
  76                                 ip_vs_service_put(svc);
  77                                 *verdict = NF_DROP;
  78                         }
  79                         return 0;
  80                 }
  81                 ip_vs_service_put(svc);
  82         }
  83         /* NF_ACCEPT */
  84         return 1;
  85 }
  86
  87
  88 static inline void
  89 tcp_fast_csum_update(int af, struct tcphdr *tcph,
  90                      const union nf_inet_addr *oldip,
  91                      const union nf_inet_addr *newip,
  92                      __be16 oldport, __be16 newport)
  93 {
  94 #ifdef CONFIG_IP_VS_IPV6
  95         if (af == AF_INET6)
  96                 tcph->check =
  97                         csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
  98                                          ip_vs_check_diff2(oldport, newport,
  99                                                 ~csum_unfold(tcph->check))));
 100         else
 101 #endif
 102         tcph->check =
 103                 csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
 104                                  ip_vs_check_diff2(oldport, newport,
 105                                                 ~csum_unfold(tcph->check))));
 106 }
 107
 108
 109 static inline void
 110 tcp_partial_csum_update(int af, struct tcphdr *tcph,
 111                      const union nf_inet_addr *oldip,
 112                      const union nf_inet_addr *newip,
 113                      __be16 oldlen, __be16 newlen)
 114 {
 115 #ifdef CONFIG_IP_VS_IPV6
 116         if (af == AF_INET6)
 117                 tcph->check =
 118                         ~csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
 119                                          ip_vs_check_diff2(oldlen, newlen,
 120                                                 csum_unfold(tcph->check))));
 121         else
 122 #endif
 123         tcph->check =
 124                 ~csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
 125                                 ip_vs_check_diff2(oldlen, newlen,
 126                                                 csum_unfold(tcph->check))));
 127 }
 128
 129
 130 static int
 131 tcp_snat_handler(struct sk_buff *skb,
 132                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
 133 {
 134         struct tcphdr *tcph;
 135         unsigned int tcphoff;
 136         int oldlen;
 137         int payload_csum = 0;
 138
 139         struct ip_vs_iphdr iph;
 140         ip_vs_fill_iph_skb(cp->af, skb, &iph);
 141         tcphoff = iph.len;
 142
 143 #ifdef CONFIG_IP_VS_IPV6
 144         if (cp->af == AF_INET6 && iph.fragoffs)
 145                 return 1;
 146 #endif
 147         oldlen = skb->len - tcphoff;
 148
 149         /* csum_check requires unshared skb */
 150         if (!skb_make_writable(skb, tcphoff+sizeof(*tcph)))
 151                 return 0;
 152
 153         if (unlikely(cp->app != NULL)) {
 154                 int ret;
 155
 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 (!(ret = ip_vs_app_pkt_out(cp, skb)))
 162                         return 0;
 163                 /* ret=2: csum update is needed after payload mangling */
 164                 if (ret == 1)
 165                         oldlen = skb->len - tcphoff;
 166                 else
 167                         payload_csum = 1;
 168         }
 169
 170         tcph = (void *)skb_network_header(skb) + tcphoff;
 171         tcph->source = cp->vport;
 172
 173         /* Adjust TCP checksums */
 174         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 175                 tcp_partial_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr,
 176                                         htons(oldlen),
 177                                         htons(skb->len - tcphoff));
 178         } else if (!payload_csum) {
 179                 /* Only port and addr are changed, do fast csum update */
 180                 tcp_fast_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr,
 181                                      cp->dport, cp->vport);
 182                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 183                         skb->ip_summed = (cp->app && pp->csum_check) ?
 184                                          CHECKSUM_UNNECESSARY : CHECKSUM_NONE;
 185         } else {
 186                 /* full checksum calculation */
 187                 tcph->check = 0;
 188                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 189 #ifdef CONFIG_IP_VS_IPV6
 190                 if (cp->af == AF_INET6)
 191                         tcph->check = csum_ipv6_magic(&cp->vaddr.in6,
 192                                                       &cp->caddr.in6,
 193                                                       skb->len - tcphoff,
 194                                                       cp->protocol, skb->csum);
 195                 else
 196 #endif
 197                         tcph->check = csum_tcpudp_magic(cp->vaddr.ip,
 198                                                         cp->caddr.ip,
 199                                                         skb->len - tcphoff,
 200                                                         cp->protocol,
 201                                                         skb->csum);
 202                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 203
 204                 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n",
 205                           pp->name, tcph->check,
 206                           (char*)&(tcph->check) - (char*)tcph);
 207         }
 208         return 1;
 209 }
 210
 211
 212 static int
 213 tcp_dnat_handler(struct sk_buff *skb,
 214                  struct ip_vs_protocol *pp, struct ip_vs_conn *cp)
 215 {
 216         struct tcphdr *tcph;
 217         unsigned int tcphoff;
 218         int oldlen;
 219         int payload_csum = 0;
 220
 221         struct ip_vs_iphdr iph;
 222         ip_vs_fill_iph_skb(cp->af, skb, &iph);
 223         tcphoff = iph.len;
 224
 225 #ifdef CONFIG_IP_VS_IPV6
 226         if (cp->af == AF_INET6 && iph.fragoffs)
 227                 return 1;
 228 #endif
 229         oldlen = skb->len - tcphoff;
 230
 231         /* csum_check requires unshared skb */
 232         if (!skb_make_writable(skb, tcphoff+sizeof(*tcph)))
 233                 return 0;
 234
 235         if (unlikely(cp->app != NULL)) {
 236                 int ret;
 237
 238                 /* Some checks before mangling */
 239                 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
 240                         return 0;
 241
 242                 /*
 243                  *      Attempt ip_vs_app call.
 244                  *      It will fix ip_vs_conn and iph ack_seq stuff
 245                  */
 246                 if (!(ret = ip_vs_app_pkt_in(cp, skb)))
 247                         return 0;
 248                 /* ret=2: csum update is needed after payload mangling */
 249                 if (ret == 1)
 250                         oldlen = skb->len - tcphoff;
 251                 else
 252                         payload_csum = 1;
 253         }
 254
 255         tcph = (void *)skb_network_header(skb) + tcphoff;
 256         tcph->dest = cp->dport;
 257
 258         /*
 259          *      Adjust TCP checksums
 260          */
 261         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 262                 tcp_partial_csum_update(cp->af, tcph, &cp->vaddr, &cp->daddr,
 263                                         htons(oldlen),
 264                                         htons(skb->len - tcphoff));
 265         } else if (!payload_csum) {
 266                 /* Only port and addr are changed, do fast csum update */
 267                 tcp_fast_csum_update(cp->af, tcph, &cp->vaddr, &cp->daddr,
 268                                      cp->vport, cp->dport);
 269                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 270                         skb->ip_summed = (cp->app && pp->csum_check) ?
 271                                          CHECKSUM_UNNECESSARY : CHECKSUM_NONE;
 272         } else {
 273                 /* full checksum calculation */
 274                 tcph->check = 0;
 275                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 276 #ifdef CONFIG_IP_VS_IPV6
 277                 if (cp->af == AF_INET6)
 278                         tcph->check = csum_ipv6_magic(&cp->caddr.in6,
 279                                                       &cp->daddr.in6,
 280                                                       skb->len - tcphoff,
 281                                                       cp->protocol, skb->csum);
 282                 else
 283 #endif
 284                         tcph->check = csum_tcpudp_magic(cp->caddr.ip,
 285                                                         cp->daddr.ip,
 286                                                         skb->len - tcphoff,
 287                                                         cp->protocol,
 288                                                         skb->csum);
 289                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 290         }
 291         return 1;
 292 }
 293
 294
 295 static int
 296 tcp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp)
 297 {
 298         unsigned int tcphoff;
 299
 300 #ifdef CONFIG_IP_VS_IPV6
 301         if (af == AF_INET6)
 302                 tcphoff = sizeof(struct ipv6hdr);
 303         else
 304 #endif
 305                 tcphoff = ip_hdrlen(skb);
 306
 307         switch (skb->ip_summed) {
 308         case CHECKSUM_NONE:
 309                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 310         case CHECKSUM_COMPLETE:
 311 #ifdef CONFIG_IP_VS_IPV6
 312                 if (af == AF_INET6) {
 313                         if (csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 314                                             &ipv6_hdr(skb)->daddr,
 315                                             skb->len - tcphoff,
 316                                             ipv6_hdr(skb)->nexthdr,
 317                                             skb->csum)) {
 318                                 IP_VS_DBG_RL_PKT(0, af, pp, skb, 0,
 319                                                  "Failed checksum for");
 320                                 return 0;
 321                         }
 322                 } else
 323 #endif
 324                         if (csum_tcpudp_magic(ip_hdr(skb)->saddr,
 325                                               ip_hdr(skb)->daddr,
 326                                               skb->len - tcphoff,
 327                                               ip_hdr(skb)->protocol,
 328                                               skb->csum)) {
 329                                 IP_VS_DBG_RL_PKT(0, af, pp, skb, 0,
 330                                                  "Failed checksum for");
 331                                 return 0;
 332                         }
 333                 break;
 334         default:
 335                 /* No need to checksum. */
 336                 break;
 337         }
 338
 339         return 1;
 340 }
 341
 342
 343 #define TCP_DIR_INPUT           0
 344 #define TCP_DIR_OUTPUT          4
 345 #define TCP_DIR_INPUT_ONLY      8
 346
 347 static const int tcp_state_off[IP_VS_DIR_LAST] = {
 348         [IP_VS_DIR_INPUT]               =       TCP_DIR_INPUT,
 349         [IP_VS_DIR_OUTPUT]              =       TCP_DIR_OUTPUT,
 350         [IP_VS_DIR_INPUT_ONLY]          =       TCP_DIR_INPUT_ONLY,
 351 };
 352
 353 /*
 354  *      Timeout table[state]
 355  */
 356 static const int tcp_timeouts[IP_VS_TCP_S_LAST+1] = {
 357         [IP_VS_TCP_S_NONE]              =       2*HZ,
 358         [IP_VS_TCP_S_ESTABLISHED]       =       15*60*HZ,
 359         [IP_VS_TCP_S_SYN_SENT]          =       2*60*HZ,
 360         [IP_VS_TCP_S_SYN_RECV]          =       1*60*HZ,
 361         [IP_VS_TCP_S_FIN_WAIT]          =       2*60*HZ,
 362         [IP_VS_TCP_S_TIME_WAIT]         =       2*60*HZ,
 363         [IP_VS_TCP_S_CLOSE]             =       10*HZ,
 364         [IP_VS_TCP_S_CLOSE_WAIT]        =       60*HZ,
 365         [IP_VS_TCP_S_LAST_ACK]          =       30*HZ,
 366         [IP_VS_TCP_S_LISTEN]            =       2*60*HZ,
 367         [IP_VS_TCP_S_SYNACK]            =       120*HZ,
 368         [IP_VS_TCP_S_LAST]              =       2*HZ,
 369 };
 370
 371 static const char *const tcp_state_name_table[IP_VS_TCP_S_LAST+1] = {
 372         [IP_VS_TCP_S_NONE]              =       "NONE",
 373         [IP_VS_TCP_S_ESTABLISHED]       =       "ESTABLISHED",
 374         [IP_VS_TCP_S_SYN_SENT]          =       "SYN_SENT",
 375         [IP_VS_TCP_S_SYN_RECV]          =       "SYN_RECV",
 376         [IP_VS_TCP_S_FIN_WAIT]          =       "FIN_WAIT",
 377         [IP_VS_TCP_S_TIME_WAIT]         =       "TIME_WAIT",
 378         [IP_VS_TCP_S_CLOSE]             =       "CLOSE",
 379         [IP_VS_TCP_S_CLOSE_WAIT]        =       "CLOSE_WAIT",
 380         [IP_VS_TCP_S_LAST_ACK]          =       "LAST_ACK",
 381         [IP_VS_TCP_S_LISTEN]            =       "LISTEN",
 382         [IP_VS_TCP_S_SYNACK]            =       "SYNACK",
 383         [IP_VS_TCP_S_LAST]              =       "BUG!",
 384 };
 385
 386 #define sNO IP_VS_TCP_S_NONE
 387 #define sES IP_VS_TCP_S_ESTABLISHED
 388 #define sSS IP_VS_TCP_S_SYN_SENT
 389 #define sSR IP_VS_TCP_S_SYN_RECV
 390 #define sFW IP_VS_TCP_S_FIN_WAIT
 391 #define sTW IP_VS_TCP_S_TIME_WAIT
 392 #define sCL IP_VS_TCP_S_CLOSE
 393 #define sCW IP_VS_TCP_S_CLOSE_WAIT
 394 #define sLA IP_VS_TCP_S_LAST_ACK
 395 #define sLI IP_VS_TCP_S_LISTEN
 396 #define sSA IP_VS_TCP_S_SYNACK
 397
 398 struct tcp_states_t {
 399         int next_state[IP_VS_TCP_S_LAST];
 400 };
 401
 402 static const char * tcp_state_name(int state)
 403 {
 404         if (state >= IP_VS_TCP_S_LAST)
 405                 return "ERR!";
 406         return tcp_state_name_table[state] ? tcp_state_name_table[state] : "?";
 407 }
 408
 409 static struct tcp_states_t tcp_states [] = {
 410 /*      INPUT */
 411 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 412 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }},
 413 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sTW }},
 414 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 415 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sSR }},
 416
 417 /*      OUTPUT */
 418 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 419 /*syn*/ {{sSS, sES, sSS, sSR, sSS, sSS, sSS, sSS, sSS, sLI, sSR }},
 420 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }},
 421 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }},
 422 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }},
 423
 424 /*      INPUT-ONLY */
 425 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 426 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }},
 427 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }},
 428 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 429 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 430 };
 431
 432 static struct tcp_states_t tcp_states_dos [] = {
 433 /*      INPUT */
 434 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 435 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSA }},
 436 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sSA }},
 437 /*ack*/ {{sCL, sES, sSS, sSR, sFW, sTW, sCL, sCW, sCL, sLI, sSA }},
 438 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 439
 440 /*      OUTPUT */
 441 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 442 /*syn*/ {{sSS, sES, sSS, sSA, sSS, sSS, sSS, sSS, sSS, sLI, sSA }},
 443 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }},
 444 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }},
 445 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }},
 446
 447 /*      INPUT-ONLY */
 448 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 449 /*syn*/ {{sSA, sES, sES, sSR, sSA, sSA, sSA, sSA, sSA, sSA, sSA }},
 450 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }},
 451 /*ack*/ {{sCL, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 452 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 453 };
 454
 455 static void tcp_timeout_change(struct ip_vs_proto_data *pd, int flags)
 456 {
 457         int on = (flags & 1);           /* secure_tcp */
 458
 459         /*
 460         ** FIXME: change secure_tcp to independent sysctl var
 461         ** or make it per-service or per-app because it is valid
 462         ** for most if not for all of the applications. Something
 463         ** like "capabilities" (flags) for each object.
 464         */
 465         pd->tcp_state_table = (on ? tcp_states_dos : tcp_states);
 466 }
 467
 468 static inline int tcp_state_idx(struct tcphdr *th)
 469 {
 470         if (th->rst)
 471                 return 3;
 472         if (th->syn)
 473                 return 0;
 474         if (th->fin)
 475                 return 1;
 476         if (th->ack)
 477                 return 2;
 478         return -1;
 479 }
 480
 481 static inline void
 482 set_tcp_state(struct ip_vs_proto_data *pd, struct ip_vs_conn *cp,
 483               int direction, struct tcphdr *th)
 484 {
 485         int state_idx;
 486         int new_state = IP_VS_TCP_S_CLOSE;
 487         int state_off = tcp_state_off[direction];
 488
 489         /*
 490          *    Update state offset to INPUT_ONLY if necessary
 491          *    or delete NO_OUTPUT flag if output packet detected
 492          */
 493         if (cp->flags & IP_VS_CONN_F_NOOUTPUT) {
 494                 if (state_off == TCP_DIR_OUTPUT)
 495                         cp->flags &= ~IP_VS_CONN_F_NOOUTPUT;
 496                 else
 497                         state_off = TCP_DIR_INPUT_ONLY;
 498         }
 499
 500         if ((state_idx = tcp_state_idx(th)) < 0) {
 501                 IP_VS_DBG(8, "tcp_state_idx=%d!!!\n", state_idx);
 502                 goto tcp_state_out;
 503         }
 504
 505         new_state =
 506                 pd->tcp_state_table[state_off+state_idx].next_state[cp->state];
 507
 508   tcp_state_out:
 509         if (new_state != cp->state) {
 510                 struct ip_vs_dest *dest = cp->dest;
 511
 512                 IP_VS_DBG_BUF(8, "%s %s [%c%c%c%c] %s:%d->"
 513                               "%s:%d state: %s->%s conn->refcnt:%d\n",
 514                               pd->pp->name,
 515                               ((state_off == TCP_DIR_OUTPUT) ?
 516                                "output " : "input "),
 517                               th->syn ? 'S' : '.',
 518                               th->fin ? 'F' : '.',
 519                               th->ack ? 'A' : '.',
 520                               th->rst ? 'R' : '.',
 521                               IP_VS_DBG_ADDR(cp->af, &cp->daddr),
 522                               ntohs(cp->dport),
 523                               IP_VS_DBG_ADDR(cp->af, &cp->caddr),
 524                               ntohs(cp->cport),
 525                               tcp_state_name(cp->state),
 526                               tcp_state_name(new_state),
 527                               atomic_read(&cp->refcnt));
 528
 529                 if (dest) {
 530                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 531                             (new_state != IP_VS_TCP_S_ESTABLISHED)) {
 532                                 atomic_dec(&dest->activeconns);
 533                                 atomic_inc(&dest->inactconns);
 534                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
 535                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
 536                                    (new_state == IP_VS_TCP_S_ESTABLISHED)) {
 537                                 atomic_inc(&dest->activeconns);
 538                                 atomic_dec(&dest->inactconns);
 539                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 540                         }
 541                 }
 542         }
 543
 544         if (likely(pd))
 545                 cp->timeout = pd->timeout_table[cp->state = new_state];
 546         else    /* What to do ? */
 547                 cp->timeout = tcp_timeouts[cp->state = new_state];
 548 }
 549
 550 /*
 551  *      Handle state transitions
 552  */
 553 static void
 554 tcp_state_transition(struct ip_vs_conn *cp, int direction,
 555                      const struct sk_buff *skb,
 556                      struct ip_vs_proto_data *pd)
 557 {
 558         struct tcphdr _tcph, *th;
 559
 560 #ifdef CONFIG_IP_VS_IPV6
 561         int ihl = cp->af == AF_INET ? ip_hdrlen(skb) : sizeof(struct ipv6hdr);
 562 #else
 563         int ihl = ip_hdrlen(skb);
 564 #endif
 565
 566         th = skb_header_pointer(skb, ihl, sizeof(_tcph), &_tcph);
 567         if (th == NULL)
 568                 return;
 569
 570         spin_lock(&cp->lock);
 571         set_tcp_state(pd, cp, direction, th);
 572         spin_unlock(&cp->lock);
 573 }
 574
 575 static inline __u16 tcp_app_hashkey(__be16 port)
 576 {
 577         return (((__force u16)port >> TCP_APP_TAB_BITS) ^ (__force u16)port)
 578                 & TCP_APP_TAB_MASK;
 579 }
 580
 581
 582 static int tcp_register_app(struct net *net, struct ip_vs_app *inc)
 583 {
 584         struct ip_vs_app *i;
 585         __u16 hash;
 586         __be16 port = inc->port;
 587         int ret = 0;
 588         struct netns_ipvs *ipvs = net_ipvs(net);
 589         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(net, IPPROTO_TCP);
 590
 591         hash = tcp_app_hashkey(port);
 592
 593         spin_lock_bh(&ipvs->tcp_app_lock);
 594         list_for_each_entry(i, &ipvs->tcp_apps[hash], p_list) {
 595                 if (i->port == port) {
 596                         ret = -EEXIST;
 597                         goto out;
 598                 }
 599         }
 600         list_add(&inc->p_list, &ipvs->tcp_apps[hash]);
 601         atomic_inc(&pd->appcnt);
 602
 603   out:
 604         spin_unlock_bh(&ipvs->tcp_app_lock);
 605         return ret;
 606 }
 607
 608
 609 static void
 610 tcp_unregister_app(struct net *net, struct ip_vs_app *inc)
 611 {
 612         struct netns_ipvs *ipvs = net_ipvs(net);
 613         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(net, IPPROTO_TCP);
 614
 615         spin_lock_bh(&ipvs->tcp_app_lock);
 616         atomic_dec(&pd->appcnt);
 617         list_del(&inc->p_list);
 618         spin_unlock_bh(&ipvs->tcp_app_lock);
 619 }
 620
 621
 622 static int
 623 tcp_app_conn_bind(struct ip_vs_conn *cp)
 624 {
 625         struct netns_ipvs *ipvs = net_ipvs(ip_vs_conn_net(cp));
 626         int hash;
 627         struct ip_vs_app *inc;
 628         int result = 0;
 629
 630         /* Default binding: bind app only for NAT */
 631         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
 632                 return 0;
 633
 634         /* Lookup application incarnations and bind the right one */
 635         hash = tcp_app_hashkey(cp->vport);
 636
 637         spin_lock(&ipvs->tcp_app_lock);
 638         list_for_each_entry(inc, &ipvs->tcp_apps[hash], p_list) {
 639                 if (inc->port == cp->vport) {
 640                         if (unlikely(!ip_vs_app_inc_get(inc)))
 641                                 break;
 642                         spin_unlock(&ipvs->tcp_app_lock);
 643
 644                         IP_VS_DBG_BUF(9, "%s(): Binding conn %s:%u->"
 645                                       "%s:%u to app %s on port %u\n",
 646                                       __func__,
 647                                       IP_VS_DBG_ADDR(cp->af, &cp->caddr),
 648                                       ntohs(cp->cport),
 649                                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
 650                                       ntohs(cp->vport),
 651                                       inc->name, ntohs(inc->port));
 652
 653                         cp->app = inc;
 654                         if (inc->init_conn)
 655                                 result = inc->init_conn(inc, cp);
 656                         goto out;
 657                 }
 658         }
 659         spin_unlock(&ipvs->tcp_app_lock);
 660
 661   out:
 662         return result;
 663 }
 664
 665
 666 /*
 667  *      Set LISTEN timeout. (ip_vs_conn_put will setup timer)
 668  */
 669 void ip_vs_tcp_conn_listen(struct net *net, struct ip_vs_conn *cp)
 670 {
 671         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(net, IPPROTO_TCP);
 672
 673         spin_lock(&cp->lock);
 674         cp->state = IP_VS_TCP_S_LISTEN;
 675         cp->timeout = (pd ? pd->timeout_table[IP_VS_TCP_S_LISTEN]
 676                            : tcp_timeouts[IP_VS_TCP_S_LISTEN]);
 677         spin_unlock(&cp->lock);
 678 }
 679
 680 /* ---------------------------------------------
 681  *   timeouts is netns related now.
 682  * ---------------------------------------------
 683  */
 684 static int __ip_vs_tcp_init(struct net *net, struct ip_vs_proto_data *pd)
 685 {
 686         struct netns_ipvs *ipvs = net_ipvs(net);
 687
 688         ip_vs_init_hash_table(ipvs->tcp_apps, TCP_APP_TAB_SIZE);
 689         spin_lock_init(&ipvs->tcp_app_lock);
 690         pd->timeout_table = ip_vs_create_timeout_table((int *)tcp_timeouts,
 691                                                         sizeof(tcp_timeouts));
 692         if (!pd->timeout_table)
 693                 return -ENOMEM;
 694         pd->tcp_state_table =  tcp_states;
 695         return 0;
 696 }
 697
 698 static void __ip_vs_tcp_exit(struct net *net, struct ip_vs_proto_data *pd)
 699 {
 700         kfree(pd->timeout_table);
 701 }
 702
 703
 704 struct ip_vs_protocol ip_vs_protocol_tcp = {
 705         .name =                 "TCP",
 706         .protocol =             IPPROTO_TCP,
 707         .num_states =           IP_VS_TCP_S_LAST,
 708         .dont_defrag =          0,
 709         .init =                 NULL,
 710         .exit =                 NULL,
 711         .init_netns =           __ip_vs_tcp_init,
 712         .exit_netns =           __ip_vs_tcp_exit,
 713         .register_app =         tcp_register_app,
 714         .unregister_app =       tcp_unregister_app,
 715         .conn_schedule =        tcp_conn_schedule,
 716         .conn_in_get =          ip_vs_conn_in_get_proto,
 717         .conn_out_get =         ip_vs_conn_out_get_proto,
 718         .snat_handler =         tcp_snat_handler,
 719         .dnat_handler =         tcp_dnat_handler,
 720         .csum_check =           tcp_csum_check,
 721         .state_name =           tcp_state_name,
 722         .state_transition =     tcp_state_transition,
 723         .app_conn_bind =        tcp_app_conn_bind,
 724         .debug_packet =         ip_vs_tcpudp_debug_packet,
 725         .timeout_change =       tcp_timeout_change,
 726 };