SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/ipv4/ipvs/ip_vs_lblcr.c

   1 /*
   2  * IPVS:        Locality-Based Least-Connection with Replication scheduler
   3  *
   4  * Version:     $Id: ip_vs_lblcr.c,v 1.11 2002/09/15 08:14:08 wensong Exp $
   5  *
   6  * Authors:     Wensong Zhang <wensong@gnuchina.org>
   7  *
   8  *              This program is free software; you can redistribute it and/or
   9  *              modify it under the terms of the GNU General Public License
  10  *              as published by the Free Software Foundation; either version
  11  *              2 of the License, or (at your option) any later version.
  12  *
  13  * Changes:
  14  *     Julian Anastasov        :    Added the missing (dest->weight>0)
  15  *                                  condition in the ip_vs_dest_set_max.
  16  *
  17  */
  18
  19 /*
  20  * The lblc/r algorithm is as follows (pseudo code):
  21  *
  22  *       if serverSet[dest_ip] is null then
  23  *               n, serverSet[dest_ip] <- {weighted least-conn node};
  24  *       else
  25  *               n <- {least-conn (alive) node in serverSet[dest_ip]};
  26  *               if (n is null) OR
  27  *                  (n.conns>n.weight AND
  28  *                   there is a node m with m.conns<m.weight/2) then
  29  *                   n <- {weighted least-conn node};
  30  *                   add n to serverSet[dest_ip];
  31  *               if |serverSet[dest_ip]| > 1 AND
  32  *                   now - serverSet[dest_ip].lastMod > T then
  33  *                   m <- {most conn node in serverSet[dest_ip]};
  34  *                   remove m from serverSet[dest_ip];
  35  *       if serverSet[dest_ip] changed then
  36  *               serverSet[dest_ip].lastMod <- now;
  37  *
  38  *       return n;
  39  *
  40  */
  41
  42 #include <linux/ip.h>
  43 #include <linux/module.h>
  44 #include <linux/kernel.h>
  45 #include <linux/skbuff.h>
  46 #include <linux/jiffies.h>
  47
  48 /* for sysctl */
  49 #include <linux/fs.h>
  50 #include <linux/sysctl.h>
  51 #include <net/net_namespace.h>
  52
  53 #include <net/ip_vs.h>
  54
  55
  56 /*
  57  *    It is for garbage collection of stale IPVS lblcr entries,
  58  *    when the table is full.
  59  */
  60 #define CHECK_EXPIRE_INTERVAL   (60*HZ)
  61 #define ENTRY_TIMEOUT           (6*60*HZ)
  62
  63 /*
  64  *    It is for full expiration check.
  65  *    When there is no partial expiration check (garbage collection)
  66  *    in a half hour, do a full expiration check to collect stale
  67  *    entries that haven't been touched for a day.
  68  */
  69 #define COUNT_FOR_FULL_EXPIRATION   30
  70 static int sysctl_ip_vs_lblcr_expiration = 24*60*60*HZ;
  71
  72
  73 /*
  74  *     for IPVS lblcr entry hash table
  75  */
  76 #ifndef CONFIG_IP_VS_LBLCR_TAB_BITS
  77 #define CONFIG_IP_VS_LBLCR_TAB_BITS      10
  78 #endif
  79 #define IP_VS_LBLCR_TAB_BITS     CONFIG_IP_VS_LBLCR_TAB_BITS
  80 #define IP_VS_LBLCR_TAB_SIZE     (1 << IP_VS_LBLCR_TAB_BITS)
  81 #define IP_VS_LBLCR_TAB_MASK     (IP_VS_LBLCR_TAB_SIZE - 1)
  82
  83
  84 /*
  85  *      IPVS destination set structure and operations
  86  */
  87 struct ip_vs_dest_list {
  88         struct ip_vs_dest_list  *next;          /* list link */
  89         struct ip_vs_dest       *dest;          /* destination server */
  90 };
  91
  92 struct ip_vs_dest_set {
  93         atomic_t                size;           /* set size */
  94         unsigned long           lastmod;        /* last modified time */
  95         struct ip_vs_dest_list  *list;          /* destination list */
  96         rwlock_t                lock;           /* lock for this list */
  97 };
  98
  99
 100 static struct ip_vs_dest_list *
 101 ip_vs_dest_set_insert(struct ip_vs_dest_set *set, struct ip_vs_dest *dest)
 102 {
 103         struct ip_vs_dest_list *e;
 104
 105         for (e=set->list; e!=NULL; e=e->next) {
 106                 if (e->dest == dest)
 107                         /* already existed */
 108                         return NULL;
 109         }
 110
 111         e = kmalloc(sizeof(struct ip_vs_dest_list), GFP_ATOMIC);
 112         if (e == NULL) {
 113                 IP_VS_ERR("ip_vs_dest_set_insert(): no memory\n");
 114                 return NULL;
 115         }
 116
 117         atomic_inc(&dest->refcnt);
 118         e->dest = dest;
 119
 120         /* link it to the list */
 121         write_lock(&set->lock);
 122         e->next = set->list;
 123         set->list = e;
 124         atomic_inc(&set->size);
 125         write_unlock(&set->lock);
 126
 127         set->lastmod = jiffies;
 128         return e;
 129 }
 130
 131 static void
 132 ip_vs_dest_set_erase(struct ip_vs_dest_set *set, struct ip_vs_dest *dest)
 133 {
 134         struct ip_vs_dest_list *e, **ep;
 135
 136         write_lock(&set->lock);
 137         for (ep=&set->list, e=*ep; e!=NULL; e=*ep) {
 138                 if (e->dest == dest) {
 139                         /* HIT */
 140                         *ep = e->next;
 141                         atomic_dec(&set->size);
 142                         set->lastmod = jiffies;
 143                         atomic_dec(&e->dest->refcnt);
 144                         kfree(e);
 145                         break;
 146                 }
 147                 ep = &e->next;
 148         }
 149         write_unlock(&set->lock);
 150 }
 151
 152 static void ip_vs_dest_set_eraseall(struct ip_vs_dest_set *set)
 153 {
 154         struct ip_vs_dest_list *e, **ep;
 155
 156         write_lock(&set->lock);
 157         for (ep=&set->list, e=*ep; e!=NULL; e=*ep) {
 158                 *ep = e->next;
 159                 /*
 160                  * We don't kfree dest because it is refered either
 161                  * by its service or by the trash dest list.
 162                  */
 163                 atomic_dec(&e->dest->refcnt);
 164                 kfree(e);
 165         }
 166         write_unlock(&set->lock);
 167 }
 168
 169 /* get weighted least-connection node in the destination set */
 170 static inline struct ip_vs_dest *ip_vs_dest_set_min(struct ip_vs_dest_set *set)
 171 {
 172         register struct ip_vs_dest_list *e;
 173         struct ip_vs_dest *dest, *least;
 174         int loh, doh;
 175
 176         if (set == NULL)
 177                 return NULL;
 178
 179         read_lock(&set->lock);
 180         /* select the first destination server, whose weight > 0 */
 181         for (e=set->list; e!=NULL; e=e->next) {
 182                 least = e->dest;
 183                 if (least->flags & IP_VS_DEST_F_OVERLOAD)
 184                         continue;
 185
 186                 if ((atomic_read(&least->weight) > 0)
 187                     && (least->flags & IP_VS_DEST_F_AVAILABLE)) {
 188                         loh = atomic_read(&least->activeconns) * 50
 189                                 + atomic_read(&least->inactconns);
 190                         goto nextstage;
 191                 }
 192         }
 193         read_unlock(&set->lock);
 194         return NULL;
 195
 196         /* find the destination with the weighted least load */
 197   nextstage:
 198         for (e=e->next; e!=NULL; e=e->next) {
 199                 dest = e->dest;
 200                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 201                         continue;
 202
 203                 doh = atomic_read(&dest->activeconns) * 50
 204                         + atomic_read(&dest->inactconns);
 205                 if ((loh * atomic_read(&dest->weight) >
 206                      doh * atomic_read(&least->weight))
 207                     && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 208                         least = dest;
 209                         loh = doh;
 210                 }
 211         }
 212         read_unlock(&set->lock);
 213
 214         IP_VS_DBG(6, "ip_vs_dest_set_min: server %d.%d.%d.%d:%d "
 215                   "activeconns %d refcnt %d weight %d overhead %d\n",
 216                   NIPQUAD(least->addr), ntohs(least->port),
 217                   atomic_read(&least->activeconns),
 218                   atomic_read(&least->refcnt),
 219                   atomic_read(&least->weight), loh);
 220         return least;
 221 }
 222
 223
 224 /* get weighted most-connection node in the destination set */
 225 static inline struct ip_vs_dest *ip_vs_dest_set_max(struct ip_vs_dest_set *set)
 226 {
 227         register struct ip_vs_dest_list *e;
 228         struct ip_vs_dest *dest, *most;
 229         int moh, doh;
 230
 231         if (set == NULL)
 232                 return NULL;
 233
 234         read_lock(&set->lock);
 235         /* select the first destination server, whose weight > 0 */
 236         for (e=set->list; e!=NULL; e=e->next) {
 237                 most = e->dest;
 238                 if (atomic_read(&most->weight) > 0) {
 239                         moh = atomic_read(&most->activeconns) * 50
 240                                 + atomic_read(&most->inactconns);
 241                         goto nextstage;
 242                 }
 243         }
 244         read_unlock(&set->lock);
 245         return NULL;
 246
 247         /* find the destination with the weighted most load */
 248   nextstage:
 249         for (e=e->next; e!=NULL; e=e->next) {
 250                 dest = e->dest;
 251                 doh = atomic_read(&dest->activeconns) * 50
 252                         + atomic_read(&dest->inactconns);
 253                 /* moh/mw < doh/dw ==> moh*dw < doh*mw, where mw,dw>0 */
 254                 if ((moh * atomic_read(&dest->weight) <
 255                      doh * atomic_read(&most->weight))
 256                     && (atomic_read(&dest->weight) > 0)) {
 257                         most = dest;
 258                         moh = doh;
 259                 }
 260         }
 261         read_unlock(&set->lock);
 262
 263         IP_VS_DBG(6, "ip_vs_dest_set_max: server %d.%d.%d.%d:%d "
 264                   "activeconns %d refcnt %d weight %d overhead %d\n",
 265                   NIPQUAD(most->addr), ntohs(most->port),
 266                   atomic_read(&most->activeconns),
 267                   atomic_read(&most->refcnt),
 268                   atomic_read(&most->weight), moh);
 269         return most;
 270 }
 271
 272
 273 /*
 274  *      IPVS lblcr entry represents an association between destination
 275  *      IP address and its destination server set
 276  */
 277 struct ip_vs_lblcr_entry {
 278         struct list_head        list;
 279         __be32                   addr;           /* destination IP address */
 280         struct ip_vs_dest_set   set;            /* destination server set */
 281         unsigned long           lastuse;        /* last used time */
 282 };
 283
 284
 285 /*
 286  *      IPVS lblcr hash table
 287  */
 288 struct ip_vs_lblcr_table {
 289         rwlock_t                lock;           /* lock for this table */
 290         struct list_head        bucket[IP_VS_LBLCR_TAB_SIZE];  /* hash bucket */
 291         atomic_t                entries;        /* number of entries */
 292         int                     max_size;       /* maximum size of entries */
 293         struct timer_list       periodic_timer; /* collect stale entries */
 294         int                     rover;          /* rover for expire check */
 295         int                     counter;        /* counter for no expire */
 296 };
 297
 298
 299 /*
 300  *      IPVS LBLCR sysctl table
 301  */
 302
 303 static ctl_table vs_vars_table[] = {
 304         {
 305                 .procname       = "lblcr_expiration",
 306                 .data           = &sysctl_ip_vs_lblcr_expiration,
 307                 .maxlen         = sizeof(int),
 308                 .mode           = 0644,
 309                 .proc_handler   = &proc_dointvec_jiffies,
 310         },
 311         { .ctl_name = 0 }
 312 };
 313
 314 static ctl_table vs_table[] = {
 315         {
 316                 .procname       = "vs",
 317                 .mode           = 0555,
 318                 .child          = vs_vars_table
 319         },
 320         { .ctl_name = 0 }
 321 };
 322
 323 static ctl_table ipvs_ipv4_table[] = {
 324         {
 325                 .ctl_name       = NET_IPV4,
 326                 .procname       = "ipv4",
 327                 .mode           = 0555,
 328                 .child          = vs_table
 329         },
 330         { .ctl_name = 0 }
 331 };
 332
 333 static ctl_table lblcr_root_table[] = {
 334         {
 335                 .ctl_name       = CTL_NET,
 336                 .procname       = "net",
 337                 .mode           = 0555,
 338                 .child          = ipvs_ipv4_table
 339         },
 340         { .ctl_name = 0 }
 341 };
 342
 343 static struct ctl_table_header * sysctl_header;
 344
 345 /*
 346  *      new/free a ip_vs_lblcr_entry, which is a mapping of a destination
 347  *      IP address to a server.
 348  */
 349 static inline struct ip_vs_lblcr_entry *ip_vs_lblcr_new(__be32 daddr)
 350 {
 351         struct ip_vs_lblcr_entry *en;
 352
 353         en = kmalloc(sizeof(struct ip_vs_lblcr_entry), GFP_ATOMIC);
 354         if (en == NULL) {
 355                 IP_VS_ERR("ip_vs_lblcr_new(): no memory\n");
 356                 return NULL;
 357         }
 358
 359         INIT_LIST_HEAD(&en->list);
 360         en->addr = daddr;
 361
 362         /* initilize its dest set */
 363         atomic_set(&(en->set.size), 0);
 364         en->set.list = NULL;
 365         rwlock_init(&en->set.lock);
 366
 367         return en;
 368 }
 369
 370
 371 static inline void ip_vs_lblcr_free(struct ip_vs_lblcr_entry *en)
 372 {
 373         list_del(&en->list);
 374         ip_vs_dest_set_eraseall(&en->set);
 375         kfree(en);
 376 }
 377
 378
 379 /*
 380  *      Returns hash value for IPVS LBLCR entry
 381  */
 382 static inline unsigned ip_vs_lblcr_hashkey(__be32 addr)
 383 {
 384         return (ntohl(addr)*2654435761UL) & IP_VS_LBLCR_TAB_MASK;
 385 }
 386
 387
 388 /*
 389  *      Hash an entry in the ip_vs_lblcr_table.
 390  *      returns bool success.
 391  */
 392 static int
 393 ip_vs_lblcr_hash(struct ip_vs_lblcr_table *tbl, struct ip_vs_lblcr_entry *en)
 394 {
 395         unsigned hash;
 396
 397         if (!list_empty(&en->list)) {
 398                 IP_VS_ERR("ip_vs_lblcr_hash(): request for already hashed, "
 399                           "called from %p\n", __builtin_return_address(0));
 400                 return 0;
 401         }
 402
 403         /*
 404          *      Hash by destination IP address
 405          */
 406         hash = ip_vs_lblcr_hashkey(en->addr);
 407
 408         write_lock(&tbl->lock);
 409         list_add(&en->list, &tbl->bucket[hash]);
 410         atomic_inc(&tbl->entries);
 411         write_unlock(&tbl->lock);
 412
 413         return 1;
 414 }
 415
 416
 417 /*
 418  *  Get ip_vs_lblcr_entry associated with supplied parameters.
 419  */
 420 static inline struct ip_vs_lblcr_entry *
 421 ip_vs_lblcr_get(struct ip_vs_lblcr_table *tbl, __be32 addr)
 422 {
 423         unsigned hash;
 424         struct ip_vs_lblcr_entry *en;
 425
 426         hash = ip_vs_lblcr_hashkey(addr);
 427
 428         read_lock(&tbl->lock);
 429
 430         list_for_each_entry(en, &tbl->bucket[hash], list) {
 431                 if (en->addr == addr) {
 432                         /* HIT */
 433                         read_unlock(&tbl->lock);
 434                         return en;
 435                 }
 436         }
 437
 438         read_unlock(&tbl->lock);
 439
 440         return NULL;
 441 }
 442
 443
 444 /*
 445  *      Flush all the entries of the specified table.
 446  */
 447 static void ip_vs_lblcr_flush(struct ip_vs_lblcr_table *tbl)
 448 {
 449         int i;
 450         struct ip_vs_lblcr_entry *en, *nxt;
 451
 452         for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) {
 453                 write_lock(&tbl->lock);
 454                 list_for_each_entry_safe(en, nxt, &tbl->bucket[i], list) {
 455                         ip_vs_lblcr_free(en);
 456                         atomic_dec(&tbl->entries);
 457                 }
 458                 write_unlock(&tbl->lock);
 459         }
 460 }
 461
 462
 463 static inline void ip_vs_lblcr_full_check(struct ip_vs_lblcr_table *tbl)
 464 {
 465         unsigned long now = jiffies;
 466         int i, j;
 467         struct ip_vs_lblcr_entry *en, *nxt;
 468
 469         for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) {
 470                 j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
 471
 472                 write_lock(&tbl->lock);
 473                 list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) {
 474                         if (time_after(en->lastuse+sysctl_ip_vs_lblcr_expiration,
 475                                        now))
 476                                 continue;
 477
 478                         ip_vs_lblcr_free(en);
 479                         atomic_dec(&tbl->entries);
 480                 }
 481                 write_unlock(&tbl->lock);
 482         }
 483         tbl->rover = j;
 484 }
 485
 486
 487 /*
 488  *      Periodical timer handler for IPVS lblcr table
 489  *      It is used to collect stale entries when the number of entries
 490  *      exceeds the maximum size of the table.
 491  *
 492  *      Fixme: we probably need more complicated algorithm to collect
 493  *             entries that have not been used for a long time even
 494  *             if the number of entries doesn't exceed the maximum size
 495  *             of the table.
 496  *      The full expiration check is for this purpose now.
 497  */
 498 static void ip_vs_lblcr_check_expire(unsigned long data)
 499 {
 500         struct ip_vs_lblcr_table *tbl;
 501         unsigned long now = jiffies;
 502         int goal;
 503         int i, j;
 504         struct ip_vs_lblcr_entry *en, *nxt;
 505
 506         tbl = (struct ip_vs_lblcr_table *)data;
 507
 508         if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) {
 509                 /* do full expiration check */
 510                 ip_vs_lblcr_full_check(tbl);
 511                 tbl->counter = 1;
 512                 goto out;
 513         }
 514
 515         if (atomic_read(&tbl->entries) <= tbl->max_size) {
 516                 tbl->counter++;
 517                 goto out;
 518         }
 519
 520         goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3;
 521         if (goal > tbl->max_size/2)
 522                 goal = tbl->max_size/2;
 523
 524         for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) {
 525                 j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
 526
 527                 write_lock(&tbl->lock);
 528                 list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) {
 529                         if (time_before(now, en->lastuse+ENTRY_TIMEOUT))
 530                                 continue;
 531
 532                         ip_vs_lblcr_free(en);
 533                         atomic_dec(&tbl->entries);
 534                         goal--;
 535                 }
 536                 write_unlock(&tbl->lock);
 537                 if (goal <= 0)
 538                         break;
 539         }
 540         tbl->rover = j;
 541
 542   out:
 543         mod_timer(&tbl->periodic_timer, jiffies+CHECK_EXPIRE_INTERVAL);
 544 }
 545
 546 static int ip_vs_lblcr_init_svc(struct ip_vs_service *svc)
 547 {
 548         int i;
 549         struct ip_vs_lblcr_table *tbl;
 550
 551         /*
 552          *    Allocate the ip_vs_lblcr_table for this service
 553          */
 554         tbl = kmalloc(sizeof(struct ip_vs_lblcr_table), GFP_ATOMIC);
 555         if (tbl == NULL) {
 556                 IP_VS_ERR("ip_vs_lblcr_init_svc(): no memory\n");
 557                 return -ENOMEM;
 558         }
 559         svc->sched_data = tbl;
 560         IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) allocated for "
 561                   "current service\n",
 562                   sizeof(struct ip_vs_lblcr_table));
 563
 564         /*
 565          *    Initialize the hash buckets
 566          */
 567         for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) {
 568                 INIT_LIST_HEAD(&tbl->bucket[i]);
 569         }
 570         rwlock_init(&tbl->lock);
 571         tbl->max_size = IP_VS_LBLCR_TAB_SIZE*16;
 572         tbl->rover = 0;
 573         tbl->counter = 1;
 574
 575         /*
 576          *    Hook periodic timer for garbage collection
 577          */
 578         init_timer(&tbl->periodic_timer);
 579         tbl->periodic_timer.data = (unsigned long)tbl;
 580         tbl->periodic_timer.function = ip_vs_lblcr_check_expire;
 581         tbl->periodic_timer.expires = jiffies+CHECK_EXPIRE_INTERVAL;
 582         add_timer(&tbl->periodic_timer);
 583
 584         return 0;
 585 }
 586
 587
 588 static int ip_vs_lblcr_done_svc(struct ip_vs_service *svc)
 589 {
 590         struct ip_vs_lblcr_table *tbl = svc->sched_data;
 591
 592         /* remove periodic timer */
 593         del_timer_sync(&tbl->periodic_timer);
 594
 595         /* got to clean up table entries here */
 596         ip_vs_lblcr_flush(tbl);
 597
 598         /* release the table itself */
 599         kfree(svc->sched_data);
 600         IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) released\n",
 601                   sizeof(struct ip_vs_lblcr_table));
 602
 603         return 0;
 604 }
 605
 606
 607 static int ip_vs_lblcr_update_svc(struct ip_vs_service *svc)
 608 {
 609         return 0;
 610 }
 611
 612
 613 static inline struct ip_vs_dest *
 614 __ip_vs_wlc_schedule(struct ip_vs_service *svc, struct iphdr *iph)
 615 {
 616         struct ip_vs_dest *dest, *least;
 617         int loh, doh;
 618
 619         /*
 620          * We think the overhead of processing active connections is fifty
 621          * times higher than that of inactive connections in average. (This
 622          * fifty times might not be accurate, we will change it later.) We
 623          * use the following formula to estimate the overhead:
 624          *                dest->activeconns*50 + dest->inactconns
 625          * and the load:
 626          *                (dest overhead) / dest->weight
 627          *
 628          * Remember -- no floats in kernel mode!!!
 629          * The comparison of h1*w2 > h2*w1 is equivalent to that of
 630          *                h1/w1 > h2/w2
 631          * if every weight is larger than zero.
 632          *
 633          * The server with weight=0 is quiesced and will not receive any
 634          * new connection.
 635          */
 636         list_for_each_entry(dest, &svc->destinations, n_list) {
 637                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 638                         continue;
 639
 640                 if (atomic_read(&dest->weight) > 0) {
 641                         least = dest;
 642                         loh = atomic_read(&least->activeconns) * 50
 643                                 + atomic_read(&least->inactconns);
 644                         goto nextstage;
 645                 }
 646         }
 647         return NULL;
 648
 649         /*
 650          *    Find the destination with the least load.
 651          */
 652   nextstage:
 653         list_for_each_entry_continue(dest, &svc->destinations, n_list) {
 654                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 655                         continue;
 656
 657                 doh = atomic_read(&dest->activeconns) * 50
 658                         + atomic_read(&dest->inactconns);
 659                 if (loh * atomic_read(&dest->weight) >
 660                     doh * atomic_read(&least->weight)) {
 661                         least = dest;
 662                         loh = doh;
 663                 }
 664         }
 665
 666         IP_VS_DBG(6, "LBLCR: server %d.%d.%d.%d:%d "
 667                   "activeconns %d refcnt %d weight %d overhead %d\n",
 668                   NIPQUAD(least->addr), ntohs(least->port),
 669                   atomic_read(&least->activeconns),
 670                   atomic_read(&least->refcnt),
 671                   atomic_read(&least->weight), loh);
 672
 673         return least;
 674 }
 675
 676
 677 /*
 678  *   If this destination server is overloaded and there is a less loaded
 679  *   server, then return true.
 680  */
 681 static inline int
 682 is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc)
 683 {
 684         if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) {
 685                 struct ip_vs_dest *d;
 686
 687                 list_for_each_entry(d, &svc->destinations, n_list) {
 688                         if (atomic_read(&d->activeconns)*2
 689                             < atomic_read(&d->weight)) {
 690                                 return 1;
 691                         }
 692                 }
 693         }
 694         return 0;
 695 }
 696
 697
 698 /*
 699  *    Locality-Based (weighted) Least-Connection scheduling
 700  */
 701 static struct ip_vs_dest *
 702 ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
 703 {
 704         struct ip_vs_dest *dest;
 705         struct ip_vs_lblcr_table *tbl;
 706         struct ip_vs_lblcr_entry *en;
 707         struct iphdr *iph = ip_hdr(skb);
 708
 709         IP_VS_DBG(6, "ip_vs_lblcr_schedule(): Scheduling...\n");
 710
 711         tbl = (struct ip_vs_lblcr_table *)svc->sched_data;
 712         en = ip_vs_lblcr_get(tbl, iph->daddr);
 713         if (en == NULL) {
 714                 dest = __ip_vs_wlc_schedule(svc, iph);
 715                 if (dest == NULL) {
 716                         IP_VS_DBG(1, "no destination available\n");
 717                         return NULL;
 718                 }
 719                 en = ip_vs_lblcr_new(iph->daddr);
 720                 if (en == NULL) {
 721                         return NULL;
 722                 }
 723                 ip_vs_dest_set_insert(&en->set, dest);
 724                 ip_vs_lblcr_hash(tbl, en);
 725         } else {
 726                 dest = ip_vs_dest_set_min(&en->set);
 727                 if (!dest || is_overloaded(dest, svc)) {
 728                         dest = __ip_vs_wlc_schedule(svc, iph);
 729                         if (dest == NULL) {
 730                                 IP_VS_DBG(1, "no destination available\n");
 731                                 return NULL;
 732                         }
 733                         ip_vs_dest_set_insert(&en->set, dest);
 734                 }
 735                 if (atomic_read(&en->set.size) > 1 &&
 736                     jiffies-en->set.lastmod > sysctl_ip_vs_lblcr_expiration) {
 737                         struct ip_vs_dest *m;
 738                         m = ip_vs_dest_set_max(&en->set);
 739                         if (m)
 740                                 ip_vs_dest_set_erase(&en->set, m);
 741                 }
 742         }
 743         en->lastuse = jiffies;
 744
 745         IP_VS_DBG(6, "LBLCR: destination IP address %u.%u.%u.%u "
 746                   "--> server %u.%u.%u.%u:%d\n",
 747                   NIPQUAD(en->addr),
 748                   NIPQUAD(dest->addr),
 749                   ntohs(dest->port));
 750
 751         return dest;
 752 }
 753
 754
 755 /*
 756  *      IPVS LBLCR Scheduler structure
 757  */
 758 static struct ip_vs_scheduler ip_vs_lblcr_scheduler =
 759 {
 760         .name =                 "lblcr",
 761         .refcnt =               ATOMIC_INIT(0),
 762         .module =               THIS_MODULE,
 763         .init_service =         ip_vs_lblcr_init_svc,
 764         .done_service =         ip_vs_lblcr_done_svc,
 765         .update_service =       ip_vs_lblcr_update_svc,
 766         .schedule =             ip_vs_lblcr_schedule,
 767 };
 768
 769
 770 static int __init ip_vs_lblcr_init(void)
 771 {
 772         INIT_LIST_HEAD(&ip_vs_lblcr_scheduler.n_list);
 773         sysctl_header = register_sysctl_table(lblcr_root_table);
 774         return register_ip_vs_scheduler(&ip_vs_lblcr_scheduler);
 775 }
 776
 777
 778 static void __exit ip_vs_lblcr_cleanup(void)
 779 {
 780         unregister_sysctl_table(sysctl_header);
 781         unregister_ip_vs_scheduler(&ip_vs_lblcr_scheduler);
 782 }
 783
 784
 785 module_init(ip_vs_lblcr_init);
 786 module_exit(ip_vs_lblcr_cleanup);
 787 MODULE_LICENSE("GPL");