SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/netfilter/ipvs/ip_vs_conn.c

   1 /*
   2  * IPVS         An implementation of the IP virtual server support for the
   3  *              LINUX operating system.  IPVS is now implemented as a module
   4  *              over the Netfilter framework. IPVS can be used to build a
   5  *              high-performance and highly available server based on a
   6  *              cluster of servers.
   7  *
   8  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
   9  *              Peter Kese <peter.kese@ijs.si>
  10  *              Julian Anastasov <ja@ssi.bg>
  11  *
  12  *              This program is free software; you can redistribute it and/or
  13  *              modify it under the terms of the GNU General Public License
  14  *              as published by the Free Software Foundation; either version
  15  *              2 of the License, or (at your option) any later version.
  16  *
  17  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
  18  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
  19  * and others. Many code here is taken from IP MASQ code of kernel 2.2.
  20  *
  21  * Changes:
  22  *
  23  */
  24
  25 #define KMSG_COMPONENT "IPVS"
  26 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  27
  28 #include <linux/interrupt.h>
  29 #include <linux/in.h>
  30 #include <linux/net.h>
  31 #include <linux/kernel.h>
  32 #include <linux/module.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/proc_fs.h>              /* for proc_net_* */
  35 #include <linux/seq_file.h>
  36 #include <linux/jhash.h>
  37 #include <linux/random.h>
  38
  39 #include <net/net_namespace.h>
  40 #include <net/ip_vs.h>
  41
  42
  43 #ifndef CONFIG_IP_VS_TAB_BITS
  44 #define CONFIG_IP_VS_TAB_BITS   12
  45 #endif
  46
  47 /*
  48  * Connection hash size. Default is what was selected at compile time.
  49 */
  50 int ip_vs_conn_tab_bits = CONFIG_IP_VS_TAB_BITS;
  51 module_param_named(conn_tab_bits, ip_vs_conn_tab_bits, int, 0444);
  52 MODULE_PARM_DESC(conn_tab_bits, "Set connections' hash size");
  53
  54 /* size and mask values */
  55 int ip_vs_conn_tab_size;
  56 int ip_vs_conn_tab_mask;
  57
  58 /*
  59  *  Connection hash table: for input and output packets lookups of IPVS
  60  */
  61 static struct list_head *ip_vs_conn_tab;
  62
  63 /*  SLAB cache for IPVS connections */
  64 static struct kmem_cache *ip_vs_conn_cachep __read_mostly;
  65
  66 /*  counter for current IPVS connections */
  67 static atomic_t ip_vs_conn_count = ATOMIC_INIT(0);
  68
  69 /*  counter for no client port connections */
  70 static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
  71
  72 /* random value for IPVS connection hash */
  73 static unsigned int ip_vs_conn_rnd;
  74
  75 /*
  76  *  Fine locking granularity for big connection hash table
  77  */
  78 #define CT_LOCKARRAY_BITS  4
  79 #define CT_LOCKARRAY_SIZE  (1<<CT_LOCKARRAY_BITS)
  80 #define CT_LOCKARRAY_MASK  (CT_LOCKARRAY_SIZE-1)
  81
  82 struct ip_vs_aligned_lock
  83 {
  84         rwlock_t        l;
  85 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
  86
  87 /* lock array for conn table */
  88 static struct ip_vs_aligned_lock
  89 __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
  90
  91 static inline void ct_read_lock(unsigned key)
  92 {
  93         read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  94 }
  95
  96 static inline void ct_read_unlock(unsigned key)
  97 {
  98         read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  99 }
 100
 101 static inline void ct_write_lock(unsigned key)
 102 {
 103         write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 104 }
 105
 106 static inline void ct_write_unlock(unsigned key)
 107 {
 108         write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 109 }
 110
 111 static inline void ct_read_lock_bh(unsigned key)
 112 {
 113         read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 114 }
 115
 116 static inline void ct_read_unlock_bh(unsigned key)
 117 {
 118         read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 119 }
 120
 121 static inline void ct_write_lock_bh(unsigned key)
 122 {
 123         write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 124 }
 125
 126 static inline void ct_write_unlock_bh(unsigned key)
 127 {
 128         write_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 129 }
 130
 131
 132 /*
 133  *      Returns hash value for IPVS connection entry
 134  */
 135 static unsigned int ip_vs_conn_hashkey(int af, unsigned proto,
 136                                        const union nf_inet_addr *addr,
 137                                        __be16 port)
 138 {
 139 #ifdef CONFIG_IP_VS_IPV6
 140         if (af == AF_INET6)
 141                 return jhash_3words(jhash(addr, 16, ip_vs_conn_rnd),
 142                                     (__force u32)port, proto, ip_vs_conn_rnd)
 143                         & ip_vs_conn_tab_mask;
 144 #endif
 145         return jhash_3words((__force u32)addr->ip, (__force u32)port, proto,
 146                             ip_vs_conn_rnd)
 147                 & ip_vs_conn_tab_mask;
 148 }
 149
 150
 151 /*
 152  *      Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port.
 153  *      returns bool success.
 154  */
 155 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
 156 {
 157         unsigned hash;
 158         int ret;
 159
 160         /* Hash by protocol, client address and port */
 161         hash = ip_vs_conn_hashkey(cp->af, cp->protocol, &cp->caddr, cp->cport);
 162
 163         ct_write_lock(hash);
 164
 165         if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
 166                 list_add(&cp->c_list, &ip_vs_conn_tab[hash]);
 167                 cp->flags |= IP_VS_CONN_F_HASHED;
 168                 atomic_inc(&cp->refcnt);
 169                 ret = 1;
 170         } else {
 171                 pr_err("%s(): request for already hashed, called from %pF\n",
 172                        __func__, __builtin_return_address(0));
 173                 ret = 0;
 174         }
 175
 176         ct_write_unlock(hash);
 177
 178         return ret;
 179 }
 180
 181
 182 /*
 183  *      UNhashes ip_vs_conn from ip_vs_conn_tab.
 184  *      returns bool success.
 185  */
 186 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
 187 {
 188         unsigned hash;
 189         int ret;
 190
 191         /* unhash it and decrease its reference counter */
 192         hash = ip_vs_conn_hashkey(cp->af, cp->protocol, &cp->caddr, cp->cport);
 193
 194         ct_write_lock(hash);
 195
 196         if (cp->flags & IP_VS_CONN_F_HASHED) {
 197                 list_del(&cp->c_list);
 198                 cp->flags &= ~IP_VS_CONN_F_HASHED;
 199                 atomic_dec(&cp->refcnt);
 200                 ret = 1;
 201         } else
 202                 ret = 0;
 203
 204         ct_write_unlock(hash);
 205
 206         return ret;
 207 }
 208
 209
 210 /*
 211  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
 212  *  Called for pkts coming from OUTside-to-INside.
 213  *      s_addr, s_port: pkt source address (foreign host)
 214  *      d_addr, d_port: pkt dest address (load balancer)
 215  */
 216 static inline struct ip_vs_conn *__ip_vs_conn_in_get
 217 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
 218  const union nf_inet_addr *d_addr, __be16 d_port)
 219 {
 220         unsigned hash;
 221         struct ip_vs_conn *cp;
 222
 223         hash = ip_vs_conn_hashkey(af, protocol, s_addr, s_port);
 224
 225         ct_read_lock(hash);
 226
 227         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
 228                 if (cp->af == af &&
 229                     ip_vs_addr_equal(af, s_addr, &cp->caddr) &&
 230                     ip_vs_addr_equal(af, d_addr, &cp->vaddr) &&
 231                     s_port == cp->cport && d_port == cp->vport &&
 232                     ((!s_port) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
 233                     protocol == cp->protocol) {
 234                         /* HIT */
 235                         atomic_inc(&cp->refcnt);
 236                         ct_read_unlock(hash);
 237                         return cp;
 238                 }
 239         }
 240
 241         ct_read_unlock(hash);
 242
 243         return NULL;
 244 }
 245
 246 struct ip_vs_conn *ip_vs_conn_in_get
 247 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
 248  const union nf_inet_addr *d_addr, __be16 d_port)
 249 {
 250         struct ip_vs_conn *cp;
 251
 252         cp = __ip_vs_conn_in_get(af, protocol, s_addr, s_port, d_addr, d_port);
 253         if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt))
 254                 cp = __ip_vs_conn_in_get(af, protocol, s_addr, 0, d_addr,
 255                                          d_port);
 256
 257         IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
 258                       ip_vs_proto_name(protocol),
 259                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
 260                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
 261                       cp ? "hit" : "not hit");
 262
 263         return cp;
 264 }
 265
 266 /* Get reference to connection template */
 267 struct ip_vs_conn *ip_vs_ct_in_get
 268 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
 269  const union nf_inet_addr *d_addr, __be16 d_port)
 270 {
 271         unsigned hash;
 272         struct ip_vs_conn *cp;
 273
 274         hash = ip_vs_conn_hashkey(af, protocol, s_addr, s_port);
 275
 276         ct_read_lock(hash);
 277
 278         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
 279                 if (cp->af == af &&
 280                     ip_vs_addr_equal(af, s_addr, &cp->caddr) &&
 281                     /* protocol should only be IPPROTO_IP if
 282                      * d_addr is a fwmark */
 283                     ip_vs_addr_equal(protocol == IPPROTO_IP ? AF_UNSPEC : af,
 284                                      d_addr, &cp->vaddr) &&
 285                     s_port == cp->cport && d_port == cp->vport &&
 286                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
 287                     protocol == cp->protocol) {
 288                         /* HIT */
 289                         atomic_inc(&cp->refcnt);
 290                         goto out;
 291                 }
 292         }
 293         cp = NULL;
 294
 295   out:
 296         ct_read_unlock(hash);
 297
 298         IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
 299                       ip_vs_proto_name(protocol),
 300                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
 301                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
 302                       cp ? "hit" : "not hit");
 303
 304         return cp;
 305 }
 306
 307 /*
 308  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
 309  *  Called for pkts coming from inside-to-OUTside.
 310  *      s_addr, s_port: pkt source address (inside host)
 311  *      d_addr, d_port: pkt dest address (foreign host)
 312  */
 313 struct ip_vs_conn *ip_vs_conn_out_get
 314 (int af, int protocol, const union nf_inet_addr *s_addr, __be16 s_port,
 315  const union nf_inet_addr *d_addr, __be16 d_port)
 316 {
 317         unsigned hash;
 318         struct ip_vs_conn *cp, *ret=NULL;
 319
 320         /*
 321          *      Check for "full" addressed entries
 322          */
 323         hash = ip_vs_conn_hashkey(af, protocol, d_addr, d_port);
 324
 325         ct_read_lock(hash);
 326
 327         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
 328                 if (cp->af == af &&
 329                     ip_vs_addr_equal(af, d_addr, &cp->caddr) &&
 330                     ip_vs_addr_equal(af, s_addr, &cp->daddr) &&
 331                     d_port == cp->cport && s_port == cp->dport &&
 332                     protocol == cp->protocol) {
 333                         /* HIT */
 334                         atomic_inc(&cp->refcnt);
 335                         ret = cp;
 336                         break;
 337                 }
 338         }
 339
 340         ct_read_unlock(hash);
 341
 342         IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
 343                       ip_vs_proto_name(protocol),
 344                       IP_VS_DBG_ADDR(af, s_addr), ntohs(s_port),
 345                       IP_VS_DBG_ADDR(af, d_addr), ntohs(d_port),
 346                       ret ? "hit" : "not hit");
 347
 348         return ret;
 349 }
 350
 351
 352 /*
 353  *      Put back the conn and restart its timer with its timeout
 354  */
 355 void ip_vs_conn_put(struct ip_vs_conn *cp)
 356 {
 357         /* reset it expire in its timeout */
 358         mod_timer(&cp->timer, jiffies+cp->timeout);
 359
 360         __ip_vs_conn_put(cp);
 361 }
 362
 363
 364 /*
 365  *      Fill a no_client_port connection with a client port number
 366  */
 367 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
 368 {
 369         if (ip_vs_conn_unhash(cp)) {
 370                 spin_lock(&cp->lock);
 371                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
 372                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 373                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
 374                         cp->cport = cport;
 375                 }
 376                 spin_unlock(&cp->lock);
 377
 378                 /* hash on new dport */
 379                 ip_vs_conn_hash(cp);
 380         }
 381 }
 382
 383
 384 /*
 385  *      Bind a connection entry with the corresponding packet_xmit.
 386  *      Called by ip_vs_conn_new.
 387  */
 388 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
 389 {
 390         switch (IP_VS_FWD_METHOD(cp)) {
 391         case IP_VS_CONN_F_MASQ:
 392                 cp->packet_xmit = ip_vs_nat_xmit;
 393                 break;
 394
 395         case IP_VS_CONN_F_TUNNEL:
 396                 cp->packet_xmit = ip_vs_tunnel_xmit;
 397                 break;
 398
 399         case IP_VS_CONN_F_DROUTE:
 400                 cp->packet_xmit = ip_vs_dr_xmit;
 401                 break;
 402
 403         case IP_VS_CONN_F_LOCALNODE:
 404                 cp->packet_xmit = ip_vs_null_xmit;
 405                 break;
 406
 407         case IP_VS_CONN_F_BYPASS:
 408                 cp->packet_xmit = ip_vs_bypass_xmit;
 409                 break;
 410         }
 411 }
 412
 413 #ifdef CONFIG_IP_VS_IPV6
 414 static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
 415 {
 416         switch (IP_VS_FWD_METHOD(cp)) {
 417         case IP_VS_CONN_F_MASQ:
 418                 cp->packet_xmit = ip_vs_nat_xmit_v6;
 419                 break;
 420
 421         case IP_VS_CONN_F_TUNNEL:
 422                 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
 423                 break;
 424
 425         case IP_VS_CONN_F_DROUTE:
 426                 cp->packet_xmit = ip_vs_dr_xmit_v6;
 427                 break;
 428
 429         case IP_VS_CONN_F_LOCALNODE:
 430                 cp->packet_xmit = ip_vs_null_xmit;
 431                 break;
 432
 433         case IP_VS_CONN_F_BYPASS:
 434                 cp->packet_xmit = ip_vs_bypass_xmit_v6;
 435                 break;
 436         }
 437 }
 438 #endif
 439
 440
 441 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
 442 {
 443         return atomic_read(&dest->activeconns)
 444                 + atomic_read(&dest->inactconns);
 445 }
 446
 447 /*
 448  *      Bind a connection entry with a virtual service destination
 449  *      Called just after a new connection entry is created.
 450  */
 451 static inline void
 452 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
 453 {
 454         /* if dest is NULL, then return directly */
 455         if (!dest)
 456                 return;
 457
 458         /* Increase the refcnt counter of the dest */
 459         atomic_inc(&dest->refcnt);
 460
 461         /* Bind with the destination and its corresponding transmitter */
 462         if ((cp->flags & IP_VS_CONN_F_SYNC) &&
 463             (!(cp->flags & IP_VS_CONN_F_TEMPLATE)))
 464                 /* if the connection is not template and is created
 465                  * by sync, preserve the activity flag.
 466                  */
 467                 cp->flags |= atomic_read(&dest->conn_flags) &
 468                              (~IP_VS_CONN_F_INACTIVE);
 469         else
 470                 cp->flags |= atomic_read(&dest->conn_flags);
 471         cp->dest = dest;
 472
 473         IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
 474                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 475                       "dest->refcnt:%d\n",
 476                       ip_vs_proto_name(cp->protocol),
 477                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 478                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 479                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
 480                       ip_vs_fwd_tag(cp), cp->state,
 481                       cp->flags, atomic_read(&cp->refcnt),
 482                       atomic_read(&dest->refcnt));
 483
 484         /* Update the connection counters */
 485         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
 486                 /* It is a normal connection, so increase the inactive
 487                    connection counter because it is in TCP SYNRECV
 488                    state (inactive) or other protocol inacive state */
 489                 if ((cp->flags & IP_VS_CONN_F_SYNC) &&
 490                     (!(cp->flags & IP_VS_CONN_F_INACTIVE)))
 491                         atomic_inc(&dest->activeconns);
 492                 else
 493                         atomic_inc(&dest->inactconns);
 494         } else {
 495                 /* It is a persistent connection/template, so increase
 496                    the peristent connection counter */
 497                 atomic_inc(&dest->persistconns);
 498         }
 499
 500         if (dest->u_threshold != 0 &&
 501             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
 502                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
 503 }
 504
 505
 506 /*
 507  * Check if there is a destination for the connection, if so
 508  * bind the connection to the destination.
 509  */
 510 struct ip_vs_dest *ip_vs_try_bind_dest(struct ip_vs_conn *cp)
 511 {
 512         struct ip_vs_dest *dest;
 513
 514         if ((cp) && (!cp->dest)) {
 515                 dest = ip_vs_find_dest(cp->af, &cp->daddr, cp->dport,
 516                                        &cp->vaddr, cp->vport,
 517                                        cp->protocol);
 518                 ip_vs_bind_dest(cp, dest);
 519                 return dest;
 520         } else
 521                 return NULL;
 522 }
 523
 524
 525 /*
 526  *      Unbind a connection entry with its VS destination
 527  *      Called by the ip_vs_conn_expire function.
 528  */
 529 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
 530 {
 531         struct ip_vs_dest *dest = cp->dest;
 532
 533         if (!dest)
 534                 return;
 535
 536         IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
 537                       "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 538                       "dest->refcnt:%d\n",
 539                       ip_vs_proto_name(cp->protocol),
 540                       IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
 541                       IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
 542                       IP_VS_DBG_ADDR(cp->af, &cp->daddr), ntohs(cp->dport),
 543                       ip_vs_fwd_tag(cp), cp->state,
 544                       cp->flags, atomic_read(&cp->refcnt),
 545                       atomic_read(&dest->refcnt));
 546
 547         /* Update the connection counters */
 548         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
 549                 /* It is a normal connection, so decrease the inactconns
 550                    or activeconns counter */
 551                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
 552                         atomic_dec(&dest->inactconns);
 553                 } else {
 554                         atomic_dec(&dest->activeconns);
 555                 }
 556         } else {
 557                 /* It is a persistent connection/template, so decrease
 558                    the peristent connection counter */
 559                 atomic_dec(&dest->persistconns);
 560         }
 561
 562         if (dest->l_threshold != 0) {
 563                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
 564                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 565         } else if (dest->u_threshold != 0) {
 566                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
 567                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 568         } else {
 569                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 570                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 571         }
 572
 573         /*
 574          * Simply decrease the refcnt of the dest, because the
 575          * dest will be either in service's destination list
 576          * or in the trash.
 577          */
 578         atomic_dec(&dest->refcnt);
 579 }
 580
 581
 582 /*
 583  *      Checking if the destination of a connection template is available.
 584  *      If available, return 1, otherwise invalidate this connection
 585  *      template and return 0.
 586  */
 587 int ip_vs_check_template(struct ip_vs_conn *ct)
 588 {
 589         struct ip_vs_dest *dest = ct->dest;
 590
 591         /*
 592          * Checking the dest server status.
 593          */
 594         if ((dest == NULL) ||
 595             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
 596             (sysctl_ip_vs_expire_quiescent_template &&
 597              (atomic_read(&dest->weight) == 0))) {
 598                 IP_VS_DBG_BUF(9, "check_template: dest not available for "
 599                               "protocol %s s:%s:%d v:%s:%d "
 600                               "-> d:%s:%d\n",
 601                               ip_vs_proto_name(ct->protocol),
 602                               IP_VS_DBG_ADDR(ct->af, &ct->caddr),
 603                               ntohs(ct->cport),
 604                               IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
 605                               ntohs(ct->vport),
 606                               IP_VS_DBG_ADDR(ct->af, &ct->daddr),
 607                               ntohs(ct->dport));
 608
 609                 /*
 610                  * Invalidate the connection template
 611                  */
 612                 if (ct->vport != htons(0xffff)) {
 613                         if (ip_vs_conn_unhash(ct)) {
 614                                 ct->dport = htons(0xffff);
 615                                 ct->vport = htons(0xffff);
 616                                 ct->cport = 0;
 617                                 ip_vs_conn_hash(ct);
 618                         }
 619                 }
 620
 621                 /*
 622                  * Simply decrease the refcnt of the template,
 623                  * don't restart its timer.
 624                  */
 625                 atomic_dec(&ct->refcnt);
 626                 return 0;
 627         }
 628         return 1;
 629 }
 630
 631 static void ip_vs_conn_expire(unsigned long data)
 632 {
 633         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
 634
 635         cp->timeout = 60*HZ;
 636
 637         /*
 638          *      hey, I'm using it
 639          */
 640         atomic_inc(&cp->refcnt);
 641
 642         /*
 643          *      do I control anybody?
 644          */
 645         if (atomic_read(&cp->n_control))
 646                 goto expire_later;
 647
 648         /*
 649          *      unhash it if it is hashed in the conn table
 650          */
 651         if (!ip_vs_conn_unhash(cp))
 652                 goto expire_later;
 653
 654         /*
 655          *      refcnt==1 implies I'm the only one referrer
 656          */
 657         if (likely(atomic_read(&cp->refcnt) == 1)) {
 658                 /* delete the timer if it is activated by other users */
 659                 if (timer_pending(&cp->timer))
 660                         del_timer(&cp->timer);
 661
 662                 /* does anybody control me? */
 663                 if (cp->control)
 664                         ip_vs_control_del(cp);
 665
 666                 if (unlikely(cp->app != NULL))
 667                         ip_vs_unbind_app(cp);
 668                 ip_vs_unbind_dest(cp);
 669                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
 670                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 671                 atomic_dec(&ip_vs_conn_count);
 672
 673                 kmem_cache_free(ip_vs_conn_cachep, cp);
 674                 return;
 675         }
 676
 677         /* hash it back to the table */
 678         ip_vs_conn_hash(cp);
 679
 680   expire_later:
 681         IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
 682                   atomic_read(&cp->refcnt)-1,
 683                   atomic_read(&cp->n_control));
 684
 685         ip_vs_conn_put(cp);
 686 }
 687
 688
 689 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
 690 {
 691         if (del_timer(&cp->timer))
 692                 mod_timer(&cp->timer, jiffies);
 693 }
 694
 695
 696 /*
 697  *      Create a new connection entry and hash it into the ip_vs_conn_tab
 698  */
 699 struct ip_vs_conn *
 700 ip_vs_conn_new(int af, int proto, const union nf_inet_addr *caddr, __be16 cport,
 701                const union nf_inet_addr *vaddr, __be16 vport,
 702                const union nf_inet_addr *daddr, __be16 dport, unsigned flags,
 703                struct ip_vs_dest *dest)
 704 {
 705         struct ip_vs_conn *cp;
 706         struct ip_vs_protocol *pp = ip_vs_proto_get(proto);
 707
 708         cp = kmem_cache_zalloc(ip_vs_conn_cachep, GFP_ATOMIC);
 709         if (cp == NULL) {
 710                 IP_VS_ERR_RL("%s(): no memory\n", __func__);
 711                 return NULL;
 712         }
 713
 714         INIT_LIST_HEAD(&cp->c_list);
 715         setup_timer(&cp->timer, ip_vs_conn_expire, (unsigned long)cp);
 716         cp->af             = af;
 717         cp->protocol       = proto;
 718         ip_vs_addr_copy(af, &cp->caddr, caddr);
 719         cp->cport          = cport;
 720         ip_vs_addr_copy(af, &cp->vaddr, vaddr);
 721         cp->vport          = vport;
 722         /* proto should only be IPPROTO_IP if d_addr is a fwmark */
 723         ip_vs_addr_copy(proto == IPPROTO_IP ? AF_UNSPEC : af,
 724                         &cp->daddr, daddr);
 725         cp->dport          = dport;
 726         cp->flags          = flags;
 727         spin_lock_init(&cp->lock);
 728
 729         /*
 730          * Set the entry is referenced by the current thread before hashing
 731          * it in the table, so that other thread run ip_vs_random_dropentry
 732          * but cannot drop this entry.
 733          */
 734         atomic_set(&cp->refcnt, 1);
 735
 736         atomic_set(&cp->n_control, 0);
 737         atomic_set(&cp->in_pkts, 0);
 738
 739         atomic_inc(&ip_vs_conn_count);
 740         if (flags & IP_VS_CONN_F_NO_CPORT)
 741                 atomic_inc(&ip_vs_conn_no_cport_cnt);
 742
 743         /* Bind the connection with a destination server */
 744         ip_vs_bind_dest(cp, dest);
 745
 746         /* Set its state and timeout */
 747         cp->state = 0;
 748         cp->timeout = 3*HZ;
 749
 750         /* Bind its packet transmitter */
 751 #ifdef CONFIG_IP_VS_IPV6
 752         if (af == AF_INET6)
 753                 ip_vs_bind_xmit_v6(cp);
 754         else
 755 #endif
 756                 ip_vs_bind_xmit(cp);
 757
 758         if (unlikely(pp && atomic_read(&pp->appcnt)))
 759                 ip_vs_bind_app(cp, pp);
 760
 761         /* Hash it in the ip_vs_conn_tab finally */
 762         ip_vs_conn_hash(cp);
 763
 764         return cp;
 765 }
 766
 767
 768 /*
 769  *      /proc/net/ip_vs_conn entries
 770  */
 771 #ifdef CONFIG_PROC_FS
 772
 773 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
 774 {
 775         int idx;
 776         struct ip_vs_conn *cp;
 777
 778         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
 779                 ct_read_lock_bh(idx);
 780                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
 781                         if (pos-- == 0) {
 782                                 seq->private = &ip_vs_conn_tab[idx];
 783                                 return cp;
 784                         }
 785                 }
 786                 ct_read_unlock_bh(idx);
 787         }
 788
 789         return NULL;
 790 }
 791
 792 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
 793 {
 794         seq->private = NULL;
 795         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
 796 }
 797
 798 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 799 {
 800         struct ip_vs_conn *cp = v;
 801         struct list_head *e, *l = seq->private;
 802         int idx;
 803
 804         ++*pos;
 805         if (v == SEQ_START_TOKEN)
 806                 return ip_vs_conn_array(seq, 0);
 807
 808         /* more on same hash chain? */
 809         if ((e = cp->c_list.next) != l)
 810                 return list_entry(e, struct ip_vs_conn, c_list);
 811
 812         idx = l - ip_vs_conn_tab;
 813         ct_read_unlock_bh(idx);
 814
 815         while (++idx < ip_vs_conn_tab_size) {
 816                 ct_read_lock_bh(idx);
 817                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
 818                         seq->private = &ip_vs_conn_tab[idx];
 819                         return cp;
 820                 }
 821                 ct_read_unlock_bh(idx);
 822         }
 823         seq->private = NULL;
 824         return NULL;
 825 }
 826
 827 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
 828 {
 829         struct list_head *l = seq->private;
 830
 831         if (l)
 832                 ct_read_unlock_bh(l - ip_vs_conn_tab);
 833 }
 834
 835 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
 836 {
 837
 838         if (v == SEQ_START_TOKEN)
 839                 seq_puts(seq,
 840    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires\n");
 841         else {
 842                 const struct ip_vs_conn *cp = v;
 843
 844 #ifdef CONFIG_IP_VS_IPV6
 845                 if (cp->af == AF_INET6)
 846                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %7lu\n",
 847                                 ip_vs_proto_name(cp->protocol),
 848                                 &cp->caddr.in6, ntohs(cp->cport),
 849                                 &cp->vaddr.in6, ntohs(cp->vport),
 850                                 &cp->daddr.in6, ntohs(cp->dport),
 851                                 ip_vs_state_name(cp->protocol, cp->state),
 852                                 (cp->timer.expires-jiffies)/HZ);
 853                 else
 854 #endif
 855                         seq_printf(seq,
 856                                 "%-3s %08X %04X %08X %04X"
 857                                 " %08X %04X %-11s %7lu\n",
 858                                 ip_vs_proto_name(cp->protocol),
 859                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
 860                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
 861                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
 862                                 ip_vs_state_name(cp->protocol, cp->state),
 863                                 (cp->timer.expires-jiffies)/HZ);
 864         }
 865         return 0;
 866 }
 867
 868 static const struct seq_operations ip_vs_conn_seq_ops = {
 869         .start = ip_vs_conn_seq_start,
 870         .next  = ip_vs_conn_seq_next,
 871         .stop  = ip_vs_conn_seq_stop,
 872         .show  = ip_vs_conn_seq_show,
 873 };
 874
 875 static int ip_vs_conn_open(struct inode *inode, struct file *file)
 876 {
 877         return seq_open(file, &ip_vs_conn_seq_ops);
 878 }
 879
 880 static const struct file_operations ip_vs_conn_fops = {
 881         .owner   = THIS_MODULE,
 882         .open    = ip_vs_conn_open,
 883         .read    = seq_read,
 884         .llseek  = seq_lseek,
 885         .release = seq_release,
 886 };
 887
 888 static const char *ip_vs_origin_name(unsigned flags)
 889 {
 890         if (flags & IP_VS_CONN_F_SYNC)
 891                 return "SYNC";
 892         else
 893                 return "LOCAL";
 894 }
 895
 896 static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
 897 {
 898
 899         if (v == SEQ_START_TOKEN)
 900                 seq_puts(seq,
 901    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Origin Expires\n");
 902         else {
 903                 const struct ip_vs_conn *cp = v;
 904
 905 #ifdef CONFIG_IP_VS_IPV6
 906                 if (cp->af == AF_INET6)
 907                         seq_printf(seq, "%-3s %pI6 %04X %pI6 %04X %pI6 %04X %-11s %-6s %7lu\n",
 908                                 ip_vs_proto_name(cp->protocol),
 909                                 &cp->caddr.in6, ntohs(cp->cport),
 910                                 &cp->vaddr.in6, ntohs(cp->vport),
 911                                 &cp->daddr.in6, ntohs(cp->dport),
 912                                 ip_vs_state_name(cp->protocol, cp->state),
 913                                 ip_vs_origin_name(cp->flags),
 914                                 (cp->timer.expires-jiffies)/HZ);
 915                 else
 916 #endif
 917                         seq_printf(seq,
 918                                 "%-3s %08X %04X %08X %04X "
 919                                 "%08X %04X %-11s %-6s %7lu\n",
 920                                 ip_vs_proto_name(cp->protocol),
 921                                 ntohl(cp->caddr.ip), ntohs(cp->cport),
 922                                 ntohl(cp->vaddr.ip), ntohs(cp->vport),
 923                                 ntohl(cp->daddr.ip), ntohs(cp->dport),
 924                                 ip_vs_state_name(cp->protocol, cp->state),
 925                                 ip_vs_origin_name(cp->flags),
 926                                 (cp->timer.expires-jiffies)/HZ);
 927         }
 928         return 0;
 929 }
 930
 931 static const struct seq_operations ip_vs_conn_sync_seq_ops = {
 932         .start = ip_vs_conn_seq_start,
 933         .next  = ip_vs_conn_seq_next,
 934         .stop  = ip_vs_conn_seq_stop,
 935         .show  = ip_vs_conn_sync_seq_show,
 936 };
 937
 938 static int ip_vs_conn_sync_open(struct inode *inode, struct file *file)
 939 {
 940         return seq_open(file, &ip_vs_conn_sync_seq_ops);
 941 }
 942
 943 static const struct file_operations ip_vs_conn_sync_fops = {
 944         .owner   = THIS_MODULE,
 945         .open    = ip_vs_conn_sync_open,
 946         .read    = seq_read,
 947         .llseek  = seq_lseek,
 948         .release = seq_release,
 949 };
 950
 951 #endif
 952
 953
 954 /*
 955  *      Randomly drop connection entries before running out of memory
 956  */
 957 static inline int todrop_entry(struct ip_vs_conn *cp)
 958 {
 959         /*
 960          * The drop rate array needs tuning for real environments.
 961          * Called from timer bh only => no locking
 962          */
 963         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
 964         static char todrop_counter[9] = {0};
 965         int i;
 966
 967         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
 968            This will leave enough time for normal connection to get
 969            through. */
 970         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
 971                 return 0;
 972
 973         /* Don't drop the entry if its number of incoming packets is not
 974            located in [0, 8] */
 975         i = atomic_read(&cp->in_pkts);
 976         if (i > 8 || i < 0) return 0;
 977
 978         if (!todrop_rate[i]) return 0;
 979         if (--todrop_counter[i] > 0) return 0;
 980
 981         todrop_counter[i] = todrop_rate[i];
 982         return 1;
 983 }
 984
 985 /* Called from keventd and must protect itself from softirqs */
 986 void ip_vs_random_dropentry(void)
 987 {
 988         int idx;
 989         struct ip_vs_conn *cp;
 990
 991         /*
 992          * Randomly scan 1/32 of the whole table every second
 993          */
 994         for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
 995                 unsigned hash = net_random() & ip_vs_conn_tab_mask;
 996
 997                 /*
 998                  *  Lock is actually needed in this loop.
 999                  */
1000                 ct_write_lock_bh(hash);
1001
1002                 list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
1003                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
1004                                 /* connection template */
1005                                 continue;
1006
1007                         if (cp->protocol == IPPROTO_TCP) {
1008                                 switch(cp->state) {
1009                                 case IP_VS_TCP_S_SYN_RECV:
1010                                 case IP_VS_TCP_S_SYNACK:
1011                                         break;
1012
1013                                 case IP_VS_TCP_S_ESTABLISHED:
1014                                         if (todrop_entry(cp))
1015                                                 break;
1016                                         continue;
1017
1018                                 default:
1019                                         continue;
1020                                 }
1021                         } else {
1022                                 if (!todrop_entry(cp))
1023                                         continue;
1024                         }
1025
1026                         IP_VS_DBG(4, "del connection\n");
1027                         ip_vs_conn_expire_now(cp);
1028                         if (cp->control) {
1029                                 IP_VS_DBG(4, "del conn template\n");
1030                                 ip_vs_conn_expire_now(cp->control);
1031                         }
1032                 }
1033                 ct_write_unlock_bh(hash);
1034         }
1035 }
1036
1037
1038 /*
1039  *      Flush all the connection entries in the ip_vs_conn_tab
1040  */
1041 static void ip_vs_conn_flush(void)
1042 {
1043         int idx;
1044         struct ip_vs_conn *cp;
1045
1046   flush_again:
1047         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1048                 /*
1049                  *  Lock is actually needed in this loop.
1050                  */
1051                 ct_write_lock_bh(idx);
1052
1053                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
1054
1055                         IP_VS_DBG(4, "del connection\n");
1056                         ip_vs_conn_expire_now(cp);
1057                         if (cp->control) {
1058                                 IP_VS_DBG(4, "del conn template\n");
1059                                 ip_vs_conn_expire_now(cp->control);
1060                         }
1061                 }
1062                 ct_write_unlock_bh(idx);
1063         }
1064
1065         /* the counter may be not NULL, because maybe some conn entries
1066            are run by slow timer handler or unhashed but still referred */
1067         if (atomic_read(&ip_vs_conn_count) != 0) {
1068                 schedule();
1069                 goto flush_again;
1070         }
1071 }
1072
1073
1074 int __init ip_vs_conn_init(void)
1075 {
1076         int idx;
1077
1078         /* Compute size and mask */
1079         ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1080         ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1081
1082         /*
1083          * Allocate the connection hash table and initialize its list heads
1084          */
1085         ip_vs_conn_tab = vmalloc(ip_vs_conn_tab_size *
1086                                  sizeof(struct list_head));
1087         if (!ip_vs_conn_tab)
1088                 return -ENOMEM;
1089
1090         /* Allocate ip_vs_conn slab cache */
1091         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
1092                                               sizeof(struct ip_vs_conn), 0,
1093                                               SLAB_HWCACHE_ALIGN, NULL);
1094         if (!ip_vs_conn_cachep) {
1095                 vfree(ip_vs_conn_tab);
1096                 return -ENOMEM;
1097         }
1098
1099         pr_info("Connection hash table configured "
1100                 "(size=%d, memory=%ldKbytes)\n",
1101                 ip_vs_conn_tab_size,
1102                 (long)(ip_vs_conn_tab_size*sizeof(struct list_head))/1024);
1103         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
1104                   sizeof(struct ip_vs_conn));
1105
1106         for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1107                 INIT_LIST_HEAD(&ip_vs_conn_tab[idx]);
1108         }
1109
1110         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
1111                 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
1112         }
1113
1114         proc_net_fops_create(&init_net, "ip_vs_conn", 0, &ip_vs_conn_fops);
1115         proc_net_fops_create(&init_net, "ip_vs_conn_sync", 0, &ip_vs_conn_sync_fops);
1116
1117         /* calculate the random value for connection hash */
1118         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
1119
1120         return 0;
1121 }
1122
1123
1124 void ip_vs_conn_cleanup(void)
1125 {
1126         /* flush all the connection entries first */
1127         ip_vs_conn_flush();
1128
1129         /* Release the empty cache */
1130         kmem_cache_destroy(ip_vs_conn_cachep);
1131         proc_net_remove(&init_net, "ip_vs_conn");
1132         proc_net_remove(&init_net, "ip_vs_conn_sync");
1133         vfree(ip_vs_conn_tab);
1134 }