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
  67         if (likely(!ip_vs_iph_inverse(iph)))
  68                 svc = ip_vs_service_find(ipvs, af, skb->mark, iph->protocol,
  69                                          &iph->daddr, ports[1]);
  70         else
  71                 svc = ip_vs_service_find(ipvs, af, skb->mark, iph->protocol,
  72                                          &iph->saddr, ports[0]);
  73
  74         if (svc) {
  75                 int ignored;
  76
  77                 if (ip_vs_todrop(ipvs)) {
  78                         /*
  79                          * It seems that we are very loaded.
  80                          * We have to drop this packet :(
  81                          */
  82                         *verdict = NF_DROP;
  83                         return 0;
  84                 }
  85
  86                 /*
  87                  * Let the virtual server select a real server for the
  88                  * incoming connection, and create a connection entry.
  89                  */
  90                 *cpp = ip_vs_schedule(svc, skb, pd, &ignored, iph);
  91                 if (!*cpp && ignored <= 0) {
  92                         if (!ignored)
  93                                 *verdict = ip_vs_leave(svc, skb, pd, iph);
  94                         else
  95                                 *verdict = NF_DROP;
  96                         return 0;
  97                 }
  98         }
  99         /* NF_ACCEPT */
 100         return 1;
 101 }
 102
 103
 104 static inline void
 105 tcp_fast_csum_update(int af, struct tcphdr *tcph,
 106                      const union nf_inet_addr *oldip,
 107                      const union nf_inet_addr *newip,
 108                      __be16 oldport, __be16 newport)
 109 {
 110 #ifdef CONFIG_IP_VS_IPV6
 111         if (af == AF_INET6)
 112                 tcph->check =
 113                         csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
 114                                          ip_vs_check_diff2(oldport, newport,
 115                                                 ~csum_unfold(tcph->check))));
 116         else
 117 #endif
 118         tcph->check =
 119                 csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
 120                                  ip_vs_check_diff2(oldport, newport,
 121                                                 ~csum_unfold(tcph->check))));
 122 }
 123
 124
 125 static inline void
 126 tcp_partial_csum_update(int af, struct tcphdr *tcph,
 127                      const union nf_inet_addr *oldip,
 128                      const union nf_inet_addr *newip,
 129                      __be16 oldlen, __be16 newlen)
 130 {
 131 #ifdef CONFIG_IP_VS_IPV6
 132         if (af == AF_INET6)
 133                 tcph->check =
 134                         ~csum_fold(ip_vs_check_diff16(oldip->ip6, newip->ip6,
 135                                          ip_vs_check_diff2(oldlen, newlen,
 136                                                 csum_unfold(tcph->check))));
 137         else
 138 #endif
 139         tcph->check =
 140                 ~csum_fold(ip_vs_check_diff4(oldip->ip, newip->ip,
 141                                 ip_vs_check_diff2(oldlen, newlen,
 142                                                 csum_unfold(tcph->check))));
 143 }
 144
 145
 146 static int
 147 tcp_snat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
 148                  struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
 149 {
 150         struct tcphdr *tcph;
 151         unsigned int tcphoff = iph->len;
 152         int oldlen;
 153         int payload_csum = 0;
 154
 155 #ifdef CONFIG_IP_VS_IPV6
 156         if (cp->af == AF_INET6 && iph->fragoffs)
 157                 return 1;
 158 #endif
 159         oldlen = skb->len - tcphoff;
 160
 161         /* csum_check requires unshared skb */
 162         if (!skb_make_writable(skb, tcphoff+sizeof(*tcph)))
 163                 return 0;
 164
 165         if (unlikely(cp->app != NULL)) {
 166                 int ret;
 167
 168                 /* Some checks before mangling */
 169                 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
 170                         return 0;
 171
 172                 /* Call application helper if needed */
 173                 if (!(ret = ip_vs_app_pkt_out(cp, skb)))
 174                         return 0;
 175                 /* ret=2: csum update is needed after payload mangling */
 176                 if (ret == 1)
 177                         oldlen = skb->len - tcphoff;
 178                 else
 179                         payload_csum = 1;
 180         }
 181
 182         tcph = (void *)skb_network_header(skb) + tcphoff;
 183         tcph->source = cp->vport;
 184
 185         /* Adjust TCP checksums */
 186         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 187                 tcp_partial_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr,
 188                                         htons(oldlen),
 189                                         htons(skb->len - tcphoff));
 190         } else if (!payload_csum) {
 191                 /* Only port and addr are changed, do fast csum update */
 192                 tcp_fast_csum_update(cp->af, tcph, &cp->daddr, &cp->vaddr,
 193                                      cp->dport, cp->vport);
 194                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 195                         skb->ip_summed = (cp->app && pp->csum_check) ?
 196                                          CHECKSUM_UNNECESSARY : CHECKSUM_NONE;
 197         } else {
 198                 /* full checksum calculation */
 199                 tcph->check = 0;
 200                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 201 #ifdef CONFIG_IP_VS_IPV6
 202                 if (cp->af == AF_INET6)
 203                         tcph->check = csum_ipv6_magic(&cp->vaddr.in6,
 204                                                       &cp->caddr.in6,
 205                                                       skb->len - tcphoff,
 206                                                       cp->protocol, skb->csum);
 207                 else
 208 #endif
 209                         tcph->check = csum_tcpudp_magic(cp->vaddr.ip,
 210                                                         cp->caddr.ip,
 211                                                         skb->len - tcphoff,
 212                                                         cp->protocol,
 213                                                         skb->csum);
 214                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 215
 216                 IP_VS_DBG(11, "O-pkt: %s O-csum=%d (+%zd)\n",
 217                           pp->name, tcph->check,
 218                           (char*)&(tcph->check) - (char*)tcph);
 219         }
 220         return 1;
 221 }
 222
 223
 224 static int
 225 tcp_dnat_handler(struct sk_buff *skb, struct ip_vs_protocol *pp,
 226                  struct ip_vs_conn *cp, struct ip_vs_iphdr *iph)
 227 {
 228         struct tcphdr *tcph;
 229         unsigned int tcphoff = iph->len;
 230         int oldlen;
 231         int payload_csum = 0;
 232
 233 #ifdef CONFIG_IP_VS_IPV6
 234         if (cp->af == AF_INET6 && iph->fragoffs)
 235                 return 1;
 236 #endif
 237         oldlen = skb->len - tcphoff;
 238
 239         /* csum_check requires unshared skb */
 240         if (!skb_make_writable(skb, tcphoff+sizeof(*tcph)))
 241                 return 0;
 242
 243         if (unlikely(cp->app != NULL)) {
 244                 int ret;
 245
 246                 /* Some checks before mangling */
 247                 if (pp->csum_check && !pp->csum_check(cp->af, skb, pp))
 248                         return 0;
 249
 250                 /*
 251                  *      Attempt ip_vs_app call.
 252                  *      It will fix ip_vs_conn and iph ack_seq stuff
 253                  */
 254                 if (!(ret = ip_vs_app_pkt_in(cp, skb)))
 255                         return 0;
 256                 /* ret=2: csum update is needed after payload mangling */
 257                 if (ret == 1)
 258                         oldlen = skb->len - tcphoff;
 259                 else
 260                         payload_csum = 1;
 261         }
 262
 263         tcph = (void *)skb_network_header(skb) + tcphoff;
 264         tcph->dest = cp->dport;
 265
 266         /*
 267          *      Adjust TCP checksums
 268          */
 269         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 270                 tcp_partial_csum_update(cp->af, tcph, &cp->vaddr, &cp->daddr,
 271                                         htons(oldlen),
 272                                         htons(skb->len - tcphoff));
 273         } else if (!payload_csum) {
 274                 /* Only port and addr are changed, do fast csum update */
 275                 tcp_fast_csum_update(cp->af, tcph, &cp->vaddr, &cp->daddr,
 276                                      cp->vport, cp->dport);
 277                 if (skb->ip_summed == CHECKSUM_COMPLETE)
 278                         skb->ip_summed = (cp->app && pp->csum_check) ?
 279                                          CHECKSUM_UNNECESSARY : CHECKSUM_NONE;
 280         } else {
 281                 /* full checksum calculation */
 282                 tcph->check = 0;
 283                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 284 #ifdef CONFIG_IP_VS_IPV6
 285                 if (cp->af == AF_INET6)
 286                         tcph->check = csum_ipv6_magic(&cp->caddr.in6,
 287                                                       &cp->daddr.in6,
 288                                                       skb->len - tcphoff,
 289                                                       cp->protocol, skb->csum);
 290                 else
 291 #endif
 292                         tcph->check = csum_tcpudp_magic(cp->caddr.ip,
 293                                                         cp->daddr.ip,
 294                                                         skb->len - tcphoff,
 295                                                         cp->protocol,
 296                                                         skb->csum);
 297                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 298         }
 299         return 1;
 300 }
 301
 302
 303 static int
 304 tcp_csum_check(int af, struct sk_buff *skb, struct ip_vs_protocol *pp)
 305 {
 306         unsigned int tcphoff;
 307
 308 #ifdef CONFIG_IP_VS_IPV6
 309         if (af == AF_INET6)
 310                 tcphoff = sizeof(struct ipv6hdr);
 311         else
 312 #endif
 313                 tcphoff = ip_hdrlen(skb);
 314
 315         switch (skb->ip_summed) {
 316         case CHECKSUM_NONE:
 317                 skb->csum = skb_checksum(skb, tcphoff, skb->len - tcphoff, 0);
 318         case CHECKSUM_COMPLETE:
 319 #ifdef CONFIG_IP_VS_IPV6
 320                 if (af == AF_INET6) {
 321                         if (csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 322                                             &ipv6_hdr(skb)->daddr,
 323                                             skb->len - tcphoff,
 324                                             ipv6_hdr(skb)->nexthdr,
 325                                             skb->csum)) {
 326                                 IP_VS_DBG_RL_PKT(0, af, pp, skb, 0,
 327                                                  "Failed checksum for");
 328                                 return 0;
 329                         }
 330                 } else
 331 #endif
 332                         if (csum_tcpudp_magic(ip_hdr(skb)->saddr,
 333                                               ip_hdr(skb)->daddr,
 334                                               skb->len - tcphoff,
 335                                               ip_hdr(skb)->protocol,
 336                                               skb->csum)) {
 337                                 IP_VS_DBG_RL_PKT(0, af, pp, skb, 0,
 338                                                  "Failed checksum for");
 339                                 return 0;
 340                         }
 341                 break;
 342         default:
 343                 /* No need to checksum. */
 344                 break;
 345         }
 346
 347         return 1;
 348 }
 349
 350
 351 #define TCP_DIR_INPUT           0
 352 #define TCP_DIR_OUTPUT          4
 353 #define TCP_DIR_INPUT_ONLY      8
 354
 355 static const int tcp_state_off[IP_VS_DIR_LAST] = {
 356         [IP_VS_DIR_INPUT]               =       TCP_DIR_INPUT,
 357         [IP_VS_DIR_OUTPUT]              =       TCP_DIR_OUTPUT,
 358         [IP_VS_DIR_INPUT_ONLY]          =       TCP_DIR_INPUT_ONLY,
 359 };
 360
 361 /*
 362  *      Timeout table[state]
 363  */
 364 static const int tcp_timeouts[IP_VS_TCP_S_LAST+1] = {
 365         [IP_VS_TCP_S_NONE]              =       2*HZ,
 366         [IP_VS_TCP_S_ESTABLISHED]       =       15*60*HZ,
 367         [IP_VS_TCP_S_SYN_SENT]          =       2*60*HZ,
 368         [IP_VS_TCP_S_SYN_RECV]          =       1*60*HZ,
 369         [IP_VS_TCP_S_FIN_WAIT]          =       2*60*HZ,
 370         [IP_VS_TCP_S_TIME_WAIT]         =       2*60*HZ,
 371         [IP_VS_TCP_S_CLOSE]             =       10*HZ,
 372         [IP_VS_TCP_S_CLOSE_WAIT]        =       60*HZ,
 373         [IP_VS_TCP_S_LAST_ACK]          =       30*HZ,
 374         [IP_VS_TCP_S_LISTEN]            =       2*60*HZ,
 375         [IP_VS_TCP_S_SYNACK]            =       120*HZ,
 376         [IP_VS_TCP_S_LAST]              =       2*HZ,
 377 };
 378
 379 static const char *const tcp_state_name_table[IP_VS_TCP_S_LAST+1] = {
 380         [IP_VS_TCP_S_NONE]              =       "NONE",
 381         [IP_VS_TCP_S_ESTABLISHED]       =       "ESTABLISHED",
 382         [IP_VS_TCP_S_SYN_SENT]          =       "SYN_SENT",
 383         [IP_VS_TCP_S_SYN_RECV]          =       "SYN_RECV",
 384         [IP_VS_TCP_S_FIN_WAIT]          =       "FIN_WAIT",
 385         [IP_VS_TCP_S_TIME_WAIT]         =       "TIME_WAIT",
 386         [IP_VS_TCP_S_CLOSE]             =       "CLOSE",
 387         [IP_VS_TCP_S_CLOSE_WAIT]        =       "CLOSE_WAIT",
 388         [IP_VS_TCP_S_LAST_ACK]          =       "LAST_ACK",
 389         [IP_VS_TCP_S_LISTEN]            =       "LISTEN",
 390         [IP_VS_TCP_S_SYNACK]            =       "SYNACK",
 391         [IP_VS_TCP_S_LAST]              =       "BUG!",
 392 };
 393
 394 static const bool tcp_state_active_table[IP_VS_TCP_S_LAST] = {
 395         [IP_VS_TCP_S_NONE]              =       false,
 396         [IP_VS_TCP_S_ESTABLISHED]       =       true,
 397         [IP_VS_TCP_S_SYN_SENT]          =       true,
 398         [IP_VS_TCP_S_SYN_RECV]          =       true,
 399         [IP_VS_TCP_S_FIN_WAIT]          =       false,
 400         [IP_VS_TCP_S_TIME_WAIT]         =       false,
 401         [IP_VS_TCP_S_CLOSE]             =       false,
 402         [IP_VS_TCP_S_CLOSE_WAIT]        =       false,
 403         [IP_VS_TCP_S_LAST_ACK]          =       false,
 404         [IP_VS_TCP_S_LISTEN]            =       false,
 405         [IP_VS_TCP_S_SYNACK]            =       true,
 406 };
 407
 408 #define sNO IP_VS_TCP_S_NONE
 409 #define sES IP_VS_TCP_S_ESTABLISHED
 410 #define sSS IP_VS_TCP_S_SYN_SENT
 411 #define sSR IP_VS_TCP_S_SYN_RECV
 412 #define sFW IP_VS_TCP_S_FIN_WAIT
 413 #define sTW IP_VS_TCP_S_TIME_WAIT
 414 #define sCL IP_VS_TCP_S_CLOSE
 415 #define sCW IP_VS_TCP_S_CLOSE_WAIT
 416 #define sLA IP_VS_TCP_S_LAST_ACK
 417 #define sLI IP_VS_TCP_S_LISTEN
 418 #define sSA IP_VS_TCP_S_SYNACK
 419
 420 struct tcp_states_t {
 421         int next_state[IP_VS_TCP_S_LAST];
 422 };
 423
 424 static const char * tcp_state_name(int state)
 425 {
 426         if (state >= IP_VS_TCP_S_LAST)
 427                 return "ERR!";
 428         return tcp_state_name_table[state] ? tcp_state_name_table[state] : "?";
 429 }
 430
 431 static bool tcp_state_active(int state)
 432 {
 433         if (state >= IP_VS_TCP_S_LAST)
 434                 return false;
 435         return tcp_state_active_table[state];
 436 }
 437
 438 static struct tcp_states_t tcp_states [] = {
 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, sSR }},
 442 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sTW }},
 443 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 444 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sSR }},
 445
 446 /*      OUTPUT */
 447 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 448 /*syn*/ {{sSS, sES, sSS, sSR, sSS, sSS, sSS, sSS, sSS, sLI, sSR }},
 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*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }},
 456 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }},
 457 /*ack*/ {{sES, 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_states_dos [] = {
 462 /*      INPUT */
 463 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 464 /*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSA }},
 465 /*fin*/ {{sCL, sCW, sSS, sTW, sTW, sTW, sCL, sCW, sLA, sLI, sSA }},
 466 /*ack*/ {{sES, sES, sSS, sSR, sFW, sTW, sCL, sCW, sCL, sLI, sSA }},
 467 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 468
 469 /*      OUTPUT */
 470 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 471 /*syn*/ {{sSS, sES, sSS, sSA, sSS, sSS, sSS, sSS, sSS, sLI, sSA }},
 472 /*fin*/ {{sTW, sFW, sSS, sTW, sFW, sTW, sCL, sTW, sLA, sLI, sTW }},
 473 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sLA, sES, sES }},
 474 /*rst*/ {{sCL, sCL, sSS, sCL, sCL, sTW, sCL, sCL, sCL, sCL, sCL }},
 475
 476 /*      INPUT-ONLY */
 477 /*        sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
 478 /*syn*/ {{sSA, sES, sES, sSR, sSA, sSA, sSA, sSA, sSA, sSA, sSA }},
 479 /*fin*/ {{sCL, sFW, sSS, sTW, sFW, sTW, sCL, sCW, sLA, sLI, sTW }},
 480 /*ack*/ {{sES, sES, sSS, sES, sFW, sTW, sCL, sCW, sCL, sLI, sES }},
 481 /*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
 482 };
 483
 484 static void tcp_timeout_change(struct ip_vs_proto_data *pd, int flags)
 485 {
 486         int on = (flags & 1);           /* secure_tcp */
 487
 488         /*
 489         ** FIXME: change secure_tcp to independent sysctl var
 490         ** or make it per-service or per-app because it is valid
 491         ** for most if not for all of the applications. Something
 492         ** like "capabilities" (flags) for each object.
 493         */
 494         pd->tcp_state_table = (on ? tcp_states_dos : tcp_states);
 495 }
 496
 497 static inline int tcp_state_idx(struct tcphdr *th)
 498 {
 499         if (th->rst)
 500                 return 3;
 501         if (th->syn)
 502                 return 0;
 503         if (th->fin)
 504                 return 1;
 505         if (th->ack)
 506                 return 2;
 507         return -1;
 508 }
 509
 510 static inline void
 511 set_tcp_state(struct ip_vs_proto_data *pd, struct ip_vs_conn *cp,
 512               int direction, struct tcphdr *th)
 513 {
 514         int state_idx;
 515         int new_state = IP_VS_TCP_S_CLOSE;
 516         int state_off = tcp_state_off[direction];
 517
 518         /*
 519          *    Update state offset to INPUT_ONLY if necessary
 520          *    or delete NO_OUTPUT flag if output packet detected
 521          */
 522         if (cp->flags & IP_VS_CONN_F_NOOUTPUT) {
 523                 if (state_off == TCP_DIR_OUTPUT)
 524                         cp->flags &= ~IP_VS_CONN_F_NOOUTPUT;
 525                 else
 526                         state_off = TCP_DIR_INPUT_ONLY;
 527         }
 528
 529         if ((state_idx = tcp_state_idx(th)) < 0) {
 530                 IP_VS_DBG(8, "tcp_state_idx=%d!!!\n", state_idx);
 531                 goto tcp_state_out;
 532         }
 533
 534         new_state =
 535                 pd->tcp_state_table[state_off+state_idx].next_state[cp->state];
 536
 537   tcp_state_out:
 538         if (new_state != cp->state) {
 539                 struct ip_vs_dest *dest = cp->dest;
 540
 541                 IP_VS_DBG_BUF(8, "%s %s [%c%c%c%c] %s:%d->"
 542                               "%s:%d state: %s->%s conn->refcnt:%d\n",
 543                               pd->pp->name,
 544                               ((state_off == TCP_DIR_OUTPUT) ?
 545                                "output " : "input "),
 546                               th->syn ? 'S' : '.',
 547                               th->fin ? 'F' : '.',
 548                               th->ack ? 'A' : '.',
 549                               th->rst ? 'R' : '.',
 550                               IP_VS_DBG_ADDR(cp->daf, &cp->daddr),
 551                               ntohs(cp->dport),
 552                               IP_VS_DBG_ADDR(cp->af, &cp->caddr),
 553                               ntohs(cp->cport),
 554                               tcp_state_name(cp->state),
 555                               tcp_state_name(new_state),
 556                               refcount_read(&cp->refcnt));
 557
 558                 if (dest) {
 559                         if (!(cp->flags & IP_VS_CONN_F_INACTIVE) &&
 560                             !tcp_state_active(new_state)) {
 561                                 atomic_dec(&dest->activeconns);
 562                                 atomic_inc(&dest->inactconns);
 563                                 cp->flags |= IP_VS_CONN_F_INACTIVE;
 564                         } else if ((cp->flags & IP_VS_CONN_F_INACTIVE) &&
 565                                    tcp_state_active(new_state)) {
 566                                 atomic_inc(&dest->activeconns);
 567                                 atomic_dec(&dest->inactconns);
 568                                 cp->flags &= ~IP_VS_CONN_F_INACTIVE;
 569                         }
 570                 }
 571         }
 572
 573         if (likely(pd))
 574                 cp->timeout = pd->timeout_table[cp->state = new_state];
 575         else    /* What to do ? */
 576                 cp->timeout = tcp_timeouts[cp->state = new_state];
 577 }
 578
 579 /*
 580  *      Handle state transitions
 581  */
 582 static void
 583 tcp_state_transition(struct ip_vs_conn *cp, int direction,
 584                      const struct sk_buff *skb,
 585                      struct ip_vs_proto_data *pd)
 586 {
 587         struct tcphdr _tcph, *th;
 588
 589 #ifdef CONFIG_IP_VS_IPV6
 590         int ihl = cp->af == AF_INET ? ip_hdrlen(skb) : sizeof(struct ipv6hdr);
 591 #else
 592         int ihl = ip_hdrlen(skb);
 593 #endif
 594
 595         th = skb_header_pointer(skb, ihl, sizeof(_tcph), &_tcph);
 596         if (th == NULL)
 597                 return;
 598
 599         spin_lock_bh(&cp->lock);
 600         set_tcp_state(pd, cp, direction, th);
 601         spin_unlock_bh(&cp->lock);
 602 }
 603
 604 static inline __u16 tcp_app_hashkey(__be16 port)
 605 {
 606         return (((__force u16)port >> TCP_APP_TAB_BITS) ^ (__force u16)port)
 607                 & TCP_APP_TAB_MASK;
 608 }
 609
 610
 611 static int tcp_register_app(struct netns_ipvs *ipvs, struct ip_vs_app *inc)
 612 {
 613         struct ip_vs_app *i;
 614         __u16 hash;
 615         __be16 port = inc->port;
 616         int ret = 0;
 617         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs, IPPROTO_TCP);
 618
 619         hash = tcp_app_hashkey(port);
 620
 621         list_for_each_entry(i, &ipvs->tcp_apps[hash], p_list) {
 622                 if (i->port == port) {
 623                         ret = -EEXIST;
 624                         goto out;
 625                 }
 626         }
 627         list_add_rcu(&inc->p_list, &ipvs->tcp_apps[hash]);
 628         atomic_inc(&pd->appcnt);
 629
 630   out:
 631         return ret;
 632 }
 633
 634
 635 static void
 636 tcp_unregister_app(struct netns_ipvs *ipvs, struct ip_vs_app *inc)
 637 {
 638         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs, IPPROTO_TCP);
 639
 640         atomic_dec(&pd->appcnt);
 641         list_del_rcu(&inc->p_list);
 642 }
 643
 644
 645 static int
 646 tcp_app_conn_bind(struct ip_vs_conn *cp)
 647 {
 648         struct netns_ipvs *ipvs = cp->ipvs;
 649         int hash;
 650         struct ip_vs_app *inc;
 651         int result = 0;
 652
 653         /* Default binding: bind app only for NAT */
 654         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ)
 655                 return 0;
 656
 657         /* Lookup application incarnations and bind the right one */
 658         hash = tcp_app_hashkey(cp->vport);
 659
 660         list_for_each_entry_rcu(inc, &ipvs->tcp_apps[hash], p_list) {
 661                 if (inc->port == cp->vport) {
 662                         if (unlikely(!ip_vs_app_inc_get(inc)))
 663                                 break;
 664
 665                         IP_VS_DBG_BUF(9, "%s(): Binding conn %s:%u->"
 666                                       "%s:%u to app %s on port %u\n",
 667                                       __func__,
 668                                       IP_VS_DBG_ADDR(cp->af, &cp->caddr),
 669                                       ntohs(cp->cport),
 670                                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr),
 671                                       ntohs(cp->vport),
 672                                       inc->name, ntohs(inc->port));
 673
 674                         cp->app = inc;
 675                         if (inc->init_conn)
 676                                 result = inc->init_conn(inc, cp);
 677                         break;
 678                 }
 679         }
 680
 681         return result;
 682 }
 683
 684
 685 /*
 686  *      Set LISTEN timeout. (ip_vs_conn_put will setup timer)
 687  */
 688 void ip_vs_tcp_conn_listen(struct ip_vs_conn *cp)
 689 {
 690         struct ip_vs_proto_data *pd = ip_vs_proto_data_get(cp->ipvs, IPPROTO_TCP);
 691
 692         spin_lock_bh(&cp->lock);
 693         cp->state = IP_VS_TCP_S_LISTEN;
 694         cp->timeout = (pd ? pd->timeout_table[IP_VS_TCP_S_LISTEN]
 695                            : tcp_timeouts[IP_VS_TCP_S_LISTEN]);
 696         spin_unlock_bh(&cp->lock);
 697 }
 698
 699 /* ---------------------------------------------
 700  *   timeouts is netns related now.
 701  * ---------------------------------------------
 702  */
 703 static int __ip_vs_tcp_init(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd)
 704 {
 705         ip_vs_init_hash_table(ipvs->tcp_apps, TCP_APP_TAB_SIZE);
 706         pd->timeout_table = ip_vs_create_timeout_table((int *)tcp_timeouts,
 707                                                         sizeof(tcp_timeouts));
 708         if (!pd->timeout_table)
 709                 return -ENOMEM;
 710         pd->tcp_state_table =  tcp_states;
 711         return 0;
 712 }
 713
 714 static void __ip_vs_tcp_exit(struct netns_ipvs *ipvs, struct ip_vs_proto_data *pd)
 715 {
 716         kfree(pd->timeout_table);
 717 }
 718
 719
 720 struct ip_vs_protocol ip_vs_protocol_tcp = {
 721         .name =                 "TCP",
 722         .protocol =             IPPROTO_TCP,
 723         .num_states =           IP_VS_TCP_S_LAST,
 724         .dont_defrag =          0,
 725         .init =                 NULL,
 726         .exit =                 NULL,
 727         .init_netns =           __ip_vs_tcp_init,
 728         .exit_netns =           __ip_vs_tcp_exit,
 729         .register_app =         tcp_register_app,
 730         .unregister_app =       tcp_unregister_app,
 731         .conn_schedule =        tcp_conn_schedule,
 732         .conn_in_get =          ip_vs_conn_in_get_proto,
 733         .conn_out_get =         ip_vs_conn_out_get_proto,
 734         .snat_handler =         tcp_snat_handler,
 735         .dnat_handler =         tcp_dnat_handler,
 736         .csum_check =           tcp_csum_check,
 737         .state_name =           tcp_state_name,
 738         .state_transition =     tcp_state_transition,
 739         .app_conn_bind =        tcp_app_conn_bind,
 740         .debug_packet =         ip_vs_tcpudp_debug_packet,
 741         .timeout_change =       tcp_timeout_change,
 742 };