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

   1 /*
   2  * IPVS:        Locality-Based Least-Connection with Replication scheduler
   3  *
   4  * Authors:     Wensong Zhang <wensong@gnuchina.org>
   5  *
   6  *              This program is free software; you can redistribute it and/or
   7  *              modify it under the terms of the GNU General Public License
   8  *              as published by the Free Software Foundation; either version
   9  *              2 of the License, or (at your option) any later version.
  10  *
  11  * Changes:
  12  *     Julian Anastasov        :    Added the missing (dest->weight>0)
  13  *                                  condition in the ip_vs_dest_set_max.
  14  *
  15  */
  16
  17 /*
  18  * The lblc/r algorithm is as follows (pseudo code):
  19  *
  20  *       if serverSet[dest_ip] is null then
  21  *               n, serverSet[dest_ip] <- {weighted least-conn node};
  22  *       else
  23  *               n <- {least-conn (alive) node in serverSet[dest_ip]};
  24  *               if (n is null) OR
  25  *                  (n.conns>n.weight AND
  26  *                   there is a node m with m.conns<m.weight/2) then
  27  *                   n <- {weighted least-conn node};
  28  *                   add n to serverSet[dest_ip];
  29  *               if |serverSet[dest_ip]| > 1 AND
  30  *                   now - serverSet[dest_ip].lastMod > T then
  31  *                   m <- {most conn node in serverSet[dest_ip]};
  32  *                   remove m from serverSet[dest_ip];
  33  *       if serverSet[dest_ip] changed then
  34  *               serverSet[dest_ip].lastMod <- now;
  35  *
  36  *       return n;
  37  *
  38  */
  39
  40 #define KMSG_COMPONENT "IPVS"
  41 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  42
  43 #include <linux/ip.h>
  44 #include <linux/module.h>
  45 #include <linux/kernel.h>
  46 #include <linux/skbuff.h>
  47 #include <linux/jiffies.h>
  48
  49 /* for sysctl */
  50 #include <linux/fs.h>
  51 #include <linux/sysctl.h>
  52 #include <net/net_namespace.h>
  53
  54 #include <net/ip_vs.h>
  55
  56
  57 /*
  58  *    It is for garbage collection of stale IPVS lblcr entries,
  59  *    when the table is full.
  60  */
  61 #define CHECK_EXPIRE_INTERVAL   (60*HZ)
  62 #define ENTRY_TIMEOUT           (6*60*HZ)
  63
  64 /*
  65  *    It is for full expiration check.
  66  *    When there is no partial expiration check (garbage collection)
  67  *    in a half hour, do a full expiration check to collect stale
  68  *    entries that haven't been touched for a day.
  69  */
  70 #define COUNT_FOR_FULL_EXPIRATION   30
  71 static int sysctl_ip_vs_lblcr_expiration = 24*60*60*HZ;
  72
  73
  74 /*
  75  *     for IPVS lblcr entry hash table
  76  */
  77 #ifndef CONFIG_IP_VS_LBLCR_TAB_BITS
  78 #define CONFIG_IP_VS_LBLCR_TAB_BITS      10
  79 #endif
  80 #define IP_VS_LBLCR_TAB_BITS     CONFIG_IP_VS_LBLCR_TAB_BITS
  81 #define IP_VS_LBLCR_TAB_SIZE     (1 << IP_VS_LBLCR_TAB_BITS)
  82 #define IP_VS_LBLCR_TAB_MASK     (IP_VS_LBLCR_TAB_SIZE - 1)
  83
  84
  85 /*
  86  *      IPVS destination set structure and operations
  87  */
  88 struct ip_vs_dest_list {
  89         struct ip_vs_dest_list  *next;          /* list link */
  90         struct ip_vs_dest       *dest;          /* destination server */
  91 };
  92
  93 struct ip_vs_dest_set {
  94         atomic_t                size;           /* set size */
  95         unsigned long           lastmod;        /* last modified time */
  96         struct ip_vs_dest_list  *list;          /* destination list */
  97         rwlock_t                lock;           /* lock for this list */
  98 };
  99
 100
 101 static struct ip_vs_dest_list *
 102 ip_vs_dest_set_insert(struct ip_vs_dest_set *set, struct ip_vs_dest *dest)
 103 {
 104         struct ip_vs_dest_list *e;
 105
 106         for (e=set->list; e!=NULL; e=e->next) {
 107                 if (e->dest == dest)
 108                         /* already existed */
 109                         return NULL;
 110         }
 111
 112         e = kmalloc(sizeof(*e), GFP_ATOMIC);
 113         if (e == NULL) {
 114                 IP_VS_ERR("ip_vs_dest_set_insert(): no memory\n");
 115                 return NULL;
 116         }
 117
 118         atomic_inc(&dest->refcnt);
 119         e->dest = dest;
 120
 121         /* link it to the list */
 122         e->next = set->list;
 123         set->list = e;
 124         atomic_inc(&set->size);
 125
 126         set->lastmod = jiffies;
 127         return e;
 128 }
 129
 130 static void
 131 ip_vs_dest_set_erase(struct ip_vs_dest_set *set, struct ip_vs_dest *dest)
 132 {
 133         struct ip_vs_dest_list *e, **ep;
 134
 135         for (ep=&set->list, e=*ep; e!=NULL; e=*ep) {
 136                 if (e->dest == dest) {
 137                         /* HIT */
 138                         *ep = e->next;
 139                         atomic_dec(&set->size);
 140                         set->lastmod = jiffies;
 141                         atomic_dec(&e->dest->refcnt);
 142                         kfree(e);
 143                         break;
 144                 }
 145                 ep = &e->next;
 146         }
 147 }
 148
 149 static void ip_vs_dest_set_eraseall(struct ip_vs_dest_set *set)
 150 {
 151         struct ip_vs_dest_list *e, **ep;
 152
 153         write_lock(&set->lock);
 154         for (ep=&set->list, e=*ep; e!=NULL; e=*ep) {
 155                 *ep = e->next;
 156                 /*
 157                  * We don't kfree dest because it is refered either
 158                  * by its service or by the trash dest list.
 159                  */
 160                 atomic_dec(&e->dest->refcnt);
 161                 kfree(e);
 162         }
 163         write_unlock(&set->lock);
 164 }
 165
 166 /* get weighted least-connection node in the destination set */
 167 static inline struct ip_vs_dest *ip_vs_dest_set_min(struct ip_vs_dest_set *set)
 168 {
 169         register struct ip_vs_dest_list *e;
 170         struct ip_vs_dest *dest, *least;
 171         int loh, doh;
 172
 173         if (set == NULL)
 174                 return NULL;
 175
 176         /* select the first destination server, whose weight > 0 */
 177         for (e=set->list; e!=NULL; e=e->next) {
 178                 least = e->dest;
 179                 if (least->flags & IP_VS_DEST_F_OVERLOAD)
 180                         continue;
 181
 182                 if ((atomic_read(&least->weight) > 0)
 183                     && (least->flags & IP_VS_DEST_F_AVAILABLE)) {
 184                         loh = atomic_read(&least->activeconns) * 50
 185                                 + atomic_read(&least->inactconns);
 186                         goto nextstage;
 187                 }
 188         }
 189         return NULL;
 190
 191         /* find the destination with the weighted least load */
 192   nextstage:
 193         for (e=e->next; e!=NULL; e=e->next) {
 194                 dest = e->dest;
 195                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 196                         continue;
 197
 198                 doh = atomic_read(&dest->activeconns) * 50
 199                         + atomic_read(&dest->inactconns);
 200                 if ((loh * atomic_read(&dest->weight) >
 201                      doh * atomic_read(&least->weight))
 202                     && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 203                         least = dest;
 204                         loh = doh;
 205                 }
 206         }
 207
 208         IP_VS_DBG_BUF(6, "ip_vs_dest_set_min: server %s:%d "
 209                       "activeconns %d refcnt %d weight %d overhead %d\n",
 210                       IP_VS_DBG_ADDR(least->af, &least->addr),
 211                       ntohs(least->port),
 212                       atomic_read(&least->activeconns),
 213                       atomic_read(&least->refcnt),
 214                       atomic_read(&least->weight), loh);
 215         return least;
 216 }
 217
 218
 219 /* get weighted most-connection node in the destination set */
 220 static inline struct ip_vs_dest *ip_vs_dest_set_max(struct ip_vs_dest_set *set)
 221 {
 222         register struct ip_vs_dest_list *e;
 223         struct ip_vs_dest *dest, *most;
 224         int moh, doh;
 225
 226         if (set == NULL)
 227                 return NULL;
 228
 229         /* select the first destination server, whose weight > 0 */
 230         for (e=set->list; e!=NULL; e=e->next) {
 231                 most = e->dest;
 232                 if (atomic_read(&most->weight) > 0) {
 233                         moh = atomic_read(&most->activeconns) * 50
 234                                 + atomic_read(&most->inactconns);
 235                         goto nextstage;
 236                 }
 237         }
 238         return NULL;
 239
 240         /* find the destination with the weighted most load */
 241   nextstage:
 242         for (e=e->next; e!=NULL; e=e->next) {
 243                 dest = e->dest;
 244                 doh = atomic_read(&dest->activeconns) * 50
 245                         + atomic_read(&dest->inactconns);
 246                 /* moh/mw < doh/dw ==> moh*dw < doh*mw, where mw,dw>0 */
 247                 if ((moh * atomic_read(&dest->weight) <
 248                      doh * atomic_read(&most->weight))
 249                     && (atomic_read(&dest->weight) > 0)) {
 250                         most = dest;
 251                         moh = doh;
 252                 }
 253         }
 254
 255         IP_VS_DBG_BUF(6, "ip_vs_dest_set_max: server %s:%d "
 256                       "activeconns %d refcnt %d weight %d overhead %d\n",
 257                       IP_VS_DBG_ADDR(most->af, &most->addr), ntohs(most->port),
 258                       atomic_read(&most->activeconns),
 259                       atomic_read(&most->refcnt),
 260                       atomic_read(&most->weight), moh);
 261         return most;
 262 }
 263
 264
 265 /*
 266  *      IPVS lblcr entry represents an association between destination
 267  *      IP address and its destination server set
 268  */
 269 struct ip_vs_lblcr_entry {
 270         struct list_head        list;
 271         int                     af;             /* address family */
 272         union nf_inet_addr      addr;           /* destination IP address */
 273         struct ip_vs_dest_set   set;            /* destination server set */
 274         unsigned long           lastuse;        /* last used time */
 275 };
 276
 277
 278 /*
 279  *      IPVS lblcr hash table
 280  */
 281 struct ip_vs_lblcr_table {
 282         struct list_head        bucket[IP_VS_LBLCR_TAB_SIZE];  /* hash bucket */
 283         atomic_t                entries;        /* number of entries */
 284         int                     max_size;       /* maximum size of entries */
 285         struct timer_list       periodic_timer; /* collect stale entries */
 286         int                     rover;          /* rover for expire check */
 287         int                     counter;        /* counter for no expire */
 288 };
 289
 290
 291 /*
 292  *      IPVS LBLCR sysctl table
 293  */
 294
 295 static ctl_table vs_vars_table[] = {
 296         {
 297                 .procname       = "lblcr_expiration",
 298                 .data           = &sysctl_ip_vs_lblcr_expiration,
 299                 .maxlen         = sizeof(int),
 300                 .mode           = 0644,
 301                 .proc_handler   = proc_dointvec_jiffies,
 302         },
 303         { .ctl_name = 0 }
 304 };
 305
 306 static struct ctl_table_header * sysctl_header;
 307
 308 static inline void ip_vs_lblcr_free(struct ip_vs_lblcr_entry *en)
 309 {
 310         list_del(&en->list);
 311         ip_vs_dest_set_eraseall(&en->set);
 312         kfree(en);
 313 }
 314
 315
 316 /*
 317  *      Returns hash value for IPVS LBLCR entry
 318  */
 319 static inline unsigned
 320 ip_vs_lblcr_hashkey(int af, const union nf_inet_addr *addr)
 321 {
 322         __be32 addr_fold = addr->ip;
 323
 324 #ifdef CONFIG_IP_VS_IPV6
 325         if (af == AF_INET6)
 326                 addr_fold = addr->ip6[0]^addr->ip6[1]^
 327                             addr->ip6[2]^addr->ip6[3];
 328 #endif
 329         return (ntohl(addr_fold)*2654435761UL) & IP_VS_LBLCR_TAB_MASK;
 330 }
 331
 332
 333 /*
 334  *      Hash an entry in the ip_vs_lblcr_table.
 335  *      returns bool success.
 336  */
 337 static void
 338 ip_vs_lblcr_hash(struct ip_vs_lblcr_table *tbl, struct ip_vs_lblcr_entry *en)
 339 {
 340         unsigned hash = ip_vs_lblcr_hashkey(en->af, &en->addr);
 341
 342         list_add(&en->list, &tbl->bucket[hash]);
 343         atomic_inc(&tbl->entries);
 344 }
 345
 346
 347 /*
 348  *  Get ip_vs_lblcr_entry associated with supplied parameters. Called under
 349  *  read lock.
 350  */
 351 static inline struct ip_vs_lblcr_entry *
 352 ip_vs_lblcr_get(int af, struct ip_vs_lblcr_table *tbl,
 353                 const union nf_inet_addr *addr)
 354 {
 355         unsigned hash = ip_vs_lblcr_hashkey(af, addr);
 356         struct ip_vs_lblcr_entry *en;
 357
 358         list_for_each_entry(en, &tbl->bucket[hash], list)
 359                 if (ip_vs_addr_equal(af, &en->addr, addr))
 360                         return en;
 361
 362         return NULL;
 363 }
 364
 365
 366 /*
 367  * Create or update an ip_vs_lblcr_entry, which is a mapping of a destination
 368  * IP address to a server. Called under write lock.
 369  */
 370 static inline struct ip_vs_lblcr_entry *
 371 ip_vs_lblcr_new(struct ip_vs_lblcr_table *tbl, const union nf_inet_addr *daddr,
 372                 struct ip_vs_dest *dest)
 373 {
 374         struct ip_vs_lblcr_entry *en;
 375
 376         en = ip_vs_lblcr_get(dest->af, tbl, daddr);
 377         if (!en) {
 378                 en = kmalloc(sizeof(*en), GFP_ATOMIC);
 379                 if (!en) {
 380                         IP_VS_ERR("ip_vs_lblcr_new(): no memory\n");
 381                         return NULL;
 382                 }
 383
 384                 en->af = dest->af;
 385                 ip_vs_addr_copy(dest->af, &en->addr, daddr);
 386                 en->lastuse = jiffies;
 387
 388                 /* initilize its dest set */
 389                 atomic_set(&(en->set.size), 0);
 390                 en->set.list = NULL;
 391                 rwlock_init(&en->set.lock);
 392
 393                 ip_vs_lblcr_hash(tbl, en);
 394         }
 395
 396         write_lock(&en->set.lock);
 397         ip_vs_dest_set_insert(&en->set, dest);
 398         write_unlock(&en->set.lock);
 399
 400         return en;
 401 }
 402
 403
 404 /*
 405  *      Flush all the entries of the specified table.
 406  */
 407 static void ip_vs_lblcr_flush(struct ip_vs_lblcr_table *tbl)
 408 {
 409         int i;
 410         struct ip_vs_lblcr_entry *en, *nxt;
 411
 412         /* No locking required, only called during cleanup. */
 413         for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) {
 414                 list_for_each_entry_safe(en, nxt, &tbl->bucket[i], list) {
 415                         ip_vs_lblcr_free(en);
 416                 }
 417         }
 418 }
 419
 420
 421 static inline void ip_vs_lblcr_full_check(struct ip_vs_service *svc)
 422 {
 423         struct ip_vs_lblcr_table *tbl = svc->sched_data;
 424         unsigned long now = jiffies;
 425         int i, j;
 426         struct ip_vs_lblcr_entry *en, *nxt;
 427
 428         for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) {
 429                 j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
 430
 431                 write_lock(&svc->sched_lock);
 432                 list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) {
 433                         if (time_after(en->lastuse+sysctl_ip_vs_lblcr_expiration,
 434                                        now))
 435                                 continue;
 436
 437                         ip_vs_lblcr_free(en);
 438                         atomic_dec(&tbl->entries);
 439                 }
 440                 write_unlock(&svc->sched_lock);
 441         }
 442         tbl->rover = j;
 443 }
 444
 445
 446 /*
 447  *      Periodical timer handler for IPVS lblcr table
 448  *      It is used to collect stale entries when the number of entries
 449  *      exceeds the maximum size of the table.
 450  *
 451  *      Fixme: we probably need more complicated algorithm to collect
 452  *             entries that have not been used for a long time even
 453  *             if the number of entries doesn't exceed the maximum size
 454  *             of the table.
 455  *      The full expiration check is for this purpose now.
 456  */
 457 static void ip_vs_lblcr_check_expire(unsigned long data)
 458 {
 459         struct ip_vs_service *svc = (struct ip_vs_service *) data;
 460         struct ip_vs_lblcr_table *tbl = svc->sched_data;
 461         unsigned long now = jiffies;
 462         int goal;
 463         int i, j;
 464         struct ip_vs_lblcr_entry *en, *nxt;
 465
 466         if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) {
 467                 /* do full expiration check */
 468                 ip_vs_lblcr_full_check(svc);
 469                 tbl->counter = 1;
 470                 goto out;
 471         }
 472
 473         if (atomic_read(&tbl->entries) <= tbl->max_size) {
 474                 tbl->counter++;
 475                 goto out;
 476         }
 477
 478         goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3;
 479         if (goal > tbl->max_size/2)
 480                 goal = tbl->max_size/2;
 481
 482         for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) {
 483                 j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
 484
 485                 write_lock(&svc->sched_lock);
 486                 list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) {
 487                         if (time_before(now, en->lastuse+ENTRY_TIMEOUT))
 488                                 continue;
 489
 490                         ip_vs_lblcr_free(en);
 491                         atomic_dec(&tbl->entries);
 492                         goal--;
 493                 }
 494                 write_unlock(&svc->sched_lock);
 495                 if (goal <= 0)
 496                         break;
 497         }
 498         tbl->rover = j;
 499
 500   out:
 501         mod_timer(&tbl->periodic_timer, jiffies+CHECK_EXPIRE_INTERVAL);
 502 }
 503
 504 static int ip_vs_lblcr_init_svc(struct ip_vs_service *svc)
 505 {
 506         int i;
 507         struct ip_vs_lblcr_table *tbl;
 508
 509         /*
 510          *    Allocate the ip_vs_lblcr_table for this service
 511          */
 512         tbl = kmalloc(sizeof(*tbl), GFP_ATOMIC);
 513         if (tbl == NULL) {
 514                 IP_VS_ERR("ip_vs_lblcr_init_svc(): no memory\n");
 515                 return -ENOMEM;
 516         }
 517         svc->sched_data = tbl;
 518         IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) allocated for "
 519                   "current service\n", sizeof(*tbl));
 520
 521         /*
 522          *    Initialize the hash buckets
 523          */
 524         for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) {
 525                 INIT_LIST_HEAD(&tbl->bucket[i]);
 526         }
 527         tbl->max_size = IP_VS_LBLCR_TAB_SIZE*16;
 528         tbl->rover = 0;
 529         tbl->counter = 1;
 530
 531         /*
 532          *    Hook periodic timer for garbage collection
 533          */
 534         setup_timer(&tbl->periodic_timer, ip_vs_lblcr_check_expire,
 535                         (unsigned long)svc);
 536         mod_timer(&tbl->periodic_timer, jiffies + CHECK_EXPIRE_INTERVAL);
 537
 538         return 0;
 539 }
 540
 541
 542 static int ip_vs_lblcr_done_svc(struct ip_vs_service *svc)
 543 {
 544         struct ip_vs_lblcr_table *tbl = svc->sched_data;
 545
 546         /* remove periodic timer */
 547         del_timer_sync(&tbl->periodic_timer);
 548
 549         /* got to clean up table entries here */
 550         ip_vs_lblcr_flush(tbl);
 551
 552         /* release the table itself */
 553         kfree(tbl);
 554         IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) released\n",
 555                   sizeof(*tbl));
 556
 557         return 0;
 558 }
 559
 560
 561 static inline struct ip_vs_dest *
 562 __ip_vs_lblcr_schedule(struct ip_vs_service *svc)
 563 {
 564         struct ip_vs_dest *dest, *least;
 565         int loh, doh;
 566
 567         /*
 568          * We think the overhead of processing active connections is fifty
 569          * times higher than that of inactive connections in average. (This
 570          * fifty times might not be accurate, we will change it later.) We
 571          * use the following formula to estimate the overhead:
 572          *                dest->activeconns*50 + dest->inactconns
 573          * and the load:
 574          *                (dest overhead) / dest->weight
 575          *
 576          * Remember -- no floats in kernel mode!!!
 577          * The comparison of h1*w2 > h2*w1 is equivalent to that of
 578          *                h1/w1 > h2/w2
 579          * if every weight is larger than zero.
 580          *
 581          * The server with weight=0 is quiesced and will not receive any
 582          * new connection.
 583          */
 584         list_for_each_entry(dest, &svc->destinations, n_list) {
 585                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 586                         continue;
 587
 588                 if (atomic_read(&dest->weight) > 0) {
 589                         least = dest;
 590                         loh = atomic_read(&least->activeconns) * 50
 591                                 + atomic_read(&least->inactconns);
 592                         goto nextstage;
 593                 }
 594         }
 595         return NULL;
 596
 597         /*
 598          *    Find the destination with the least load.
 599          */
 600   nextstage:
 601         list_for_each_entry_continue(dest, &svc->destinations, n_list) {
 602                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 603                         continue;
 604
 605                 doh = atomic_read(&dest->activeconns) * 50
 606                         + atomic_read(&dest->inactconns);
 607                 if (loh * atomic_read(&dest->weight) >
 608                     doh * atomic_read(&least->weight)) {
 609                         least = dest;
 610                         loh = doh;
 611                 }
 612         }
 613
 614         IP_VS_DBG_BUF(6, "LBLCR: server %s:%d "
 615                       "activeconns %d refcnt %d weight %d overhead %d\n",
 616                       IP_VS_DBG_ADDR(least->af, &least->addr),
 617                       ntohs(least->port),
 618                       atomic_read(&least->activeconns),
 619                       atomic_read(&least->refcnt),
 620                       atomic_read(&least->weight), loh);
 621
 622         return least;
 623 }
 624
 625
 626 /*
 627  *   If this destination server is overloaded and there is a less loaded
 628  *   server, then return true.
 629  */
 630 static inline int
 631 is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc)
 632 {
 633         if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) {
 634                 struct ip_vs_dest *d;
 635
 636                 list_for_each_entry(d, &svc->destinations, n_list) {
 637                         if (atomic_read(&d->activeconns)*2
 638                             < atomic_read(&d->weight)) {
 639                                 return 1;
 640                         }
 641                 }
 642         }
 643         return 0;
 644 }
 645
 646
 647 /*
 648  *    Locality-Based (weighted) Least-Connection scheduling
 649  */
 650 static struct ip_vs_dest *
 651 ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
 652 {
 653         struct ip_vs_lblcr_table *tbl = svc->sched_data;
 654         struct ip_vs_iphdr iph;
 655         struct ip_vs_dest *dest = NULL;
 656         struct ip_vs_lblcr_entry *en;
 657
 658         ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
 659
 660         IP_VS_DBG(6, "ip_vs_lblcr_schedule(): Scheduling...\n");
 661
 662         /* First look in our cache */
 663         read_lock(&svc->sched_lock);
 664         en = ip_vs_lblcr_get(svc->af, tbl, &iph.daddr);
 665         if (en) {
 666                 /* We only hold a read lock, but this is atomic */
 667                 en->lastuse = jiffies;
 668
 669                 /* Get the least loaded destination */
 670                 read_lock(&en->set.lock);
 671                 dest = ip_vs_dest_set_min(&en->set);
 672                 read_unlock(&en->set.lock);
 673
 674                 /* More than one destination + enough time passed by, cleanup */
 675                 if (atomic_read(&en->set.size) > 1 &&
 676                                 time_after(jiffies, en->set.lastmod +
 677                                 sysctl_ip_vs_lblcr_expiration)) {
 678                         struct ip_vs_dest *m;
 679
 680                         write_lock(&en->set.lock);
 681                         m = ip_vs_dest_set_max(&en->set);
 682                         if (m)
 683                                 ip_vs_dest_set_erase(&en->set, m);
 684                         write_unlock(&en->set.lock);
 685                 }
 686
 687                 /* If the destination is not overloaded, use it */
 688                 if (dest && !is_overloaded(dest, svc)) {
 689                         read_unlock(&svc->sched_lock);
 690                         goto out;
 691                 }
 692
 693                 /* The cache entry is invalid, time to schedule */
 694                 dest = __ip_vs_lblcr_schedule(svc);
 695                 if (!dest) {
 696                         IP_VS_ERR_RL("LBLCR: no destination available\n");
 697                         read_unlock(&svc->sched_lock);
 698                         return NULL;
 699                 }
 700
 701                 /* Update our cache entry */
 702                 write_lock(&en->set.lock);
 703                 ip_vs_dest_set_insert(&en->set, dest);
 704                 write_unlock(&en->set.lock);
 705         }
 706         read_unlock(&svc->sched_lock);
 707
 708         if (dest)
 709                 goto out;
 710
 711         /* No cache entry, time to schedule */
 712         dest = __ip_vs_lblcr_schedule(svc);
 713         if (!dest) {
 714                 IP_VS_DBG(1, "no destination available\n");
 715                 return NULL;
 716         }
 717
 718         /* If we fail to create a cache entry, we'll just use the valid dest */
 719         write_lock(&svc->sched_lock);
 720         ip_vs_lblcr_new(tbl, &iph.daddr, dest);
 721         write_unlock(&svc->sched_lock);
 722
 723 out:
 724         IP_VS_DBG_BUF(6, "LBLCR: destination IP address %s --> server %s:%d\n",
 725                       IP_VS_DBG_ADDR(svc->af, &iph.daddr),
 726                       IP_VS_DBG_ADDR(svc->af, &dest->addr), ntohs(dest->port));
 727
 728         return dest;
 729 }
 730
 731
 732 /*
 733  *      IPVS LBLCR Scheduler structure
 734  */
 735 static struct ip_vs_scheduler ip_vs_lblcr_scheduler =
 736 {
 737         .name =                 "lblcr",
 738         .refcnt =               ATOMIC_INIT(0),
 739         .module =               THIS_MODULE,
 740         .n_list =               LIST_HEAD_INIT(ip_vs_lblcr_scheduler.n_list),
 741         .init_service =         ip_vs_lblcr_init_svc,
 742         .done_service =         ip_vs_lblcr_done_svc,
 743         .schedule =             ip_vs_lblcr_schedule,
 744 };
 745
 746
 747 static int __init ip_vs_lblcr_init(void)
 748 {
 749         int ret;
 750
 751         sysctl_header = register_sysctl_paths(net_vs_ctl_path, vs_vars_table);
 752         ret = register_ip_vs_scheduler(&ip_vs_lblcr_scheduler);
 753         if (ret)
 754                 unregister_sysctl_table(sysctl_header);
 755         return ret;
 756 }
 757
 758
 759 static void __exit ip_vs_lblcr_cleanup(void)
 760 {
 761         unregister_sysctl_table(sysctl_header);
 762         unregister_ip_vs_scheduler(&ip_vs_lblcr_scheduler);
 763 }
 764
 765
 766 module_init(ip_vs_lblcr_init);
 767 module_exit(ip_vs_lblcr_cleanup);
 768 MODULE_LICENSE("GPL");