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(struct netns_ipvs *ipvs, int af, struct sk_buff *skb,
  36                   struct ip_vs_proto_data *pd,
  37                   int *verdict, struct ip_vs_conn **cpp,
  38                   struct ip_vs_iphdr *iph)
  39 {
  40         struct ip_vs_service *svc;
  41         struct tcphdr _tcph, *th;
  42         __be16 _ports[2], *ports = NULL;
  43
  44         /* In the event of icmp, we're only guaranteed to have the first 8
  45          * bytes of the transport header, so we only check the rest of the
  46          * TCP packet for non-ICMP packets
  47          */
  48         if (likely(!ip_vs_iph_icmp(iph))) {
  49                 th = skb_header_pointer(skb, iph->len, sizeof(_tcph), &_tcph);
  50                 if (th) {
  51                         if (th->rst || !(sysctl_sloppy_tcp(ipvs) || th->syn))
  52                                 return 1;
  53                         ports = &th->source;
  54                 }
  55         } else {
  56                 ports = skb_header_pointer(
  57                         skb, iph->len, sizeof(_ports), &_ports);
  58         }
  59
  60         if (!ports) {
  61                 *verdict = NF_DROP;
  62                 return 0;
  63         }
  64
  65         /* No !th->ack check to allow scheduling on SYN+ACK for Active FTP */
  66         rcu_read_lock();
  67
  68         if (likely(!ip_vs_iph_inverse(iph)))
  69                 svc = ip_vs_service_find(ipvs, af, skb->mark, iph->protocol,
  70                                          &iph->daddr, ports[1]);
  71         else
  72                 svc = ip_vs_service_find(ipvs, af, skb->mark, iph->protocol,
  73                                          &iph->saddr, ports[0]);
  74
  75         if (svc) {
  76                 int ignored;
  77
  78                 if (ip_vs_todrop(ipvs)) {
  79                         /*
  80                          * It seems that we are very loaded.
  81                          * We have to drop this packet :(
  82                          */
  83                         rcu_read_unlock();
  84                         *verdict = NF_DROP;
  85                         return 0;
  86                 }
  87
  88                 /*
  89                  * Let the virtual server select a real server for the
  90                  * incoming connection, and create a connection entry.
  91                  */
  92                 *cpp = ip_vs_schedule(svc, skb, pd, &ignored, iph);
  93                 if (!*cpp && ignored <= 0) {
  94                         if (!ignored)
  95                                 *verdict = ip_vs_leave(svc, skb, pd, iph);
  96                         else
  97                                 *verdict = NF_DROP;
  98                         rcu_read_unlock();
  99                         return 0;
 100                 }
 101         }
 102         rcu_read_unlock();
 103         /* NF_ACCEPT */
 104         return 1;
 105 }
 106
 107
 108 static inline void
 109 tcp_fast_csum_update(int af, struct tcphdr *tcph,
 110                      const union nf_inet_addr *oldip,
 111                      const union nf_inet_addr *newip,
 112                      __be16 oldport, __be16 newport)
 113 {
 114 #ifdef CONFIG_IP_VS_IPV6
 115         if (af == AF_INET6)
 116                 tcph->check =
 117                         csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
 118                                          ip_vs_check_diff2(oldport, newport,
 119                                                 ~csum_unfold(tcph->check))));
 120         else
 121 #endif
 122         tcph->check =
 123                 csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
 124                                  ip_vs_check_diff2(oldport, newport,
 125                                                 ~csum_unfold(tcph->check))));
 126 }
 127
 128
 129 static inline void
 130 tcp_partial_csum_update(int af, struct tcphdr *tcph,
 131                      const union nf_inet_addr *oldip,
 132                      const union nf_inet_addr *newip,
 133                      __be16 oldlen, __be16 newlen)
 134 {
 135 #ifdef CONFIG_IP_VS_IPV6
 136         if (af == AF_INET6)
 137                 tcph->check =
 138                         ~csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
 139                                          ip_vs_check_diff2(oldlen, newlen,
 140                                                 csum_unfold(tcph->check))));
 141         else
 142 #endif
 143         tcph->check =
 144                 ~csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
 145                                 ip_vs_check_diff2(oldlen, newlen,
 146                                                 csum_unfold(tcph->check))));
 147 }
 148
 149
 150 static int
 151 tcp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
 152                  struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
 153 {
 154         struct tcphdr *tcph;
 155         unsigned int tcphoff = iph->len;
 156         int oldlen;
 157         int payload_csum = 0;
 158
 159 #ifdef CONFIG_IP_VS_IPV6
 160         if (cp->af == AF_INET6 && iph->fragoffs)
 161                 return 1;
 162 #endif
 163         oldlen = skb->len - tcphoff;
 164
 165         /* csum_check requires unshared skb */
 166         if (!skb_make_writable(skb, tcphoff+sizeof(*tcph)))
 167                 return 0;
 168
 169         if (unlikely(cp->app != NULL)) {
 170                 int ret;
 171
 172                 /* Some checks before mangling */
 173                 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
 174                         return 0;
 175
 176                 /* Call application helper if needed */
 177                 if (!(ret = ip_vs_app_pkt_out(cp, skb)))
 178                         return 0;
 179                 /* ret=2: csum update is needed after payload mangling */
 180                 if (ret == 1)
 181                         oldlen = skb->len - tcphoff;
 182                 else
 183                         payload_csum = 1;
 184         }
 185
 186         tcph = (void *)skb_network_header(skb) + tcphoff;
 187         tcph->source = cp->vport;
 188
 189         /* Adjust TCP checksums */
 190         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 191                 tcp_partial_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr,
 192                                         htons(oldlen),
 193                                         htons(skb->len - tcphoff));
 194         } else if (!payload_csum) {
 195                 /* Only port and addr are changed, do fast csum update */
 196                 tcp_fast_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr,
 197                                      cp->dport, cp->vport);
 198                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 199                         skb->ip_summed = (cp->app && pp->csum_check) ?
 200                                          CHECKSUM_UNNECESSARY : CHECKSUM_NONE;
 201         } else {
 202                 /* full checksum calculation */
 203                 tcph->check = 0;
 204                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 205 #ifdef CONFIG_IP_VS_IPV6
 206                 if (cp->af == AF_INET6)
 207                         tcph->check = csum_ipv6_magic(&cp->vaddr.in6,
 208                                                       &cp->caddr.in6,
 209                                                       skb->len - tcphoff,
 210                                                       cp->protocol, skb->csum);
 211                 else
 212 #endif
 213                         tcph->check = csum_tcpudp_magic(cp->vaddr.ip,
 214                                                         cp->caddr.ip,
 215                                                         skb->len - tcphoff,
 216                                                         cp->protocol,
 217                                                         skb->csum);
 218                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 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, struct ip_vs_protocol *pp,
 230                  struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
 231 {
 232         struct tcphdr *tcph;
 233         unsigned int tcphoff = iph->len;
 234         int oldlen;
 235         int payload_csum = 0;
 236
 237 #ifdef CONFIG_IP_VS_IPV6
 238         if (cp->af == AF_INET6 && iph->fragoffs)
 239                 return 1;
 240 #endif
 241         oldlen = skb->len - tcphoff;
 242
 243         /* csum_check requires unshared skb */
 244         if (!skb_make_writable(skb, tcphoff+sizeof(*tcph)))
 245                 return 0;
 246
 247         if (unlikely(cp->app != NULL)) {
 248                 int ret;
 249
 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 (!(ret = ip_vs_app_pkt_in(cp, skb)))
 259                         return 0;
 260                 /* ret=2: csum update is needed after payload mangling */
 261                 if (ret == 1)
 262                         oldlen = skb->len - tcphoff;
 263                 else
 264                         payload_csum = 1;
 265         }
 266
 267         tcph = (void *)skb_network_header(skb) + tcphoff;
 268         tcph->dest = cp->dport;
 269
 270         /*
 271          *      Adjust TCP checksums
 272          */
 273         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 274                 tcp_partial_csum_update(cp->af, tcph, &cp->vaddr, &cp->daddr,
 275                                         htons(oldlen),
 276                                         htons(skb->len - tcphoff));
 277         } else if (!payload_csum) {
 278                 /* Only port and addr are changed, do fast csum update */
 279                 tcp_fast_csum_update(cp->af, tcph, &cp->vaddr, &cp->daddr,
 280                                      cp->vport, cp->dport);
 281                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 282                         skb->ip_summed = (cp->app && pp->csum_check) ?
 283                                          CHECKSUM_UNNECESSARY : CHECKSUM_NONE;
 284         } else {
 285                 /* full checksum calculation */
 286                 tcph->check = 0;
 287                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 288 #ifdef CONFIG_IP_VS_IPV6
 289                 if (cp->af == AF_INET6)
 290                         tcph->check = csum_ipv6_magic(&cp->caddr.in6,
 291                                                       &cp->daddr.in6,
 292                                                       skb->len - tcphoff,
 293                                                       cp->protocol, skb->csum);
 294                 else
 295 #endif
 296                         tcph->check = csum_tcpudp_magic(cp->caddr.ip,
 297                                                         cp->daddr.ip,
 298                                                         skb->len - tcphoff,
 299                                                         cp->protocol,
 300                                                         skb->csum);
 301                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 302         }
 303         return 1;
 304 }
 305
 306
 307 static int
 308 tcp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp)
 309 {
 310         unsigned int tcphoff;
 311
 312 #ifdef CONFIG_IP_VS_IPV6
 313         if (af == AF_INET6)
 314                 tcphoff = sizeof(struct ipv6hdr);
 315         else
 316 #endif
 317                 tcphoff = ip_hdrlen(skb);
 318
 319         switch (skb->ip_summed) {
 320         case CHECKSUM_NONE:
 321                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 322         case CHECKSUM_COMPLETE:
 323 #ifdef CONFIG_IP_VS_IPV6
 324                 if (af == AF_INET6) {
 325                         if (csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 326                                             &ipv6_hdr(skb)->daddr,
 327                                             skb->len - tcphoff,
 328                                             ipv6_hdr(skb)->nexthdr,
 329                                             skb->csum)) {
 330                                 IP_VS_DBG_RL_PKT(0, af, pp, skb, 0,
 331                                                  "Failed checksum for");
 332                                 return 0;
 333                         }
 334                 } else
 335 #endif
 336                         if (csum_tcpudp_magic(ip_hdr(skb)->saddr,
 337                                               ip_hdr(skb)->daddr,
 338                                               skb->len - tcphoff,
 339                                               ip_hdr(skb)->protocol,
 340                                               skb->csum)) {
 341                                 IP_VS_DBG_RL_PKT(0, af, pp, skb, 0,
 342                                                  "Failed checksum for");
 343                                 return 0;
 344                         }
 345                 break;
 346         default:
 347                 /* No need to checksum. */
 348                 break;
 349         }
 350
 351         return 1;
 352 }
 353
 354
 355 #define TCP_DIR_INPUT           0
 356 #define TCP_DIR_OUTPUT          4
 357 #define TCP_DIR_INPUT_ONLY      8
 358
 359 static const int tcp_state_off[IP_VS_DIR_LAST] = {
 360         [IP_VS_DIR_INPUT]               =       TCP_DIR_INPUT,
 361         [IP_VS_DIR_OUTPUT]              =       TCP_DIR_OUTPUT,
 362         [IP_VS_DIR_INPUT_ONLY]          =       TCP_DIR_INPUT_ONLY,
 363 };
 364
 365 /*
 366  *      Timeout table[state]
 367  */
 368 static const int tcp_timeouts[IP_VS_TCP_S_LAST+1] = {
 369         [IP_VS_TCP_S_NONE]              =       2*HZ,
 370         [IP_VS_TCP_S_ESTABLISHED]       =       15*60*HZ,
 371         [IP_VS_TCP_S_SYN_SENT]          =       2*60*HZ,
 372         [IP_VS_TCP_S_SYN_RECV]          =       1*60*HZ,
 373         [IP_VS_TCP_S_FIN_WAIT]          =       2*60*HZ,
 374         [IP_VS_TCP_S_TIME_WAIT]         =       2*60*HZ,
 375         [IP_VS_TCP_S_CLOSE]             =       10*HZ,
 376         [IP_VS_TCP_S_CLOSE_WAIT]        =       60*HZ,
 377         [IP_VS_TCP_S_LAST_ACK]          =       30*HZ,
 378         [IP_VS_TCP_S_LISTEN]            =       2*60*HZ,
 379         [IP_VS_TCP_S_SYNACK]            =       120*HZ,
 380         [IP_VS_TCP_S_LAST]              =       2*HZ,
 381 };
 382
 383 static const char *const tcp_state_name_table[IP_VS_TCP_S_LAST+1] = {
 384         [IP_VS_TCP_S_NONE]              =       "NONE",
 385         [IP_VS_TCP_S_ESTABLISHED]       =       "ESTABLISHED",
 386         [IP_VS_TCP_S_SYN_SENT]          =       "SYN_SENT",
 387         [IP_VS_TCP_S_SYN_RECV]          =       "SYN_RECV",
 388         [IP_VS_TCP_S_FIN_WAIT]          =       "FIN_WAIT",
 389         [IP_VS_TCP_S_TIME_WAIT]         =       "TIME_WAIT",
 390         [IP_VS_TCP_S_CLOSE]             =       "CLOSE",
 391         [IP_VS_TCP_S_CLOSE_WAIT]        =       "CLOSE_WAIT",
 392         [IP_VS_TCP_S_LAST_ACK]          =       "LAST_ACK",
 393         [IP_VS_TCP_S_LISTEN]            =       "LISTEN",
 394         [IP_VS_TCP_S_SYNACK]            =       "SYNACK",
 395         [IP_VS_TCP_S_LAST]              =       "BUG!",
 396 };
 397
 398 static const bool tcp_state_active_table[IP_VS_TCP_S_LAST] = {
 399         [IP_VS_TCP_S_NONE]              =       false,
 400         [IP_VS_TCP_S_ESTABLISHED]       =       true,
 401         [IP_VS_TCP_S_SYN_SENT]          =       true,
 402         [IP_VS_TCP_S_SYN_RECV]          =       true,
 403         [IP_VS_TCP_S_FIN_WAIT]          =       false,
 404         [IP_VS_TCP_S_TIME_WAIT]         =       false,
 405         [IP_VS_TCP_S_CLOSE]             =       false,
 406         [IP_VS_TCP_S_CLOSE_WAIT]        =       false,
 407         [IP_VS_TCP_S_LAST_ACK]          =       false,
 408         [IP_VS_TCP_S_LISTEN]            =       false,
 409         [IP_VS_TCP_S_SYNACK]            =       true,
 410 };
 411
 412 #define sNO IP_VS_TCP_S_NONE
 413 #define sES IP_VS_TCP_S_ESTABLISHED
 414 #define sSS IP_VS_TCP_S_SYN_SENT
 415 #define sSR IP_VS_TCP_S_SYN_RECV
 416 #define sFW IP_VS_TCP_S_FIN_WAIT
 417 #define sTW IP_VS_TCP_S_TIME_WAIT
 418 #define sCL IP_VS_TCP_S_CLOSE
 419 #define sCW IP_VS_TCP_S_CLOSE_WAIT
 420 #define sLA IP_VS_TCP_S_LAST_ACK
 421 #define sLI IP_VS_TCP_S_LISTEN
 422 #define sSA IP_VS_TCP_S_SYNACK
 423
 424 struct tcp_states_t {
 425         int next_state[IP_VS_TCP_S_LAST];
 426 };
 427
 428 static const char * tcp_state_name(int state)
 429 {
 430         if (state >= IP_VS_TCP_S_LAST)
 431                 return "ERR!";
 432         return tcp_state_name_table[state] ? tcp_state_name_table[state] : "?";
 433 }
 434
 435 static bool tcp_state_active(int state)
 436 {
 437         if (state >= IP_VS_TCP_S_LAST)
 438                 return false;
 439         return tcp_state_active_table[state];
 440 }
 441
 442 static struct tcp_states_t tcp_states [] = {
 443 /*      INPUT */
 444 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 445 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }},
 446 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sTW }},
 447 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 448 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sSR }},
 449
 450 /*      OUTPUT */
 451 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 452 /*syn*/ {{sSS, sES, sSS, sSR, sSS, sSS, sSS, sSS, sSS, sLI, sSR }},
 453 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }},
 454 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }},
 455 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }},
 456
 457 /*      INPUT-ONLY */
 458 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 459 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }},
 460 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }},
 461 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 462 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 463 };
 464
 465 static struct tcp_states_t tcp_states_dos [] = {
 466 /*      INPUT */
 467 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 468 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSA }},
 469 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sSA }},
 470 /*ack*/ {{sES, sES, sSS, sSR, sFW, sTW, sCL, sCW, sCL, sLI, sSA }},
 471 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 472
 473 /*      OUTPUT */
 474 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 475 /*syn*/ {{sSS, sES, sSS, sSA, sSS, sSS, sSS, sSS, sSS, sLI, sSA }},
 476 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }},
 477 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }},
 478 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }},
 479
 480 /*      INPUT-ONLY */
 481 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 482 /*syn*/ {{sSA, sES, sES, sSR, sSA, sSA, sSA, sSA, sSA, sSA, sSA }},
 483 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }},
 484 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 485 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 486 };
 487
 488 static void tcp_timeout_change(struct ip_vs_proto_data *pd, int flags)
 489 {
 490         int on = (flags & 1);           /* secure_tcp */
 491
 492         /*
 493         ** FIXME: change secure_tcp to independent sysctl var
 494         ** or make it per-service or per-app because it is valid
 495         ** for most if not for all of the applications. Something
 496         ** like "capabilities" (flags) for each object.
 497         */
 498         pd->tcp_state_table = (on ? tcp_states_dos : tcp_states);
 499 }
 500
 501 static inline int tcp_state_idx(struct tcphdr *th)
 502 {
 503         if (th->rst)
 504                 return 3;
 505         if (th->syn)
 506                 return 0;
 507         if (th->fin)
 508                 return 1;
 509         if (th->ack)
 510                 return 2;
 511         return -1;
 512 }
 513
 514 static inline void
 515 set_tcp_state(struct ip_vs_proto_data *pd, struct ip_vs_conn *cp,
 516               int direction, struct tcphdr *th)
 517 {
 518         int state_idx;
 519         int new_state = IP_VS_TCP_S_CLOSE;
 520         int state_off = tcp_state_off[direction];
 521
 522         /*
 523          *    Update state offset to INPUT_ONLY if necessary
 524          *    or delete NO_OUTPUT flag if output packet detected
 525          */
 526         if (cp->flags & IP_VS_CONN_F_NOOUTPUT) {
 527                 if (state_off == TCP_DIR_OUTPUT)
 528                         cp->flags &= ~IP_VS_CONN_F_NOOUTPUT;
 529                 else
 530                         state_off = TCP_DIR_INPUT_ONLY;
 531         }
 532
 533         if ((state_idx = tcp_state_idx(th)) < 0) {
 534                 IP_VS_DBG(8, "tcp_state_idx=%d!!!\n", state_idx);
 535                 goto tcp_state_out;
 536         }
 537
 538         new_state =
 539                 pd->tcp_state_table[state_off+state_idx].next_state[cp->state];
 540
 541   tcp_state_out:
 542         if (new_state != cp->state) {
 543                 struct ip_vs_dest *dest = cp->dest;
 544
 545                 IP_VS_DBG_BUF(8, "%s %s [%c%c%c%c] %s:%d->"
 546                               "%s:%d state: %s->%s conn->refcnt:%d\n",
 547                               pd->pp->name,
 548                               ((state_off == TCP_DIR_OUTPUT) ?
 549                                "output " : "input "),
 550                               th->syn ? 'S' : '.',
 551                               th->fin ? 'F' : '.',
 552                               th->ack ? 'A' : '.',
 553                               th->rst ? 'R' : '.',
 554                               IP_VS_DBG_ADDR(cp->daf, &cp->daddr),
 555                               ntohs(cp->dport),
 556                               IP_VS_DBG_ADDR(cp->af, &cp->caddr),
 557                               ntohs(cp->cport),
 558                               tcp_state_name(cp->state),
 559                               tcp_state_name(new_state),
 560                               atomic_read(&cp->refcnt));
 561
 562                 if (dest) {
 563                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 564                             !tcp_state_active(new_state)) {
 565                                 atomic_dec(&dest->activeconns);
 566                                 atomic_inc(&dest->inactconns);
 567                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
 568                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
 569                                    tcp_state_active(new_state)) {
 570                                 atomic_inc(&dest->activeconns);
 571                                 atomic_dec(&dest->inactconns);
 572                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 573                         }
 574                 }
 575         }
 576
 577         if (likely(pd))
 578                 cp->timeout = pd->timeout_table[cp->state = new_state];
 579         else    /* What to do ? */
 580                 cp->timeout = tcp_timeouts[cp->state = new_state];
 581 }
 582
 583 /*
 584  *      Handle state transitions
 585  */
 586 static void
 587 tcp_state_transition(struct ip_vs_conn *cp, int direction,
 588                      const struct sk_buff *skb,
 589                      struct ip_vs_proto_data *pd)
 590 {
 591         struct tcphdr _tcph, *th;
 592
 593 #ifdef CONFIG_IP_VS_IPV6
 594         int ihl = cp->af == AF_INET ? ip_hdrlen(skb) : sizeof(struct ipv6hdr);
 595 #else
 596         int ihl = ip_hdrlen(skb);
 597 #endif
 598
 599         th = skb_header_pointer(skb, ihl, sizeof(_tcph), &_tcph);
 600         if (th == NULL)
 601                 return;
 602
 603         spin_lock_bh(&cp->lock);
 604         set_tcp_state(pd, cp, direction, th);
 605         spin_unlock_bh(&cp->lock);
 606 }
 607
 608 static inline __u16 tcp_app_hashkey(__be16 port)
 609 {
 610         return (((__force u16)port >> TCP_APP_TAB_BITS) ^ (__force u16)port)
 611                 & TCP_APP_TAB_MASK;
 612 }
 613
 614
 615 static int tcp_register_app(struct netns_ipvs *ipvs, struct ip_vs_app *inc)
 616 {
 617         struct ip_vs_app *i;
 618         __u16 hash;
 619         __be16 port = inc->port;
 620         int ret = 0;
 621         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs, IPPROTO_TCP);
 622
 623         hash = tcp_app_hashkey(port);
 624
 625         list_for_each_entry(i, &ipvs->tcp_apps[hash], p_list) {
 626                 if (i->port == port) {
 627                         ret = -EEXIST;
 628                         goto out;
 629                 }
 630         }
 631         list_add_rcu(&inc->p_list, &ipvs->tcp_apps[hash]);
 632         atomic_inc(&pd->appcnt);
 633
 634   out:
 635         return ret;
 636 }
 637
 638
 639 static void
 640 tcp_unregister_app(struct netns_ipvs *ipvs, struct ip_vs_app *inc)
 641 {
 642         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs, IPPROTO_TCP);
 643
 644         atomic_dec(&pd->appcnt);
 645         list_del_rcu(&inc->p_list);
 646 }
 647
 648
 649 static int
 650 tcp_app_conn_bind(struct ip_vs_conn *cp)
 651 {
 652         struct netns_ipvs *ipvs = cp->ipvs;
 653         int hash;
 654         struct ip_vs_app *inc;
 655         int result = 0;
 656
 657         /* Default binding: bind app only for NAT */
 658         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
 659                 return 0;
 660
 661         /* Lookup application incarnations and bind the right one */
 662         hash = tcp_app_hashkey(cp->vport);
 663
 664         rcu_read_lock();
 665         list_for_each_entry_rcu(inc, &ipvs->tcp_apps[hash], p_list) {
 666                 if (inc->port == cp->vport) {
 667                         if (unlikely(!ip_vs_app_inc_get(inc)))
 668                                 break;
 669                         rcu_read_unlock();
 670
 671                         IP_VS_DBG_BUF(9, "%s(): Binding conn %s:%u->"
 672                                       "%s:%u to app %s on port %u\n",
 673                                       __func__,
 674                                       IP_VS_DBG_ADDR(cp->af, &cp->caddr),
 675                                       ntohs(cp->cport),
 676                                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
 677                                       ntohs(cp->vport),
 678                                       inc->name, ntohs(inc->port));
 679
 680                         cp->app = inc;
 681                         if (inc->init_conn)
 682                                 result = inc->init_conn(inc, cp);
 683                         goto out;
 684                 }
 685         }
 686         rcu_read_unlock();
 687
 688   out:
 689         return result;
 690 }
 691
 692
 693 /*
 694  *      Set LISTEN timeout. (ip_vs_conn_put will setup timer)
 695  */
 696 void ip_vs_tcp_conn_listen(struct ip_vs_conn *cp)
 697 {
 698         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(cp->ipvs, IPPROTO_TCP);
 699
 700         spin_lock_bh(&cp->lock);
 701         cp->state = IP_VS_TCP_S_LISTEN;
 702         cp->timeout = (pd ? pd->timeout_table[IP_VS_TCP_S_LISTEN]
 703                            : tcp_timeouts[IP_VS_TCP_S_LISTEN]);
 704         spin_unlock_bh(&cp->lock);
 705 }
 706
 707 /* ---------------------------------------------
 708  *   timeouts is netns related now.
 709  * ---------------------------------------------
 710  */
 711 static int __ip_vs_tcp_init(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd)
 712 {
 713         ip_vs_init_hash_table(ipvs->tcp_apps, TCP_APP_TAB_SIZE);
 714         pd->timeout_table = ip_vs_create_timeout_table((int *)tcp_timeouts,
 715                                                         sizeof(tcp_timeouts));
 716         if (!pd->timeout_table)
 717                 return -ENOMEM;
 718         pd->tcp_state_table =  tcp_states;
 719         return 0;
 720 }
 721
 722 static void __ip_vs_tcp_exit(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd)
 723 {
 724         kfree(pd->timeout_table);
 725 }
 726
 727
 728 struct ip_vs_protocol ip_vs_protocol_tcp = {
 729         .name =                 "TCP",
 730         .protocol =             IPPROTO_TCP,
 731         .num_states =           IP_VS_TCP_S_LAST,
 732         .dont_defrag =          0,
 733         .init =                 NULL,
 734         .exit =                 NULL,
 735         .init_netns =           __ip_vs_tcp_init,
 736         .exit_netns =           __ip_vs_tcp_exit,
 737         .register_app =         tcp_register_app,
 738         .unregister_app =       tcp_unregister_app,
 739         .conn_schedule =        tcp_conn_schedule,
 740         .conn_in_get =          ip_vs_conn_in_get_proto,
 741         .conn_out_get =         ip_vs_conn_out_get_proto,
 742         .snat_handler =         tcp_snat_handler,
 743         .dnat_handler =         tcp_dnat_handler,
 744         .csum_check =           tcp_csum_check,
 745         .state_name =           tcp_state_name,
 746         .state_transition =     tcp_state_transition,
 747         .app_conn_bind =        tcp_app_conn_bind,
 748         .debug_packet =         ip_vs_tcpudp_debug_packet,
 749         .timeout_change =       tcp_timeout_change,
 750 };