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 #include <linux/list.h>
  49 #include <linux/slab.h>
  50
  51 /* for sysctl */
  52 #include <linux/fs.h>
  53 #include <linux/sysctl.h>
  54 #include <net/net_namespace.h>
  55
  56 #include <net/ip_vs.h>
  57
  58
  59 /*
  60  *    It is for garbage collection of stale IPVS lblcr entries,
  61  *    when the table is full.
  62  */
  63 #define CHECK_EXPIRE_INTERVAL   (60*HZ)
  64 #define ENTRY_TIMEOUT           (6*60*HZ)
  65
  66 #define DEFAULT_EXPIRATION      (24*60*60*HZ)
  67
  68 /*
  69  *    It is for full expiration check.
  70  *    When there is no partial expiration check (garbage collection)
  71  *    in a half hour, do a full expiration check to collect stale
  72  *    entries that haven't been touched for a day.
  73  */
  74 #define COUNT_FOR_FULL_EXPIRATION   30
  75
  76 /*
  77  *     for IPVS lblcr entry hash table
  78  */
  79 #ifndef CONFIG_IP_VS_LBLCR_TAB_BITS
  80 #define CONFIG_IP_VS_LBLCR_TAB_BITS      10
  81 #endif
  82 #define IP_VS_LBLCR_TAB_BITS     CONFIG_IP_VS_LBLCR_TAB_BITS
  83 #define IP_VS_LBLCR_TAB_SIZE     (1 << IP_VS_LBLCR_TAB_BITS)
  84 #define IP_VS_LBLCR_TAB_MASK     (IP_VS_LBLCR_TAB_SIZE - 1)
  85
  86
  87 /*
  88  *      IPVS destination set structure and operations
  89  */
  90 struct ip_vs_dest_set_elem {
  91         struct list_head        list;          /* list link */
  92         struct ip_vs_dest       *dest;          /* destination server */
  93 };
  94
  95 struct ip_vs_dest_set {
  96         atomic_t                size;           /* set size */
  97         unsigned long           lastmod;        /* last modified time */
  98         struct list_head        list;           /* destination list */
  99         rwlock_t                lock;           /* lock for this list */
 100 };
 101
 102
 103 static struct ip_vs_dest_set_elem *
 104 ip_vs_dest_set_insert(struct ip_vs_dest_set *set, struct ip_vs_dest *dest)
 105 {
 106         struct ip_vs_dest_set_elem *e;
 107
 108         list_for_each_entry(e, &set->list, list) {
 109                 if (e->dest == dest)
 110                         /* already existed */
 111                         return NULL;
 112         }
 113
 114         e = kmalloc(sizeof(*e), GFP_ATOMIC);
 115         if (e == NULL)
 116                 return NULL;
 117
 118         atomic_inc(&dest->refcnt);
 119         e->dest = dest;
 120
 121         list_add(&e->list, &set->list);
 122         atomic_inc(&set->size);
 123
 124         set->lastmod = jiffies;
 125         return e;
 126 }
 127
 128 static void
 129 ip_vs_dest_set_erase(struct ip_vs_dest_set *set, struct ip_vs_dest *dest)
 130 {
 131         struct ip_vs_dest_set_elem *e;
 132
 133         list_for_each_entry(e, &set->list, list) {
 134                 if (e->dest == dest) {
 135                         /* HIT */
 136                         atomic_dec(&set->size);
 137                         set->lastmod = jiffies;
 138                         atomic_dec(&e->dest->refcnt);
 139                         list_del(&e->list);
 140                         kfree(e);
 141                         break;
 142                 }
 143         }
 144 }
 145
 146 static void ip_vs_dest_set_eraseall(struct ip_vs_dest_set *set)
 147 {
 148         struct ip_vs_dest_set_elem *e, *ep;
 149
 150         write_lock(&set->lock);
 151         list_for_each_entry_safe(e, ep, &set->list, list) {
 152                 /*
 153                  * We don't kfree dest because it is referred either
 154                  * by its service or by the trash dest list.
 155                  */
 156                 atomic_dec(&e->dest->refcnt);
 157                 list_del(&e->list);
 158                 kfree(e);
 159         }
 160         write_unlock(&set->lock);
 161 }
 162
 163 /* get weighted least-connection node in the destination set */
 164 static inline struct ip_vs_dest *ip_vs_dest_set_min(struct ip_vs_dest_set *set)
 165 {
 166         register struct ip_vs_dest_set_elem *e;
 167         struct ip_vs_dest *dest, *least;
 168         int loh, doh;
 169
 170         if (set == NULL)
 171                 return NULL;
 172
 173         /* select the first destination server, whose weight > 0 */
 174         list_for_each_entry(e, &set->list, list) {
 175                 least = e->dest;
 176                 if (least->flags & IP_VS_DEST_F_OVERLOAD)
 177                         continue;
 178
 179                 if ((atomic_read(&least->weight) > 0)
 180                     && (least->flags & IP_VS_DEST_F_AVAILABLE)) {
 181                         loh = ip_vs_dest_conn_overhead(least);
 182                         goto nextstage;
 183                 }
 184         }
 185         return NULL;
 186
 187         /* find the destination with the weighted least load */
 188   nextstage:
 189         list_for_each_entry(e, &set->list, list) {
 190                 dest = e->dest;
 191                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 192                         continue;
 193
 194                 doh = ip_vs_dest_conn_overhead(dest);
 195                 if ((loh * atomic_read(&dest->weight) >
 196                      doh * atomic_read(&least->weight))
 197                     && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
 198                         least = dest;
 199                         loh = doh;
 200                 }
 201         }
 202
 203         IP_VS_DBG_BUF(6, "%s(): server %s:%d "
 204                       "activeconns %d refcnt %d weight %d overhead %d\n",
 205                       __func__,
 206                       IP_VS_DBG_ADDR(least->af, &least->addr),
 207                       ntohs(least->port),
 208                       atomic_read(&least->activeconns),
 209                       atomic_read(&least->refcnt),
 210                       atomic_read(&least->weight), loh);
 211         return least;
 212 }
 213
 214
 215 /* get weighted most-connection node in the destination set */
 216 static inline struct ip_vs_dest *ip_vs_dest_set_max(struct ip_vs_dest_set *set)
 217 {
 218         register struct ip_vs_dest_set_elem *e;
 219         struct ip_vs_dest *dest, *most;
 220         int moh, doh;
 221
 222         if (set == NULL)
 223                 return NULL;
 224
 225         /* select the first destination server, whose weight > 0 */
 226         list_for_each_entry(e, &set->list, list) {
 227                 most = e->dest;
 228                 if (atomic_read(&most->weight) > 0) {
 229                         moh = ip_vs_dest_conn_overhead(most);
 230                         goto nextstage;
 231                 }
 232         }
 233         return NULL;
 234
 235         /* find the destination with the weighted most load */
 236   nextstage:
 237         list_for_each_entry(e, &set->list, list) {
 238                 dest = e->dest;
 239                 doh = ip_vs_dest_conn_overhead(dest);
 240                 /* moh/mw < doh/dw ==> moh*dw < doh*mw, where mw,dw>0 */
 241                 if ((moh * atomic_read(&dest->weight) <
 242                      doh * atomic_read(&most->weight))
 243                     && (atomic_read(&dest->weight) > 0)) {
 244                         most = dest;
 245                         moh = doh;
 246                 }
 247         }
 248
 249         IP_VS_DBG_BUF(6, "%s(): server %s:%d "
 250                       "activeconns %d refcnt %d weight %d overhead %d\n",
 251                       __func__,
 252                       IP_VS_DBG_ADDR(most->af, &most->addr), ntohs(most->port),
 253                       atomic_read(&most->activeconns),
 254                       atomic_read(&most->refcnt),
 255                       atomic_read(&most->weight), moh);
 256         return most;
 257 }
 258
 259
 260 /*
 261  *      IPVS lblcr entry represents an association between destination
 262  *      IP address and its destination server set
 263  */
 264 struct ip_vs_lblcr_entry {
 265         struct list_head        list;
 266         int                     af;             /* address family */
 267         union nf_inet_addr      addr;           /* destination IP address */
 268         struct ip_vs_dest_set   set;            /* destination server set */
 269         unsigned long           lastuse;        /* last used time */
 270 };
 271
 272
 273 /*
 274  *      IPVS lblcr hash table
 275  */
 276 struct ip_vs_lblcr_table {
 277         struct list_head        bucket[IP_VS_LBLCR_TAB_SIZE];  /* hash bucket */
 278         atomic_t                entries;        /* number of entries */
 279         int                     max_size;       /* maximum size of entries */
 280         struct timer_list       periodic_timer; /* collect stale entries */
 281         int                     rover;          /* rover for expire check */
 282         int                     counter;        /* counter for no expire */
 283 };
 284
 285
 286 #ifdef CONFIG_SYSCTL
 287 /*
 288  *      IPVS LBLCR sysctl table
 289  */
 290
 291 static ctl_table vs_vars_table[] = {
 292         {
 293                 .procname       = "lblcr_expiration",
 294                 .data           = NULL,
 295                 .maxlen         = sizeof(int),
 296                 .mode           = 0644,
 297                 .proc_handler   = proc_dointvec_jiffies,
 298         },
 299         { }
 300 };
 301 #endif
 302
 303 static inline void ip_vs_lblcr_free(struct ip_vs_lblcr_entry *en)
 304 {
 305         list_del(&en->list);
 306         ip_vs_dest_set_eraseall(&en->set);
 307         kfree(en);
 308 }
 309
 310
 311 /*
 312  *      Returns hash value for IPVS LBLCR entry
 313  */
 314 static inline unsigned
 315 ip_vs_lblcr_hashkey(int af, const union nf_inet_addr *addr)
 316 {
 317         __be32 addr_fold = addr->ip;
 318
 319 #ifdef CONFIG_IP_VS_IPV6
 320         if (af == AF_INET6)
 321                 addr_fold = addr->ip6[0]^addr->ip6[1]^
 322                             addr->ip6[2]^addr->ip6[3];
 323 #endif
 324         return (ntohl(addr_fold)*2654435761UL) & IP_VS_LBLCR_TAB_MASK;
 325 }
 326
 327
 328 /*
 329  *      Hash an entry in the ip_vs_lblcr_table.
 330  *      returns bool success.
 331  */
 332 static void
 333 ip_vs_lblcr_hash(struct ip_vs_lblcr_table *tbl, struct ip_vs_lblcr_entry *en)
 334 {
 335         unsigned hash = ip_vs_lblcr_hashkey(en->af, &en->addr);
 336
 337         list_add(&en->list, &tbl->bucket[hash]);
 338         atomic_inc(&tbl->entries);
 339 }
 340
 341
 342 /*
 343  *  Get ip_vs_lblcr_entry associated with supplied parameters. Called under
 344  *  read lock.
 345  */
 346 static inline struct ip_vs_lblcr_entry *
 347 ip_vs_lblcr_get(int af, struct ip_vs_lblcr_table *tbl,
 348                 const union nf_inet_addr *addr)
 349 {
 350         unsigned hash = ip_vs_lblcr_hashkey(af, addr);
 351         struct ip_vs_lblcr_entry *en;
 352
 353         list_for_each_entry(en, &tbl->bucket[hash], list)
 354                 if (ip_vs_addr_equal(af, &en->addr, addr))
 355                         return en;
 356
 357         return NULL;
 358 }
 359
 360
 361 /*
 362  * Create or update an ip_vs_lblcr_entry, which is a mapping of a destination
 363  * IP address to a server. Called under write lock.
 364  */
 365 static inline struct ip_vs_lblcr_entry *
 366 ip_vs_lblcr_new(struct ip_vs_lblcr_table *tbl, const union nf_inet_addr *daddr,
 367                 struct ip_vs_dest *dest)
 368 {
 369         struct ip_vs_lblcr_entry *en;
 370
 371         en = ip_vs_lblcr_get(dest->af, tbl, daddr);
 372         if (!en) {
 373                 en = kmalloc(sizeof(*en), GFP_ATOMIC);
 374                 if (!en)
 375                         return NULL;
 376
 377                 en->af = dest->af;
 378                 ip_vs_addr_copy(dest->af, &en->addr, daddr);
 379                 en->lastuse = jiffies;
 380
 381                 /* initialize its dest set */
 382                 atomic_set(&(en->set.size), 0);
 383                 INIT_LIST_HEAD(&en->set.list);
 384                 rwlock_init(&en->set.lock);
 385
 386                 ip_vs_lblcr_hash(tbl, en);
 387         }
 388
 389         write_lock(&en->set.lock);
 390         ip_vs_dest_set_insert(&en->set, dest);
 391         write_unlock(&en->set.lock);
 392
 393         return en;
 394 }
 395
 396
 397 /*
 398  *      Flush all the entries of the specified table.
 399  */
 400 static void ip_vs_lblcr_flush(struct ip_vs_lblcr_table *tbl)
 401 {
 402         int i;
 403         struct ip_vs_lblcr_entry *en, *nxt;
 404
 405         /* No locking required, only called during cleanup. */
 406         for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) {
 407                 list_for_each_entry_safe(en, nxt, &tbl->bucket[i], list) {
 408                         ip_vs_lblcr_free(en);
 409                 }
 410         }
 411 }
 412
 413 static int sysctl_lblcr_expiration(struct ip_vs_service *svc)
 414 {
 415 #ifdef CONFIG_SYSCTL
 416         struct netns_ipvs *ipvs = net_ipvs(svc->net);
 417         return ipvs->sysctl_lblcr_expiration;
 418 #else
 419         return DEFAULT_EXPIRATION;
 420 #endif
 421 }
 422
 423 static inline void ip_vs_lblcr_full_check(struct ip_vs_service *svc)
 424 {
 425         struct ip_vs_lblcr_table *tbl = svc->sched_data;
 426         unsigned long now = jiffies;
 427         int i, j;
 428         struct ip_vs_lblcr_entry *en, *nxt;
 429
 430         for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) {
 431                 j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
 432
 433                 write_lock(&svc->sched_lock);
 434                 list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) {
 435                         if (time_after(en->lastuse +
 436                                        sysctl_lblcr_expiration(svc), now))
 437                                 continue;
 438
 439                         ip_vs_lblcr_free(en);
 440                         atomic_dec(&tbl->entries);
 441                 }
 442                 write_unlock(&svc->sched_lock);
 443         }
 444         tbl->rover = j;
 445 }
 446
 447
 448 /*
 449  *      Periodical timer handler for IPVS lblcr table
 450  *      It is used to collect stale entries when the number of entries
 451  *      exceeds the maximum size of the table.
 452  *
 453  *      Fixme: we probably need more complicated algorithm to collect
 454  *             entries that have not been used for a long time even
 455  *             if the number of entries doesn't exceed the maximum size
 456  *             of the table.
 457  *      The full expiration check is for this purpose now.
 458  */
 459 static void ip_vs_lblcr_check_expire(unsigned long data)
 460 {
 461         struct ip_vs_service *svc = (struct ip_vs_service *) data;
 462         struct ip_vs_lblcr_table *tbl = svc->sched_data;
 463         unsigned long now = jiffies;
 464         int goal;
 465         int i, j;
 466         struct ip_vs_lblcr_entry *en, *nxt;
 467
 468         if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) {
 469                 /* do full expiration check */
 470                 ip_vs_lblcr_full_check(svc);
 471                 tbl->counter = 1;
 472                 goto out;
 473         }
 474
 475         if (atomic_read(&tbl->entries) <= tbl->max_size) {
 476                 tbl->counter++;
 477                 goto out;
 478         }
 479
 480         goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3;
 481         if (goal > tbl->max_size/2)
 482                 goal = tbl->max_size/2;
 483
 484         for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) {
 485                 j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
 486
 487                 write_lock(&svc->sched_lock);
 488                 list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) {
 489                         if (time_before(now, en->lastuse+ENTRY_TIMEOUT))
 490                                 continue;
 491
 492                         ip_vs_lblcr_free(en);
 493                         atomic_dec(&tbl->entries);
 494                         goal--;
 495                 }
 496                 write_unlock(&svc->sched_lock);
 497                 if (goal <= 0)
 498                         break;
 499         }
 500         tbl->rover = j;
 501
 502   out:
 503         mod_timer(&tbl->periodic_timer, jiffies+CHECK_EXPIRE_INTERVAL);
 504 }
 505
 506 static int ip_vs_lblcr_init_svc(struct ip_vs_service *svc)
 507 {
 508         int i;
 509         struct ip_vs_lblcr_table *tbl;
 510
 511         /*
 512          *    Allocate the ip_vs_lblcr_table for this service
 513          */
 514         tbl = kmalloc(sizeof(*tbl), GFP_ATOMIC);
 515         if (tbl == NULL)
 516                 return -ENOMEM;
 517
 518         svc->sched_data = tbl;
 519         IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) allocated for "
 520                   "current service\n", sizeof(*tbl));
 521
 522         /*
 523          *    Initialize the hash buckets
 524          */
 525         for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) {
 526                 INIT_LIST_HEAD(&tbl->bucket[i]);
 527         }
 528         tbl->max_size = IP_VS_LBLCR_TAB_SIZE*16;
 529         tbl->rover = 0;
 530         tbl->counter = 1;
 531
 532         /*
 533          *    Hook periodic timer for garbage collection
 534          */
 535         setup_timer(&tbl->periodic_timer, ip_vs_lblcr_check_expire,
 536                         (unsigned long)svc);
 537         mod_timer(&tbl->periodic_timer, jiffies + CHECK_EXPIRE_INTERVAL);
 538
 539         return 0;
 540 }
 541
 542
 543 static int ip_vs_lblcr_done_svc(struct ip_vs_service *svc)
 544 {
 545         struct ip_vs_lblcr_table *tbl = svc->sched_data;
 546
 547         /* remove periodic timer */
 548         del_timer_sync(&tbl->periodic_timer);
 549
 550         /* got to clean up table entries here */
 551         ip_vs_lblcr_flush(tbl);
 552
 553         /* release the table itself */
 554         kfree(tbl);
 555         IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) released\n",
 556                   sizeof(*tbl));
 557
 558         return 0;
 559 }
 560
 561
 562 static inline struct ip_vs_dest *
 563 __ip_vs_lblcr_schedule(struct ip_vs_service *svc)
 564 {
 565         struct ip_vs_dest *dest, *least;
 566         int loh, doh;
 567
 568         /*
 569          * We use the following formula to estimate the load:
 570          *                (dest overhead) / dest->weight
 571          *
 572          * Remember -- no floats in kernel mode!!!
 573          * The comparison of h1*w2 > h2*w1 is equivalent to that of
 574          *                h1/w1 > h2/w2
 575          * if every weight is larger than zero.
 576          *
 577          * The server with weight=0 is quiesced and will not receive any
 578          * new connection.
 579          */
 580         list_for_each_entry(dest, &svc->destinations, n_list) {
 581                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 582                         continue;
 583
 584                 if (atomic_read(&dest->weight) > 0) {
 585                         least = dest;
 586                         loh = ip_vs_dest_conn_overhead(least);
 587                         goto nextstage;
 588                 }
 589         }
 590         return NULL;
 591
 592         /*
 593          *    Find the destination with the least load.
 594          */
 595   nextstage:
 596         list_for_each_entry_continue(dest, &svc->destinations, n_list) {
 597                 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
 598                         continue;
 599
 600                 doh = ip_vs_dest_conn_overhead(dest);
 601                 if (loh * atomic_read(&dest->weight) >
 602                     doh * atomic_read(&least->weight)) {
 603                         least = dest;
 604                         loh = doh;
 605                 }
 606         }
 607
 608         IP_VS_DBG_BUF(6, "LBLCR: server %s:%d "
 609                       "activeconns %d refcnt %d weight %d overhead %d\n",
 610                       IP_VS_DBG_ADDR(least->af, &least->addr),
 611                       ntohs(least->port),
 612                       atomic_read(&least->activeconns),
 613                       atomic_read(&least->refcnt),
 614                       atomic_read(&least->weight), loh);
 615
 616         return least;
 617 }
 618
 619
 620 /*
 621  *   If this destination server is overloaded and there is a less loaded
 622  *   server, then return true.
 623  */
 624 static inline int
 625 is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc)
 626 {
 627         if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) {
 628                 struct ip_vs_dest *d;
 629
 630                 list_for_each_entry(d, &svc->destinations, n_list) {
 631                         if (atomic_read(&d->activeconns)*2
 632                             < atomic_read(&d->weight)) {
 633                                 return 1;
 634                         }
 635                 }
 636         }
 637         return 0;
 638 }
 639
 640
 641 /*
 642  *    Locality-Based (weighted) Least-Connection scheduling
 643  */
 644 static struct ip_vs_dest *
 645 ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
 646 {
 647         struct ip_vs_lblcr_table *tbl = svc->sched_data;
 648         struct ip_vs_iphdr iph;
 649         struct ip_vs_dest *dest = NULL;
 650         struct ip_vs_lblcr_entry *en;
 651
 652         ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
 653
 654         IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
 655
 656         /* First look in our cache */
 657         read_lock(&svc->sched_lock);
 658         en = ip_vs_lblcr_get(svc->af, tbl, &iph.daddr);
 659         if (en) {
 660                 /* We only hold a read lock, but this is atomic */
 661                 en->lastuse = jiffies;
 662
 663                 /* Get the least loaded destination */
 664                 read_lock(&en->set.lock);
 665                 dest = ip_vs_dest_set_min(&en->set);
 666                 read_unlock(&en->set.lock);
 667
 668                 /* More than one destination + enough time passed by, cleanup */
 669                 if (atomic_read(&en->set.size) > 1 &&
 670                                 time_after(jiffies, en->set.lastmod +
 671                                 sysctl_lblcr_expiration(svc))) {
 672                         struct ip_vs_dest *m;
 673
 674                         write_lock(&en->set.lock);
 675                         m = ip_vs_dest_set_max(&en->set);
 676                         if (m)
 677                                 ip_vs_dest_set_erase(&en->set, m);
 678                         write_unlock(&en->set.lock);
 679                 }
 680
 681                 /* If the destination is not overloaded, use it */
 682                 if (dest && !is_overloaded(dest, svc)) {
 683                         read_unlock(&svc->sched_lock);
 684                         goto out;
 685                 }
 686
 687                 /* The cache entry is invalid, time to schedule */
 688                 dest = __ip_vs_lblcr_schedule(svc);
 689                 if (!dest) {
 690                         ip_vs_scheduler_err(svc, "no destination available");
 691                         read_unlock(&svc->sched_lock);
 692                         return NULL;
 693                 }
 694
 695                 /* Update our cache entry */
 696                 write_lock(&en->set.lock);
 697                 ip_vs_dest_set_insert(&en->set, dest);
 698                 write_unlock(&en->set.lock);
 699         }
 700         read_unlock(&svc->sched_lock);
 701
 702         if (dest)
 703                 goto out;
 704
 705         /* No cache entry, time to schedule */
 706         dest = __ip_vs_lblcr_schedule(svc);
 707         if (!dest) {
 708                 IP_VS_DBG(1, "no destination available\n");
 709                 return NULL;
 710         }
 711
 712         /* If we fail to create a cache entry, we'll just use the valid dest */
 713         write_lock(&svc->sched_lock);
 714         ip_vs_lblcr_new(tbl, &iph.daddr, dest);
 715         write_unlock(&svc->sched_lock);
 716
 717 out:
 718         IP_VS_DBG_BUF(6, "LBLCR: destination IP address %s --> server %s:%d\n",
 719                       IP_VS_DBG_ADDR(svc->af, &iph.daddr),
 720                       IP_VS_DBG_ADDR(svc->af, &dest->addr), ntohs(dest->port));
 721
 722         return dest;
 723 }
 724
 725
 726 /*
 727  *      IPVS LBLCR Scheduler structure
 728  */
 729 static struct ip_vs_scheduler ip_vs_lblcr_scheduler =
 730 {
 731         .name =                 "lblcr",
 732         .refcnt =               ATOMIC_INIT(0),
 733         .module =               THIS_MODULE,
 734         .n_list =               LIST_HEAD_INIT(ip_vs_lblcr_scheduler.n_list),
 735         .init_service =         ip_vs_lblcr_init_svc,
 736         .done_service =         ip_vs_lblcr_done_svc,
 737         .schedule =             ip_vs_lblcr_schedule,
 738 };
 739
 740 /*
 741  *  per netns init.
 742  */
 743 #ifdef CONFIG_SYSCTL
 744 static int __net_init __ip_vs_lblcr_init(struct net *net)
 745 {
 746         struct netns_ipvs *ipvs = net_ipvs(net);
 747
 748         if (!net_eq(net, &init_net)) {
 749                 ipvs->lblcr_ctl_table = kmemdup(vs_vars_table,
 750                                                 sizeof(vs_vars_table),
 751                                                 GFP_KERNEL);
 752                 if (ipvs->lblcr_ctl_table == NULL)
 753                         return -ENOMEM;
 754         } else
 755                 ipvs->lblcr_ctl_table = vs_vars_table;
 756         ipvs->sysctl_lblcr_expiration = DEFAULT_EXPIRATION;
 757         ipvs->lblcr_ctl_table[0].data = &ipvs->sysctl_lblcr_expiration;
 758
 759         ipvs->lblcr_ctl_header =
 760                 register_net_sysctl_table(net, net_vs_ctl_path,
 761                                           ipvs->lblcr_ctl_table);
 762         if (!ipvs->lblcr_ctl_header) {
 763                 if (!net_eq(net, &init_net))
 764                         kfree(ipvs->lblcr_ctl_table);
 765                 return -ENOMEM;
 766         }
 767
 768         return 0;
 769 }
 770
 771 static void __net_exit __ip_vs_lblcr_exit(struct net *net)
 772 {
 773         struct netns_ipvs *ipvs = net_ipvs(net);
 774
 775         unregister_net_sysctl_table(ipvs->lblcr_ctl_header);
 776
 777         if (!net_eq(net, &init_net))
 778                 kfree(ipvs->lblcr_ctl_table);
 779 }
 780
 781 #else
 782
 783 static int __net_init __ip_vs_lblcr_init(struct net *net) { return 0; }
 784 static void __net_exit __ip_vs_lblcr_exit(struct net *net) { }
 785
 786 #endif
 787
 788 static struct pernet_operations ip_vs_lblcr_ops = {
 789         .init = __ip_vs_lblcr_init,
 790         .exit = __ip_vs_lblcr_exit,
 791 };
 792
 793 static int __init ip_vs_lblcr_init(void)
 794 {
 795         int ret;
 796
 797         ret = register_pernet_subsys(&ip_vs_lblcr_ops);
 798         if (ret)
 799                 return ret;
 800
 801         ret = register_ip_vs_scheduler(&ip_vs_lblcr_scheduler);
 802         if (ret)
 803                 unregister_pernet_subsys(&ip_vs_lblcr_ops);
 804         return ret;
 805 }
 806
 807 static void __exit ip_vs_lblcr_cleanup(void)
 808 {
 809         unregister_ip_vs_scheduler(&ip_vs_lblcr_scheduler);
 810         unregister_pernet_subsys(&ip_vs_lblcr_ops);
 811 }
 812
 813
 814 module_init(ip_vs_lblcr_init);
 815 module_exit(ip_vs_lblcr_cleanup);
 816 MODULE_LICENSE("GPL");