net/ipv4/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  * Version:     $Id: ip_vs_conn.c,v 1.31 2003/04/18 09:03:16 wensong Exp $
   9  *
  10  * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
  11  *              Peter Kese <peter.kese@ijs.si>
  12  *              Julian Anastasov <ja@ssi.bg>
  13  *
  14  *              This program is free software; you can redistribute it and/or
  15  *              modify it under the terms of the GNU General Public License
  16  *              as published by the Free Software Foundation; either version
  17  *              2 of the License, or (at your option) any later version.
  18  *
  19  * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
  20  * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
  21  * and others. Many code here is taken from IP MASQ code of kernel 2.2.
  22  *
  23  * Changes:
  24  *
  25  */
  26
  27 #include <linux/interrupt.h>
  28 #include <linux/in.h>
  29 #include <linux/net.h>
  30 #include <linux/kernel.h>
  31 #include <linux/module.h>
  32 #include <linux/vmalloc.h>
  33 #include <linux/proc_fs.h>              /* for proc_net_* */
  34 #include <linux/seq_file.h>
  35 #include <linux/jhash.h>
  36 #include <linux/random.h>
  37
  38 #include <net/ip_vs.h>
  39
  40
  41 /*
  42  *  Connection hash table: for input and output packets lookups of IPVS
  43  */
  44 static struct list_head *ip_vs_conn_tab;
  45
  46 /*  SLAB cache for IPVS connections */
  47 static kmem_cache_t *ip_vs_conn_cachep __read_mostly;
  48
  49 /*  counter for current IPVS connections */
  50 static atomic_t ip_vs_conn_count = ATOMIC_INIT(0);
  51
  52 /*  counter for no client port connections */
  53 static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
  54
  55 /* random value for IPVS connection hash */
  56 static unsigned int ip_vs_conn_rnd;
  57
  58 /*
  59  *  Fine locking granularity for big connection hash table
  60  */
  61 #define CT_LOCKARRAY_BITS  4
  62 #define CT_LOCKARRAY_SIZE  (1<<CT_LOCKARRAY_BITS)
  63 #define CT_LOCKARRAY_MASK  (CT_LOCKARRAY_SIZE-1)
  64
  65 struct ip_vs_aligned_lock
  66 {
  67         rwlock_t        l;
  68 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
  69
  70 /* lock array for conn table */
  71 static struct ip_vs_aligned_lock
  72 __ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
  73
  74 static inline void ct_read_lock(unsigned key)
  75 {
  76         read_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  77 }
  78
  79 static inline void ct_read_unlock(unsigned key)
  80 {
  81         read_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  82 }
  83
  84 static inline void ct_write_lock(unsigned key)
  85 {
  86         write_lock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  87 }
  88
  89 static inline void ct_write_unlock(unsigned key)
  90 {
  91         write_unlock(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  92 }
  93
  94 static inline void ct_read_lock_bh(unsigned key)
  95 {
  96         read_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
  97 }
  98
  99 static inline void ct_read_unlock_bh(unsigned key)
 100 {
 101         read_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 102 }
 103
 104 static inline void ct_write_lock_bh(unsigned key)
 105 {
 106         write_lock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 107 }
 108
 109 static inline void ct_write_unlock_bh(unsigned key)
 110 {
 111         write_unlock_bh(&__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
 112 }
 113
 114
 115 /*
 116  *      Returns hash value for IPVS connection entry
 117  */
 118 static unsigned int ip_vs_conn_hashkey(unsigned proto, __be32 addr, __be16 port)
 119 {
 120         return jhash_3words((__force u32)addr, (__force u32)port, proto, ip_vs_conn_rnd)
 121                 & IP_VS_CONN_TAB_MASK;
 122 }
 123
 124
 125 /*
 126  *      Hashes ip_vs_conn in ip_vs_conn_tab by proto,addr,port.
 127  *      returns bool success.
 128  */
 129 static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
 130 {
 131         unsigned hash;
 132         int ret;
 133
 134         /* Hash by protocol, client address and port */
 135         hash = ip_vs_conn_hashkey(cp->protocol, cp->caddr, cp->cport);
 136
 137         ct_write_lock(hash);
 138
 139         if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
 140                 list_add(&cp->c_list, &ip_vs_conn_tab[hash]);
 141                 cp->flags |= IP_VS_CONN_F_HASHED;
 142                 atomic_inc(&cp->refcnt);
 143                 ret = 1;
 144         } else {
 145                 IP_VS_ERR("ip_vs_conn_hash(): request for already hashed, "
 146                           "called from %p\n", __builtin_return_address(0));
 147                 ret = 0;
 148         }
 149
 150         ct_write_unlock(hash);
 151
 152         return ret;
 153 }
 154
 155
 156 /*
 157  *      UNhashes ip_vs_conn from ip_vs_conn_tab.
 158  *      returns bool success.
 159  */
 160 static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
 161 {
 162         unsigned hash;
 163         int ret;
 164
 165         /* unhash it and decrease its reference counter */
 166         hash = ip_vs_conn_hashkey(cp->protocol, cp->caddr, cp->cport);
 167
 168         ct_write_lock(hash);
 169
 170         if (cp->flags & IP_VS_CONN_F_HASHED) {
 171                 list_del(&cp->c_list);
 172                 cp->flags &= ~IP_VS_CONN_F_HASHED;
 173                 atomic_dec(&cp->refcnt);
 174                 ret = 1;
 175         } else
 176                 ret = 0;
 177
 178         ct_write_unlock(hash);
 179
 180         return ret;
 181 }
 182
 183
 184 /*
 185  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
 186  *  Called for pkts coming from OUTside-to-INside.
 187  *      s_addr, s_port: pkt source address (foreign host)
 188  *      d_addr, d_port: pkt dest address (load balancer)
 189  */
 190 static inline struct ip_vs_conn *__ip_vs_conn_in_get
 191 (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port)
 192 {
 193         unsigned hash;
 194         struct ip_vs_conn *cp;
 195
 196         hash = ip_vs_conn_hashkey(protocol, s_addr, s_port);
 197
 198         ct_read_lock(hash);
 199
 200         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
 201                 if (s_addr==cp->caddr && s_port==cp->cport &&
 202                     d_port==cp->vport && d_addr==cp->vaddr &&
 203                     ((!s_port) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
 204                     protocol==cp->protocol) {
 205                         /* HIT */
 206                         atomic_inc(&cp->refcnt);
 207                         ct_read_unlock(hash);
 208                         return cp;
 209                 }
 210         }
 211
 212         ct_read_unlock(hash);
 213
 214         return NULL;
 215 }
 216
 217 struct ip_vs_conn *ip_vs_conn_in_get
 218 (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port)
 219 {
 220         struct ip_vs_conn *cp;
 221
 222         cp = __ip_vs_conn_in_get(protocol, s_addr, s_port, d_addr, d_port);
 223         if (!cp && atomic_read(&ip_vs_conn_no_cport_cnt))
 224                 cp = __ip_vs_conn_in_get(protocol, s_addr, 0, d_addr, d_port);
 225
 226         IP_VS_DBG(9, "lookup/in %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
 227                   ip_vs_proto_name(protocol),
 228                   NIPQUAD(s_addr), ntohs(s_port),
 229                   NIPQUAD(d_addr), ntohs(d_port),
 230                   cp?"hit":"not hit");
 231
 232         return cp;
 233 }
 234
 235 /* Get reference to connection template */
 236 struct ip_vs_conn *ip_vs_ct_in_get
 237 (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port)
 238 {
 239         unsigned hash;
 240         struct ip_vs_conn *cp;
 241
 242         hash = ip_vs_conn_hashkey(protocol, s_addr, s_port);
 243
 244         ct_read_lock(hash);
 245
 246         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
 247                 if (s_addr==cp->caddr && s_port==cp->cport &&
 248                     d_port==cp->vport && d_addr==cp->vaddr &&
 249                     cp->flags & IP_VS_CONN_F_TEMPLATE &&
 250                     protocol==cp->protocol) {
 251                         /* HIT */
 252                         atomic_inc(&cp->refcnt);
 253                         goto out;
 254                 }
 255         }
 256         cp = NULL;
 257
 258   out:
 259         ct_read_unlock(hash);
 260
 261         IP_VS_DBG(9, "template lookup/in %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
 262                   ip_vs_proto_name(protocol),
 263                   NIPQUAD(s_addr), ntohs(s_port),
 264                   NIPQUAD(d_addr), ntohs(d_port),
 265                   cp?"hit":"not hit");
 266
 267         return cp;
 268 }
 269
 270 /*
 271  *  Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
 272  *  Called for pkts coming from inside-to-OUTside.
 273  *      s_addr, s_port: pkt source address (inside host)
 274  *      d_addr, d_port: pkt dest address (foreign host)
 275  */
 276 struct ip_vs_conn *ip_vs_conn_out_get
 277 (int protocol, __be32 s_addr, __be16 s_port, __be32 d_addr, __be16 d_port)
 278 {
 279         unsigned hash;
 280         struct ip_vs_conn *cp, *ret=NULL;
 281
 282         /*
 283          *      Check for "full" addressed entries
 284          */
 285         hash = ip_vs_conn_hashkey(protocol, d_addr, d_port);
 286
 287         ct_read_lock(hash);
 288
 289         list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
 290                 if (d_addr == cp->caddr && d_port == cp->cport &&
 291                     s_port == cp->dport && s_addr == cp->daddr &&
 292                     protocol == cp->protocol) {
 293                         /* HIT */
 294                         atomic_inc(&cp->refcnt);
 295                         ret = cp;
 296                         break;
 297                 }
 298         }
 299
 300         ct_read_unlock(hash);
 301
 302         IP_VS_DBG(9, "lookup/out %s %u.%u.%u.%u:%d->%u.%u.%u.%u:%d %s\n",
 303                   ip_vs_proto_name(protocol),
 304                   NIPQUAD(s_addr), ntohs(s_port),
 305                   NIPQUAD(d_addr), ntohs(d_port),
 306                   ret?"hit":"not hit");
 307
 308         return ret;
 309 }
 310
 311
 312 /*
 313  *      Put back the conn and restart its timer with its timeout
 314  */
 315 void ip_vs_conn_put(struct ip_vs_conn *cp)
 316 {
 317         /* reset it expire in its timeout */
 318         mod_timer(&cp->timer, jiffies+cp->timeout);
 319
 320         __ip_vs_conn_put(cp);
 321 }
 322
 323
 324 /*
 325  *      Fill a no_client_port connection with a client port number
 326  */
 327 void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
 328 {
 329         if (ip_vs_conn_unhash(cp)) {
 330                 spin_lock(&cp->lock);
 331                 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
 332                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 333                         cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
 334                         cp->cport = cport;
 335                 }
 336                 spin_unlock(&cp->lock);
 337
 338                 /* hash on new dport */
 339                 ip_vs_conn_hash(cp);
 340         }
 341 }
 342
 343
 344 /*
 345  *      Bind a connection entry with the corresponding packet_xmit.
 346  *      Called by ip_vs_conn_new.
 347  */
 348 static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
 349 {
 350         switch (IP_VS_FWD_METHOD(cp)) {
 351         case IP_VS_CONN_F_MASQ:
 352                 cp->packet_xmit = ip_vs_nat_xmit;
 353                 break;
 354
 355         case IP_VS_CONN_F_TUNNEL:
 356                 cp->packet_xmit = ip_vs_tunnel_xmit;
 357                 break;
 358
 359         case IP_VS_CONN_F_DROUTE:
 360                 cp->packet_xmit = ip_vs_dr_xmit;
 361                 break;
 362
 363         case IP_VS_CONN_F_LOCALNODE:
 364                 cp->packet_xmit = ip_vs_null_xmit;
 365                 break;
 366
 367         case IP_VS_CONN_F_BYPASS:
 368                 cp->packet_xmit = ip_vs_bypass_xmit;
 369                 break;
 370         }
 371 }
 372
 373
 374 static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
 375 {
 376         return atomic_read(&dest->activeconns)
 377                 + atomic_read(&dest->inactconns);
 378 }
 379
 380 /*
 381  *      Bind a connection entry with a virtual service destination
 382  *      Called just after a new connection entry is created.
 383  */
 384 static inline void
 385 ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
 386 {
 387         /* if dest is NULL, then return directly */
 388         if (!dest)
 389                 return;
 390
 391         /* Increase the refcnt counter of the dest */
 392         atomic_inc(&dest->refcnt);
 393
 394         /* Bind with the destination and its corresponding transmitter */
 395         cp->flags |= atomic_read(&dest->conn_flags);
 396         cp->dest = dest;
 397
 398         IP_VS_DBG(7, "Bind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
 399                   "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 400                   "dest->refcnt:%d\n",
 401                   ip_vs_proto_name(cp->protocol),
 402                   NIPQUAD(cp->caddr), ntohs(cp->cport),
 403                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
 404                   NIPQUAD(cp->daddr), ntohs(cp->dport),
 405                   ip_vs_fwd_tag(cp), cp->state,
 406                   cp->flags, atomic_read(&cp->refcnt),
 407                   atomic_read(&dest->refcnt));
 408
 409         /* Update the connection counters */
 410         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
 411                 /* It is a normal connection, so increase the inactive
 412                    connection counter because it is in TCP SYNRECV
 413                    state (inactive) or other protocol inacive state */
 414                 atomic_inc(&dest->inactconns);
 415         } else {
 416                 /* It is a persistent connection/template, so increase
 417                    the peristent connection counter */
 418                 atomic_inc(&dest->persistconns);
 419         }
 420
 421         if (dest->u_threshold != 0 &&
 422             ip_vs_dest_totalconns(dest) >= dest->u_threshold)
 423                 dest->flags |= IP_VS_DEST_F_OVERLOAD;
 424 }
 425
 426
 427 /*
 428  *      Unbind a connection entry with its VS destination
 429  *      Called by the ip_vs_conn_expire function.
 430  */
 431 static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
 432 {
 433         struct ip_vs_dest *dest = cp->dest;
 434
 435         if (!dest)
 436                 return;
 437
 438         IP_VS_DBG(7, "Unbind-dest %s c:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
 439                   "d:%u.%u.%u.%u:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
 440                   "dest->refcnt:%d\n",
 441                   ip_vs_proto_name(cp->protocol),
 442                   NIPQUAD(cp->caddr), ntohs(cp->cport),
 443                   NIPQUAD(cp->vaddr), ntohs(cp->vport),
 444                   NIPQUAD(cp->daddr), ntohs(cp->dport),
 445                   ip_vs_fwd_tag(cp), cp->state,
 446                   cp->flags, atomic_read(&cp->refcnt),
 447                   atomic_read(&dest->refcnt));
 448
 449         /* Update the connection counters */
 450         if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
 451                 /* It is a normal connection, so decrease the inactconns
 452                    or activeconns counter */
 453                 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
 454                         atomic_dec(&dest->inactconns);
 455                 } else {
 456                         atomic_dec(&dest->activeconns);
 457                 }
 458         } else {
 459                 /* It is a persistent connection/template, so decrease
 460                    the peristent connection counter */
 461                 atomic_dec(&dest->persistconns);
 462         }
 463
 464         if (dest->l_threshold != 0) {
 465                 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
 466                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 467         } else if (dest->u_threshold != 0) {
 468                 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
 469                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 470         } else {
 471                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 472                         dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
 473         }
 474
 475         /*
 476          * Simply decrease the refcnt of the dest, because the
 477          * dest will be either in service's destination list
 478          * or in the trash.
 479          */
 480         atomic_dec(&dest->refcnt);
 481 }
 482
 483
 484 /*
 485  *      Checking if the destination of a connection template is available.
 486  *      If available, return 1, otherwise invalidate this connection
 487  *      template and return 0.
 488  */
 489 int ip_vs_check_template(struct ip_vs_conn *ct)
 490 {
 491         struct ip_vs_dest *dest = ct->dest;
 492
 493         /*
 494          * Checking the dest server status.
 495          */
 496         if ((dest == NULL) ||
 497             !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
 498             (sysctl_ip_vs_expire_quiescent_template &&
 499              (atomic_read(&dest->weight) == 0))) {
 500                 IP_VS_DBG(9, "check_template: dest not available for "
 501                           "protocol %s s:%u.%u.%u.%u:%d v:%u.%u.%u.%u:%d "
 502                           "-> d:%u.%u.%u.%u:%d\n",
 503                           ip_vs_proto_name(ct->protocol),
 504                           NIPQUAD(ct->caddr), ntohs(ct->cport),
 505                           NIPQUAD(ct->vaddr), ntohs(ct->vport),
 506                           NIPQUAD(ct->daddr), ntohs(ct->dport));
 507
 508                 /*
 509                  * Invalidate the connection template
 510                  */
 511                 if (ct->vport != htons(0xffff)) {
 512                         if (ip_vs_conn_unhash(ct)) {
 513                                 ct->dport = htons(0xffff);
 514                                 ct->vport = htons(0xffff);
 515                                 ct->cport = 0;
 516                                 ip_vs_conn_hash(ct);
 517                         }
 518                 }
 519
 520                 /*
 521                  * Simply decrease the refcnt of the template,
 522                  * don't restart its timer.
 523                  */
 524                 atomic_dec(&ct->refcnt);
 525                 return 0;
 526         }
 527         return 1;
 528 }
 529
 530 static void ip_vs_conn_expire(unsigned long data)
 531 {
 532         struct ip_vs_conn *cp = (struct ip_vs_conn *)data;
 533
 534         cp->timeout = 60*HZ;
 535
 536         /*
 537          *      hey, I'm using it
 538          */
 539         atomic_inc(&cp->refcnt);
 540
 541         /*
 542          *      do I control anybody?
 543          */
 544         if (atomic_read(&cp->n_control))
 545                 goto expire_later;
 546
 547         /*
 548          *      unhash it if it is hashed in the conn table
 549          */
 550         if (!ip_vs_conn_unhash(cp))
 551                 goto expire_later;
 552
 553         /*
 554          *      refcnt==1 implies I'm the only one referrer
 555          */
 556         if (likely(atomic_read(&cp->refcnt) == 1)) {
 557                 /* delete the timer if it is activated by other users */
 558                 if (timer_pending(&cp->timer))
 559                         del_timer(&cp->timer);
 560
 561                 /* does anybody control me? */
 562                 if (cp->control)
 563                         ip_vs_control_del(cp);
 564
 565                 if (unlikely(cp->app != NULL))
 566                         ip_vs_unbind_app(cp);
 567                 ip_vs_unbind_dest(cp);
 568                 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
 569                         atomic_dec(&ip_vs_conn_no_cport_cnt);
 570                 atomic_dec(&ip_vs_conn_count);
 571
 572                 kmem_cache_free(ip_vs_conn_cachep, cp);
 573                 return;
 574         }
 575
 576         /* hash it back to the table */
 577         ip_vs_conn_hash(cp);
 578
 579   expire_later:
 580         IP_VS_DBG(7, "delayed: conn->refcnt-1=%d conn->n_control=%d\n",
 581                   atomic_read(&cp->refcnt)-1,
 582                   atomic_read(&cp->n_control));
 583
 584         ip_vs_conn_put(cp);
 585 }
 586
 587
 588 void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
 589 {
 590         if (del_timer(&cp->timer))
 591                 mod_timer(&cp->timer, jiffies);
 592 }
 593
 594
 595 /*
 596  *      Create a new connection entry and hash it into the ip_vs_conn_tab
 597  */
 598 struct ip_vs_conn *
 599 ip_vs_conn_new(int proto, __be32 caddr, __be16 cport, __be32 vaddr, __be16 vport,
 600                __be32 daddr, __be16 dport, unsigned flags,
 601                struct ip_vs_dest *dest)
 602 {
 603         struct ip_vs_conn *cp;
 604         struct ip_vs_protocol *pp = ip_vs_proto_get(proto);
 605
 606         cp = kmem_cache_alloc(ip_vs_conn_cachep, GFP_ATOMIC);
 607         if (cp == NULL) {
 608                 IP_VS_ERR_RL("ip_vs_conn_new: no memory available.\n");
 609                 return NULL;
 610         }
 611
 612         memset(cp, 0, sizeof(*cp));
 613         INIT_LIST_HEAD(&cp->c_list);
 614         init_timer(&cp->timer);
 615         cp->timer.data     = (unsigned long)cp;
 616         cp->timer.function = ip_vs_conn_expire;
 617         cp->protocol       = proto;
 618         cp->caddr          = caddr;
 619         cp->cport          = cport;
 620         cp->vaddr          = vaddr;
 621         cp->vport          = vport;
 622         cp->daddr          = daddr;
 623         cp->dport          = dport;
 624         cp->flags          = flags;
 625         spin_lock_init(&cp->lock);
 626
 627         /*
 628          * Set the entry is referenced by the current thread before hashing
 629          * it in the table, so that other thread run ip_vs_random_dropentry
 630          * but cannot drop this entry.
 631          */
 632         atomic_set(&cp->refcnt, 1);
 633
 634         atomic_set(&cp->n_control, 0);
 635         atomic_set(&cp->in_pkts, 0);
 636
 637         atomic_inc(&ip_vs_conn_count);
 638         if (flags & IP_VS_CONN_F_NO_CPORT)
 639                 atomic_inc(&ip_vs_conn_no_cport_cnt);
 640
 641         /* Bind the connection with a destination server */
 642         ip_vs_bind_dest(cp, dest);
 643
 644         /* Set its state and timeout */
 645         cp->state = 0;
 646         cp->timeout = 3*HZ;
 647
 648         /* Bind its packet transmitter */
 649         ip_vs_bind_xmit(cp);
 650
 651         if (unlikely(pp && atomic_read(&pp->appcnt)))
 652                 ip_vs_bind_app(cp, pp);
 653
 654         /* Hash it in the ip_vs_conn_tab finally */
 655         ip_vs_conn_hash(cp);
 656
 657         return cp;
 658 }
 659
 660
 661 /*
 662  *      /proc/net/ip_vs_conn entries
 663  */
 664 #ifdef CONFIG_PROC_FS
 665
 666 static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
 667 {
 668         int idx;
 669         struct ip_vs_conn *cp;
 670
 671         for(idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) {
 672                 ct_read_lock_bh(idx);
 673                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
 674                         if (pos-- == 0) {
 675                                 seq->private = &ip_vs_conn_tab[idx];
 676                                 return cp;
 677                         }
 678                 }
 679                 ct_read_unlock_bh(idx);
 680         }
 681
 682         return NULL;
 683 }
 684
 685 static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
 686 {
 687         seq->private = NULL;
 688         return *pos ? ip_vs_conn_array(seq, *pos - 1) :SEQ_START_TOKEN;
 689 }
 690
 691 static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 692 {
 693         struct ip_vs_conn *cp = v;
 694         struct list_head *e, *l = seq->private;
 695         int idx;
 696
 697         ++*pos;
 698         if (v == SEQ_START_TOKEN)
 699                 return ip_vs_conn_array(seq, 0);
 700
 701         /* more on same hash chain? */
 702         if ((e = cp->c_list.next) != l)
 703                 return list_entry(e, struct ip_vs_conn, c_list);
 704
 705         idx = l - ip_vs_conn_tab;
 706         ct_read_unlock_bh(idx);
 707
 708         while (++idx < IP_VS_CONN_TAB_SIZE) {
 709                 ct_read_lock_bh(idx);
 710                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
 711                         seq->private = &ip_vs_conn_tab[idx];
 712                         return cp;
 713                 }
 714                 ct_read_unlock_bh(idx);
 715         }
 716         seq->private = NULL;
 717         return NULL;
 718 }
 719
 720 static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
 721 {
 722         struct list_head *l = seq->private;
 723
 724         if (l)
 725                 ct_read_unlock_bh(l - ip_vs_conn_tab);
 726 }
 727
 728 static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
 729 {
 730
 731         if (v == SEQ_START_TOKEN)
 732                 seq_puts(seq,
 733    "Pro FromIP   FPrt ToIP     TPrt DestIP   DPrt State       Expires\n");
 734         else {
 735                 const struct ip_vs_conn *cp = v;
 736
 737                 seq_printf(seq,
 738                         "%-3s %08X %04X %08X %04X %08X %04X %-11s %7lu\n",
 739                                 ip_vs_proto_name(cp->protocol),
 740                                 ntohl(cp->caddr), ntohs(cp->cport),
 741                                 ntohl(cp->vaddr), ntohs(cp->vport),
 742                                 ntohl(cp->daddr), ntohs(cp->dport),
 743                                 ip_vs_state_name(cp->protocol, cp->state),
 744                                 (cp->timer.expires-jiffies)/HZ);
 745         }
 746         return 0;
 747 }
 748
 749 static struct seq_operations ip_vs_conn_seq_ops = {
 750         .start = ip_vs_conn_seq_start,
 751         .next  = ip_vs_conn_seq_next,
 752         .stop  = ip_vs_conn_seq_stop,
 753         .show  = ip_vs_conn_seq_show,
 754 };
 755
 756 static int ip_vs_conn_open(struct inode *inode, struct file *file)
 757 {
 758         return seq_open(file, &ip_vs_conn_seq_ops);
 759 }
 760
 761 static struct file_operations ip_vs_conn_fops = {
 762         .owner   = THIS_MODULE,
 763         .open    = ip_vs_conn_open,
 764         .read    = seq_read,
 765         .llseek  = seq_lseek,
 766         .release = seq_release,
 767 };
 768 #endif
 769
 770
 771 /*
 772  *      Randomly drop connection entries before running out of memory
 773  */
 774 static inline int todrop_entry(struct ip_vs_conn *cp)
 775 {
 776         /*
 777          * The drop rate array needs tuning for real environments.
 778          * Called from timer bh only => no locking
 779          */
 780         static const char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
 781         static char todrop_counter[9] = {0};
 782         int i;
 783
 784         /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
 785            This will leave enough time for normal connection to get
 786            through. */
 787         if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
 788                 return 0;
 789
 790         /* Don't drop the entry if its number of incoming packets is not
 791            located in [0, 8] */
 792         i = atomic_read(&cp->in_pkts);
 793         if (i > 8 || i < 0) return 0;
 794
 795         if (!todrop_rate[i]) return 0;
 796         if (--todrop_counter[i] > 0) return 0;
 797
 798         todrop_counter[i] = todrop_rate[i];
 799         return 1;
 800 }
 801
 802 /* Called from keventd and must protect itself from softirqs */
 803 void ip_vs_random_dropentry(void)
 804 {
 805         int idx;
 806         struct ip_vs_conn *cp;
 807
 808         /*
 809          * Randomly scan 1/32 of the whole table every second
 810          */
 811         for (idx = 0; idx < (IP_VS_CONN_TAB_SIZE>>5); idx++) {
 812                 unsigned hash = net_random() & IP_VS_CONN_TAB_MASK;
 813
 814                 /*
 815                  *  Lock is actually needed in this loop.
 816                  */
 817                 ct_write_lock_bh(hash);
 818
 819                 list_for_each_entry(cp, &ip_vs_conn_tab[hash], c_list) {
 820                         if (cp->flags & IP_VS_CONN_F_TEMPLATE)
 821                                 /* connection template */
 822                                 continue;
 823
 824                         if (cp->protocol == IPPROTO_TCP) {
 825                                 switch(cp->state) {
 826                                 case IP_VS_TCP_S_SYN_RECV:
 827                                 case IP_VS_TCP_S_SYNACK:
 828                                         break;
 829
 830                                 case IP_VS_TCP_S_ESTABLISHED:
 831                                         if (todrop_entry(cp))
 832                                                 break;
 833                                         continue;
 834
 835                                 default:
 836                                         continue;
 837                                 }
 838                         } else {
 839                                 if (!todrop_entry(cp))
 840                                         continue;
 841                         }
 842
 843                         IP_VS_DBG(4, "del connection\n");
 844                         ip_vs_conn_expire_now(cp);
 845                         if (cp->control) {
 846                                 IP_VS_DBG(4, "del conn template\n");
 847                                 ip_vs_conn_expire_now(cp->control);
 848                         }
 849                 }
 850                 ct_write_unlock_bh(hash);
 851         }
 852 }
 853
 854
 855 /*
 856  *      Flush all the connection entries in the ip_vs_conn_tab
 857  */
 858 static void ip_vs_conn_flush(void)
 859 {
 860         int idx;
 861         struct ip_vs_conn *cp;
 862
 863   flush_again:
 864         for (idx=0; idx<IP_VS_CONN_TAB_SIZE; idx++) {
 865                 /*
 866                  *  Lock is actually needed in this loop.
 867                  */
 868                 ct_write_lock_bh(idx);
 869
 870                 list_for_each_entry(cp, &ip_vs_conn_tab[idx], c_list) {
 871
 872                         IP_VS_DBG(4, "del connection\n");
 873                         ip_vs_conn_expire_now(cp);
 874                         if (cp->control) {
 875                                 IP_VS_DBG(4, "del conn template\n");
 876                                 ip_vs_conn_expire_now(cp->control);
 877                         }
 878                 }
 879                 ct_write_unlock_bh(idx);
 880         }
 881
 882         /* the counter may be not NULL, because maybe some conn entries
 883            are run by slow timer handler or unhashed but still referred */
 884         if (atomic_read(&ip_vs_conn_count) != 0) {
 885                 schedule();
 886                 goto flush_again;
 887         }
 888 }
 889
 890
 891 int ip_vs_conn_init(void)
 892 {
 893         int idx;
 894
 895         /*
 896          * Allocate the connection hash table and initialize its list heads
 897          */
 898         ip_vs_conn_tab = vmalloc(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head));
 899         if (!ip_vs_conn_tab)
 900                 return -ENOMEM;
 901
 902         /* Allocate ip_vs_conn slab cache */
 903         ip_vs_conn_cachep = kmem_cache_create("ip_vs_conn",
 904                                               sizeof(struct ip_vs_conn), 0,
 905                                               SLAB_HWCACHE_ALIGN, NULL, NULL);
 906         if (!ip_vs_conn_cachep) {
 907                 vfree(ip_vs_conn_tab);
 908                 return -ENOMEM;
 909         }
 910
 911         IP_VS_INFO("Connection hash table configured "
 912                    "(size=%d, memory=%ldKbytes)\n",
 913                    IP_VS_CONN_TAB_SIZE,
 914                    (long)(IP_VS_CONN_TAB_SIZE*sizeof(struct list_head))/1024);
 915         IP_VS_DBG(0, "Each connection entry needs %Zd bytes at least\n",
 916                   sizeof(struct ip_vs_conn));
 917
 918         for (idx = 0; idx < IP_VS_CONN_TAB_SIZE; idx++) {
 919                 INIT_LIST_HEAD(&ip_vs_conn_tab[idx]);
 920         }
 921
 922         for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++)  {
 923                 rwlock_init(&__ip_vs_conntbl_lock_array[idx].l);
 924         }
 925
 926         proc_net_fops_create("ip_vs_conn", 0, &ip_vs_conn_fops);
 927
 928         /* calculate the random value for connection hash */
 929         get_random_bytes(&ip_vs_conn_rnd, sizeof(ip_vs_conn_rnd));
 930
 931         return 0;
 932 }
 933
 934
 935 void ip_vs_conn_cleanup(void)
 936 {
 937         /* flush all the connection entries first */
 938         ip_vs_conn_flush();
 939
 940         /* Release the empty cache */
 941         kmem_cache_destroy(ip_vs_conn_cachep);
 942         proc_net_remove("ip_vs_conn");
 943         vfree(ip_vs_conn_tab);
 944 }