net/ipv4/ipvs/ip_vs_lblc.c

   1 /*
   2  * IPVS:        Locality-Based Least-Connection scheduling module
   3  *
   4  * Version:     $Id: ip_vs_lblc.c,v 1.10 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  *     Martin Hamilton         :    fixed the terrible locking bugs
  15  *                                   *lock(tbl->lock) ==> *lock(&tbl->lock)
  16  *     Wensong Zhang           :    fixed the uninitilized tbl->lock bug
  17  *     Wensong Zhang           :    added doing full expiration check to
  18  *                                   collect stale entries of 24+ hours when
  19  *                                   no partial expire check in a half hour
  20  *     Julian Anastasov        :    replaced del_timer call with del_timer_sync
  21  *                                   to avoid the possible race between timer
  22  *                                   handler and del_timer thread in SMP
  23  *
  24  */
  25
  26 /*
  27  * The lblc algorithm is as follows (pseudo code):
  28  *
  29  *       if cachenode[dest_ip] is null then
  30  *               n, cachenode[dest_ip] <- {weighted least-conn node};
  31  *       else
  32  *               n <- cachenode[dest_ip];
  33  *               if (n is dead) OR
  34  *                  (n.conns>n.weight AND
  35  *                   there is a node m with m.conns<m.weight/2) then
  36  *                 n, cachenode[dest_ip] <- {weighted least-conn node};
  37  *
  38  *       return n;
  39  *
  40  * Thanks must go to Wenzhuo Zhang for talking WCCP to me and pushing
  41  * me to write this module.
  42  */
  43
  44 #include <linux/ip.h>
  45 #include <linux/module.h>
  46 #include <linux/kernel.h>
  47 #include <linux/skbuff.h>
  48
  49 /* for sysctl */
  50 #include <linux/fs.h>
  51 #include <linux/sysctl.h>
  52
  53 #include <net/ip_vs.h>
  54
  55
  56 /*
  57  *    It is for garbage collection of stale IPVS lblc 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_lblc_expiration = 24*60*60*HZ;
  71
  72
  73 /*
  74  *     for IPVS lblc entry hash table
  75  */
  76 #ifndef CONFIG_IP_VS_LBLC_TAB_BITS
  77 #define CONFIG_IP_VS_LBLC_TAB_BITS      10
  78 #endif
  79 #define IP_VS_LBLC_TAB_BITS     CONFIG_IP_VS_LBLC_TAB_BITS
  80 #define IP_VS_LBLC_TAB_SIZE     (1 << IP_VS_LBLC_TAB_BITS)
  81 #define IP_VS_LBLC_TAB_MASK     (IP_VS_LBLC_TAB_SIZE - 1)
  82
  83
  84 /*
  85  *      IPVS lblc entry represents an association between destination
  86  *      IP address and its destination server
  87  */
  88 struct ip_vs_lblc_entry {
  89         struct list_head        list;
  90         __u32                   addr;           /* destination IP address */
  91         struct ip_vs_dest       *dest;          /* real server (cache) */
  92         unsigned long           lastuse;        /* last used time */
  93 };
  94
  95
  96 /*
  97  *      IPVS lblc hash table
  98  */
  99 struct ip_vs_lblc_table {
 100         rwlock_t                lock;           /* lock for this table */
 101         struct list_head        bucket[IP_VS_LBLC_TAB_SIZE];  /* hash bucket */
 102         atomic_t                entries;        /* number of entries */
 103         int                     max_size;       /* maximum size of entries */
 104         struct timer_list       periodic_timer; /* collect stale entries */
 105         int                     rover;          /* rover for expire check */
 106         int                     counter;        /* counter for no expire */
 107 };
 108
 109
 110 /*
 111  *      IPVS LBLC sysctl table
 112  */
 113
 114 static ctl_table vs_vars_table[] = {
 115         {
 116                 .ctl_name       = NET_IPV4_VS_LBLC_EXPIRE,
 117                 .procname       = "lblc_expiration",
 118                 .data           = &sysctl_ip_vs_lblc_expiration,
 119                 .maxlen         = sizeof(int),
 120                 .mode           = 0644,
 121                 .proc_handler   = &proc_dointvec_jiffies,
 122         },
 123         { .ctl_name = 0 }
 124 };
 125
 126 static ctl_table vs_table[] = {
 127         {
 128                 .ctl_name       = NET_IPV4_VS,
 129                 .procname       = "vs",
 130                 .mode           = 0555,
 131                 .child          = vs_vars_table
 132         },
 133         { .ctl_name = 0 }
 134 };
 135
 136 static ctl_table ipvs_ipv4_table[] = {
 137         {
 138                 .ctl_name       = NET_IPV4,
 139                 .procname       = "ipv4",
 140                 .mode           = 0555,
 141                 .child          = vs_table
 142         },
 143         { .ctl_name = 0 }
 144 };
 145
 146 static ctl_table lblc_root_table[] = {
 147         {
 148                 .ctl_name       = CTL_NET,
 149                 .procname       = "net",
 150                 .mode           = 0555,
 151                 .child          = ipvs_ipv4_table
 152         },
 153         { .ctl_name = 0 }
 154 };
 155
 156 static struct ctl_table_header * sysctl_header;
 157
 158 /*
 159  *      new/free a ip_vs_lblc_entry, which is a mapping of a destionation
 160  *      IP address to a server.
 161  */
 162 static inline struct ip_vs_lblc_entry *
 163 ip_vs_lblc_new(__u32 daddr, struct ip_vs_dest *dest)
 164 {
 165         struct ip_vs_lblc_entry *en;
 166
 167         en = kmalloc(sizeof(struct ip_vs_lblc_entry), GFP_ATOMIC);
 168         if (en == NULL) {
 169                 IP_VS_ERR("ip_vs_lblc_new(): no memory\n");
 170                 return NULL;
 171         }
 172
 173         INIT_LIST_HEAD(&en->list);
 174         en->addr = daddr;
 175
 176         atomic_inc(&dest->refcnt);
 177         en->dest = dest;
 178
 179         return en;
 180 }
 181
 182
 183 static inline void ip_vs_lblc_free(struct ip_vs_lblc_entry *en)
 184 {
 185         list_del(&en->list);
 186         /*
 187          * We don't kfree dest because it is refered either by its service
 188          * or the trash dest list.
 189          */
 190         atomic_dec(&en->dest->refcnt);
 191         kfree(en);
 192 }
 193
 194
 195 /*
 196  *      Returns hash value for IPVS LBLC entry
 197  */
 198 static inline unsigned ip_vs_lblc_hashkey(__u32 addr)
 199 {
 200         return (ntohl(addr)*2654435761UL) & IP_VS_LBLC_TAB_MASK;
 201 }
 202
 203
 204 /*
 205  *      Hash an entry in the ip_vs_lblc_table.
 206  *      returns bool success.
 207  */
 208 static int
 209 ip_vs_lblc_hash(struct ip_vs_lblc_table *tbl, struct ip_vs_lblc_entry *en)
 210 {
 211         unsigned hash;
 212
 213         if (!list_empty(&en->list)) {
 214                 IP_VS_ERR("ip_vs_lblc_hash(): request for already hashed, "
 215                           "called from %p\n", __builtin_return_address(0));
 216                 return 0;
 217         }
 218
 219         /*
 220          *      Hash by destination IP address
 221          */
 222         hash = ip_vs_lblc_hashkey(en->addr);
 223
 224         write_lock(&tbl->lock);
 225         list_add(&en->list, &tbl->bucket[hash]);
 226         atomic_inc(&tbl->entries);
 227         write_unlock(&tbl->lock);
 228
 229         return 1;
 230 }
 231
 232
 233 /*
 234  *  Get ip_vs_lblc_entry associated with supplied parameters.
 235  */
 236 static inline struct ip_vs_lblc_entry *
 237 ip_vs_lblc_get(struct ip_vs_lblc_table *tbl, __u32 addr)
 238 {
 239         unsigned hash;
 240         struct ip_vs_lblc_entry *en;
 241
 242         hash = ip_vs_lblc_hashkey(addr);
 243
 244         read_lock(&tbl->lock);
 245
 246         list_for_each_entry(en, &tbl->bucket[hash], list) {
 247                 if (en->addr == addr) {
 248                         /* HIT */
 249                         read_unlock(&tbl->lock);
 250                         return en;
 251                 }
 252         }
 253
 254         read_unlock(&tbl->lock);
 255
 256         return NULL;
 257 }
 258
 259
 260 /*
 261  *      Flush all the entries of the specified table.
 262  */
 263 static void ip_vs_lblc_flush(struct ip_vs_lblc_table *tbl)
 264 {
 265         int i;
 266         struct ip_vs_lblc_entry *en, *nxt;
 267
 268         for (i=0; i<IP_VS_LBLC_TAB_SIZE; i++) {
 269                 write_lock(&tbl->lock);
 270                 list_for_each_entry_safe(en, nxt, &tbl->bucket[i], list) {
 271                         ip_vs_lblc_free(en);
 272                         atomic_dec(&tbl->entries);
 273                 }
 274                 write_unlock(&tbl->lock);
 275         }
 276 }
 277
 278
 279 static inline void ip_vs_lblc_full_check(struct ip_vs_lblc_table *tbl)
 280 {
 281         unsigned long now = jiffies;
 282         int i, j;
 283         struct ip_vs_lblc_entry *en, *nxt;
 284
 285         for (i=0, j=tbl->rover; i<IP_VS_LBLC_TAB_SIZE; i++) {
 286                 j = (j + 1) & IP_VS_LBLC_TAB_MASK;
 287
 288                 write_lock(&tbl->lock);
 289                 list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) {
 290                         if (time_before(now,
 291                                         en->lastuse + sysctl_ip_vs_lblc_expiration))
 292                                 continue;
 293
 294                         ip_vs_lblc_free(en);
 295                         atomic_dec(&tbl->entries);
 296                 }
 297                 write_unlock(&tbl->lock);
 298         }
 299         tbl->rover = j;
 300 }
 301
 302
 303 /*
 304  *      Periodical timer handler for IPVS lblc table
 305  *      It is used to collect stale entries when the number of entries
 306  *      exceeds the maximum size of the table.
 307  *
 308  *      Fixme: we probably need more complicated algorithm to collect
 309  *             entries that have not been used for a long time even
 310  *             if the number of entries doesn't exceed the maximum size
 311  *             of the table.
 312  *      The full expiration check is for this purpose now.
 313  */
 314 static void ip_vs_lblc_check_expire(unsigned long data)
 315 {
 316         struct ip_vs_lblc_table *tbl;
 317         unsigned long now = jiffies;
 318         int goal;
 319         int i, j;
 320         struct ip_vs_lblc_entry *en, *nxt;
 321
 322         tbl = (struct ip_vs_lblc_table *)data;
 323
 324         if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) {
 325                 /* do full expiration check */
 326                 ip_vs_lblc_full_check(tbl);
 327                 tbl->counter = 1;
 328                 goto out;
 329         }
 330
 331         if (atomic_read(&tbl->entries) <= tbl->max_size) {
 332                 tbl->counter++;
 333                 goto out;
 334         }
 335
 336         goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3;
 337         if (goal > tbl->max_size/2)
 338                 goal = tbl->max_size/2;
 339
 340         for (i=0, j=tbl->rover; i<IP_VS_LBLC_TAB_SIZE; i++) {
 341                 j = (j + 1) & IP_VS_LBLC_TAB_MASK;
 342
 343                 write_lock(&tbl->lock);
 344                 list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) {
 345                         if (time_before(now, en->lastuse + ENTRY_TIMEOUT))
 346                                 continue;
 347
 348                         ip_vs_lblc_free(en);
 349                         atomic_dec(&tbl->entries);
 350                         goal--;
 351                 }
 352                 write_unlock(&tbl->lock);
 353                 if (goal <= 0)
 354                         break;
 355         }
 356         tbl->rover = j;
 357
 358   out:
 359         mod_timer(&tbl->periodic_timer, jiffies+CHECK_EXPIRE_INTERVAL);
 360 }
 361
 362
 363 static int ip_vs_lblc_init_svc(struct ip_vs_service *svc)
 364 {
 365         int i;
 366         struct ip_vs_lblc_table *tbl;
 367
 368         /*
 369          *    Allocate the ip_vs_lblc_table for this service
 370          */
 371         tbl = kmalloc(sizeof(struct ip_vs_lblc_table), GFP_ATOMIC);
 372         if (tbl == NULL) {
 373                 IP_VS_ERR("ip_vs_lblc_init_svc(): no memory\n");
 374                 return -ENOMEM;
 375         }
 376         svc->sched_data = tbl;
 377         IP_VS_DBG(6, "LBLC hash table (memory=%Zdbytes) allocated for "
 378                   "current service\n",
 379                   sizeof(struct ip_vs_lblc_table));
 380
 381         /*
 382          *    Initialize the hash buckets
 383          */
 384         for (i=0; i<IP_VS_LBLC_TAB_SIZE; i++) {
 385                 INIT_LIST_HEAD(&tbl->bucket[i]);
 386         }
 387         rwlock_init(&tbl->lock);
 388         tbl->max_size = IP_VS_LBLC_TAB_SIZE*16;
 389         tbl->rover = 0;
 390         tbl->counter = 1;
 391
 392         /*
 393          *    Hook periodic timer for garbage collection
 394          */
 395         init_timer(&tbl->periodic_timer);
 396         tbl->periodic_timer.data = (unsigned long)tbl;
 397         tbl->periodic_timer.function = ip_vs_lblc_check_expire;
 398         tbl->periodic_timer.expires = jiffies+CHECK_EXPIRE_INTERVAL;
 399         add_timer(&tbl->periodic_timer);
 400
 401         return 0;
 402 }
 403
 404
 405 static int ip_vs_lblc_done_svc(struct ip_vs_service *svc)
 406 {
 407         struct ip_vs_lblc_table *tbl = svc->sched_data;
 408
 409         /* remove periodic timer */
 410         del_timer_sync(&tbl->periodic_timer);
 411
 412         /* got to clean up table entries here */
 413         ip_vs_lblc_flush(tbl);
 414
 415         /* release the table itself */
 416         kfree(svc->sched_data);
 417         IP_VS_DBG(6, "LBLC hash table (memory=%Zdbytes) released\n",
 418                   sizeof(struct ip_vs_lblc_table));
 419
 420         return 0;
 421 }
 422
 423
 424 static int ip_vs_lblc_update_svc(struct ip_vs_service *svc)
 425 {
 426         return 0;
 427 }
 428
 429
 430 static inline struct ip_vs_dest *
 431 __ip_vs_wlc_schedule(struct ip_vs_service *svc, struct iphdr *iph)
 432 {
 433         struct ip_vs_dest *dest, *least;
 434         int loh, doh;
 435
 436         /*
 437          * We think the overhead of processing active connections is fifty
 438          * times higher than that of inactive connections in average. (This
 439          * fifty times might not be accurate, we will change it later.) We
 440          * use the following formula to estimate the overhead:
 441          *                dest->activeconns*50 + dest->inactconns
 442          * and the load:
 443          *                (dest overhead) / dest->weight
 444          *
 445          * Remember -- no floats in kernel mode!!!
 446          * The comparison of h1*w2 > h2*w1 is equivalent to that of
 447          *                h1/w1 > h2/w2
 448          * if every weight is larger than zero.
 449          *
 450          * The server with weight=0 is quiesced and will not receive any
 451          * new connection.
 452          */
 453         list_for_each_entry(dest, &svc->destinations, n_list) {
 454                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 455                         continue;
 456                 if (atomic_read(&dest->weight) > 0) {
 457                         least = dest;
 458                         loh = atomic_read(&least->activeconns) * 50
 459                                 + atomic_read(&least->inactconns);
 460                         goto nextstage;
 461                 }
 462         }
 463         return NULL;
 464
 465         /*
 466          *    Find the destination with the least load.
 467          */
 468   nextstage:
 469         list_for_each_entry_continue(dest, &svc->destinations, n_list) {
 470                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 471                         continue;
 472
 473                 doh = atomic_read(&dest->activeconns) * 50
 474                         + atomic_read(&dest->inactconns);
 475                 if (loh * atomic_read(&dest->weight) >
 476                     doh * atomic_read(&least->weight)) {
 477                         least = dest;
 478                         loh = doh;
 479                 }
 480         }
 481
 482         IP_VS_DBG(6, "LBLC: server %d.%d.%d.%d:%d "
 483                   "activeconns %d refcnt %d weight %d overhead %d\n",
 484                   NIPQUAD(least->addr), ntohs(least->port),
 485                   atomic_read(&least->activeconns),
 486                   atomic_read(&least->refcnt),
 487                   atomic_read(&least->weight), loh);
 488
 489         return least;
 490 }
 491
 492
 493 /*
 494  *   If this destination server is overloaded and there is a less loaded
 495  *   server, then return true.
 496  */
 497 static inline int
 498 is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc)
 499 {
 500         if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) {
 501                 struct ip_vs_dest *d;
 502
 503                 list_for_each_entry(d, &svc->destinations, n_list) {
 504                         if (atomic_read(&d->activeconns)*2
 505                             < atomic_read(&d->weight)) {
 506                                 return 1;
 507                         }
 508                 }
 509         }
 510         return 0;
 511 }
 512
 513
 514 /*
 515  *    Locality-Based (weighted) Least-Connection scheduling
 516  */
 517 static struct ip_vs_dest *
 518 ip_vs_lblc_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
 519 {
 520         struct ip_vs_dest *dest;
 521         struct ip_vs_lblc_table *tbl;
 522         struct ip_vs_lblc_entry *en;
 523         struct iphdr *iph = skb->nh.iph;
 524
 525         IP_VS_DBG(6, "ip_vs_lblc_schedule(): Scheduling...\n");
 526
 527         tbl = (struct ip_vs_lblc_table *)svc->sched_data;
 528         en = ip_vs_lblc_get(tbl, iph->daddr);
 529         if (en == NULL) {
 530                 dest = __ip_vs_wlc_schedule(svc, iph);
 531                 if (dest == NULL) {
 532                         IP_VS_DBG(1, "no destination available\n");
 533                         return NULL;
 534                 }
 535                 en = ip_vs_lblc_new(iph->daddr, dest);
 536                 if (en == NULL) {
 537                         return NULL;
 538                 }
 539                 ip_vs_lblc_hash(tbl, en);
 540         } else {
 541                 dest = en->dest;
 542                 if (!(dest->flags & IP_VS_DEST_F_AVAILABLE)
 543                     || atomic_read(&dest->weight) <= 0
 544                     || is_overloaded(dest, svc)) {
 545                         dest = __ip_vs_wlc_schedule(svc, iph);
 546                         if (dest == NULL) {
 547                                 IP_VS_DBG(1, "no destination available\n");
 548                                 return NULL;
 549                         }
 550                         atomic_dec(&en->dest->refcnt);
 551                         atomic_inc(&dest->refcnt);
 552                         en->dest = dest;
 553                 }
 554         }
 555         en->lastuse = jiffies;
 556
 557         IP_VS_DBG(6, "LBLC: destination IP address %u.%u.%u.%u "
 558                   "--> server %u.%u.%u.%u:%d\n",
 559                   NIPQUAD(en->addr),
 560                   NIPQUAD(dest->addr),
 561                   ntohs(dest->port));
 562
 563         return dest;
 564 }
 565
 566
 567 /*
 568  *      IPVS LBLC Scheduler structure
 569  */
 570 static struct ip_vs_scheduler ip_vs_lblc_scheduler =
 571 {
 572         .name =                 "lblc",
 573         .refcnt =               ATOMIC_INIT(0),
 574         .module =               THIS_MODULE,
 575         .init_service =         ip_vs_lblc_init_svc,
 576         .done_service =         ip_vs_lblc_done_svc,
 577         .update_service =       ip_vs_lblc_update_svc,
 578         .schedule =             ip_vs_lblc_schedule,
 579 };
 580
 581
 582 static int __init ip_vs_lblc_init(void)
 583 {
 584         INIT_LIST_HEAD(&ip_vs_lblc_scheduler.n_list);
 585         sysctl_header = register_sysctl_table(lblc_root_table, 0);
 586         return register_ip_vs_scheduler(&ip_vs_lblc_scheduler);
 587 }
 588
 589
 590 static void __exit ip_vs_lblc_cleanup(void)
 591 {
 592         unregister_sysctl_table(sysctl_header);
 593         unregister_ip_vs_scheduler(&ip_vs_lblc_scheduler);
 594 }
 595
 596
 597 module_init(ip_vs_lblc_init);
 598 module_exit(ip_vs_lblc_cleanup);
 599 MODULE_LICENSE("GPL");