1// SPDX-License-Identifier: GPL-2.0-or-later 2/* 3 * IPVS: Locality-Based Least-Connection scheduling module 4 * 5 * Authors: Wensong Zhang <wensong@gnuchina.org> 6 * 7 * Changes: 8 * Martin Hamilton : fixed the terrible locking bugs 9 * *lock(tbl->lock) ==> *lock(&tbl->lock) 10 * Wensong Zhang : fixed the uninitialized tbl->lock bug 11 * Wensong Zhang : added doing full expiration check to 12 * collect stale entries of 24+ hours when 13 * no partial expire check in a half hour 14 * Julian Anastasov : replaced del_timer call with del_timer_sync 15 * to avoid the possible race between timer 16 * handler and del_timer thread in SMP 17 */ 18 19/* 20 * The lblc algorithm is as follows (pseudo code): 21 * 22 * if cachenode[dest_ip] is null then 23 * n, cachenode[dest_ip] <- {weighted least-conn node}; 24 * else 25 * n <- cachenode[dest_ip]; 26 * if (n is dead) OR 27 * (n.conns>n.weight AND 28 * there is a node m with m.conns<m.weight/2) then 29 * n, cachenode[dest_ip] <- {weighted least-conn node}; 30 * 31 * return n; 32 * 33 * Thanks must go to Wenzhuo Zhang for talking WCCP to me and pushing 34 * me to write this module. 35 */ 36 37#define KMSG_COMPONENT "IPVS" 38#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 39 40#include <linux/ip.h> 41#include <linux/slab.h> 42#include <linux/module.h> 43#include <linux/kernel.h> 44#include <linux/skbuff.h> 45#include <linux/jiffies.h> 46#include <linux/hash.h> 47 48/* for sysctl */ 49#include <linux/fs.h> 50#include <linux/sysctl.h> 51 52#include <net/ip_vs.h> 53 54 55/* 56 * It is for garbage collection of stale IPVS lblc entries, 57 * when the table is full. 58 */ 59#define CHECK_EXPIRE_INTERVAL (60*HZ) 60#define ENTRY_TIMEOUT (6*60*HZ) 61 62#define DEFAULT_EXPIRATION (24*60*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 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 */ 88struct ip_vs_lblc_entry { 89 struct hlist_node list; 90 int af; /* address family */ 91 union nf_inet_addr addr; /* destination IP address */ 92 struct ip_vs_dest *dest; /* real server (cache) */ 93 unsigned long lastuse; /* last used time */ 94 struct rcu_head rcu_head; 95}; 96 97 98/* 99 * IPVS lblc hash table 100 */ 101struct ip_vs_lblc_table { 102 struct rcu_head rcu_head; 103 struct hlist_head bucket[IP_VS_LBLC_TAB_SIZE]; /* hash bucket */ 104 struct timer_list periodic_timer; /* collect stale entries */ 105 struct ip_vs_service *svc; /* pointer back to service */ 106 atomic_t entries; /* number of entries */ 107 int max_size; /* maximum size of entries */ 108 int rover; /* rover for expire check */ 109 int counter; /* counter for no expire */ 110 bool dead; 111}; 112 113 114/* 115 * IPVS LBLC sysctl table 116 */ 117#ifdef CONFIG_SYSCTL 118static struct ctl_table vs_vars_table[] = { 119 { 120 .procname = "lblc_expiration", 121 .data = NULL, 122 .maxlen = sizeof(int), 123 .mode = 0644, 124 .proc_handler = proc_dointvec_jiffies, 125 }, 126 { } 127}; 128#endif 129 130static void ip_vs_lblc_rcu_free(struct rcu_head *head) 131{ 132 struct ip_vs_lblc_entry *en = container_of(head, 133 struct ip_vs_lblc_entry, 134 rcu_head); 135 136 ip_vs_dest_put_and_free(en->dest); 137 kfree(en); 138} 139 140static inline void ip_vs_lblc_del(struct ip_vs_lblc_entry *en) 141{ 142 hlist_del_rcu(&en->list); 143 call_rcu(&en->rcu_head, ip_vs_lblc_rcu_free); 144} 145 146/* 147 * Returns hash value for IPVS LBLC entry 148 */ 149static inline unsigned int 150ip_vs_lblc_hashkey(int af, const union nf_inet_addr *addr) 151{ 152 __be32 addr_fold = addr->ip; 153 154#ifdef CONFIG_IP_VS_IPV6 155 if (af == AF_INET6) 156 addr_fold = addr->ip6[0]^addr->ip6[1]^ 157 addr->ip6[2]^addr->ip6[3]; 158#endif 159 return hash_32(ntohl(addr_fold), IP_VS_LBLC_TAB_BITS); 160} 161 162 163/* 164 * Hash an entry in the ip_vs_lblc_table. 165 * returns bool success. 166 */ 167static void 168ip_vs_lblc_hash(struct ip_vs_lblc_table *tbl, struct ip_vs_lblc_entry *en) 169{ 170 unsigned int hash = ip_vs_lblc_hashkey(en->af, &en->addr); 171 172 hlist_add_head_rcu(&en->list, &tbl->bucket[hash]); 173 atomic_inc(&tbl->entries); 174} 175 176 177/* Get ip_vs_lblc_entry associated with supplied parameters. */ 178static inline struct ip_vs_lblc_entry * 179ip_vs_lblc_get(int af, struct ip_vs_lblc_table *tbl, 180 const union nf_inet_addr *addr) 181{ 182 unsigned int hash = ip_vs_lblc_hashkey(af, addr); 183 struct ip_vs_lblc_entry *en; 184 185 hlist_for_each_entry_rcu(en, &tbl->bucket[hash], list) 186 if (ip_vs_addr_equal(af, &en->addr, addr)) 187 return en; 188 189 return NULL; 190} 191 192 193/* 194 * Create or update an ip_vs_lblc_entry, which is a mapping of a destination IP 195 * address to a server. Called under spin lock. 196 */ 197static inline struct ip_vs_lblc_entry * 198ip_vs_lblc_new(struct ip_vs_lblc_table *tbl, const union nf_inet_addr *daddr, 199 u16 af, struct ip_vs_dest *dest) 200{ 201 struct ip_vs_lblc_entry *en; 202 203 en = ip_vs_lblc_get(af, tbl, daddr); 204 if (en) { 205 if (en->dest == dest) 206 return en; 207 ip_vs_lblc_del(en); 208 } 209 en = kmalloc(sizeof(*en), GFP_ATOMIC); 210 if (!en) 211 return NULL; 212 213 en->af = af; 214 ip_vs_addr_copy(af, &en->addr, daddr); 215 en->lastuse = jiffies; 216 217 ip_vs_dest_hold(dest); 218 en->dest = dest; 219 220 ip_vs_lblc_hash(tbl, en); 221 222 return en; 223} 224 225 226/* 227 * Flush all the entries of the specified table. 228 */ 229static void ip_vs_lblc_flush(struct ip_vs_service *svc) 230{ 231 struct ip_vs_lblc_table *tbl = svc->sched_data; 232 struct ip_vs_lblc_entry *en; 233 struct hlist_node *next; 234 int i; 235 236 spin_lock_bh(&svc->sched_lock); 237 tbl->dead = true; 238 for (i = 0; i < IP_VS_LBLC_TAB_SIZE; i++) { 239 hlist_for_each_entry_safe(en, next, &tbl->bucket[i], list) { 240 ip_vs_lblc_del(en); 241 atomic_dec(&tbl->entries); 242 } 243 } 244 spin_unlock_bh(&svc->sched_lock); 245} 246 247static int sysctl_lblc_expiration(struct ip_vs_service *svc) 248{ 249#ifdef CONFIG_SYSCTL 250 return svc->ipvs->sysctl_lblc_expiration; 251#else 252 return DEFAULT_EXPIRATION; 253#endif 254} 255 256static inline void ip_vs_lblc_full_check(struct ip_vs_service *svc) 257{ 258 struct ip_vs_lblc_table *tbl = svc->sched_data; 259 struct ip_vs_lblc_entry *en; 260 struct hlist_node *next; 261 unsigned long now = jiffies; 262 int i, j; 263 264 for (i = 0, j = tbl->rover; i < IP_VS_LBLC_TAB_SIZE; i++) { 265 j = (j + 1) & IP_VS_LBLC_TAB_MASK; 266 267 spin_lock(&svc->sched_lock); 268 hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) { 269 if (time_before(now, 270 en->lastuse + 271 sysctl_lblc_expiration(svc))) 272 continue; 273 274 ip_vs_lblc_del(en); 275 atomic_dec(&tbl->entries); 276 } 277 spin_unlock(&svc->sched_lock); 278 } 279 tbl->rover = j; 280} 281 282 283/* 284 * Periodical timer handler for IPVS lblc table 285 * It is used to collect stale entries when the number of entries 286 * exceeds the maximum size of the table. 287 * 288 * Fixme: we probably need more complicated algorithm to collect 289 * entries that have not been used for a long time even 290 * if the number of entries doesn't exceed the maximum size 291 * of the table. 292 * The full expiration check is for this purpose now. 293 */ 294static void ip_vs_lblc_check_expire(struct timer_list *t) 295{ 296 struct ip_vs_lblc_table *tbl = from_timer(tbl, t, periodic_timer); 297 struct ip_vs_service *svc = tbl->svc; 298 unsigned long now = jiffies; 299 int goal; 300 int i, j; 301 struct ip_vs_lblc_entry *en; 302 struct hlist_node *next; 303 304 if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) { 305 /* do full expiration check */ 306 ip_vs_lblc_full_check(svc); 307 tbl->counter = 1; 308 goto out; 309 } 310 311 if (atomic_read(&tbl->entries) <= tbl->max_size) { 312 tbl->counter++; 313 goto out; 314 } 315 316 goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3; 317 if (goal > tbl->max_size/2) 318 goal = tbl->max_size/2; 319 320 for (i = 0, j = tbl->rover; i < IP_VS_LBLC_TAB_SIZE; i++) { 321 j = (j + 1) & IP_VS_LBLC_TAB_MASK; 322 323 spin_lock(&svc->sched_lock); 324 hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) { 325 if (time_before(now, en->lastuse + ENTRY_TIMEOUT)) 326 continue; 327 328 ip_vs_lblc_del(en); 329 atomic_dec(&tbl->entries); 330 goal--; 331 } 332 spin_unlock(&svc->sched_lock); 333 if (goal <= 0) 334 break; 335 } 336 tbl->rover = j; 337 338 out: 339 mod_timer(&tbl->periodic_timer, jiffies + CHECK_EXPIRE_INTERVAL); 340} 341 342 343static int ip_vs_lblc_init_svc(struct ip_vs_service *svc) 344{ 345 int i; 346 struct ip_vs_lblc_table *tbl; 347 348 /* 349 * Allocate the ip_vs_lblc_table for this service 350 */ 351 tbl = kmalloc(sizeof(*tbl), GFP_KERNEL); 352 if (tbl == NULL) 353 return -ENOMEM; 354 355 svc->sched_data = tbl; 356 IP_VS_DBG(6, "LBLC hash table (memory=%zdbytes) allocated for " 357 "current service\n", sizeof(*tbl)); 358 359 /* 360 * Initialize the hash buckets 361 */ 362 for (i = 0; i < IP_VS_LBLC_TAB_SIZE; i++) { 363 INIT_HLIST_HEAD(&tbl->bucket[i]); 364 } 365 tbl->max_size = IP_VS_LBLC_TAB_SIZE*16; 366 tbl->rover = 0; 367 tbl->counter = 1; 368 tbl->dead = false; 369 tbl->svc = svc; 370 atomic_set(&tbl->entries, 0); 371 372 /* 373 * Hook periodic timer for garbage collection 374 */ 375 timer_setup(&tbl->periodic_timer, ip_vs_lblc_check_expire, 0); 376 mod_timer(&tbl->periodic_timer, jiffies + CHECK_EXPIRE_INTERVAL); 377 378 return 0; 379} 380 381 382static void ip_vs_lblc_done_svc(struct ip_vs_service *svc) 383{ 384 struct ip_vs_lblc_table *tbl = svc->sched_data; 385 386 /* remove periodic timer */ 387 timer_shutdown_sync(&tbl->periodic_timer); 388 389 /* got to clean up table entries here */ 390 ip_vs_lblc_flush(svc); 391 392 /* release the table itself */ 393 kfree_rcu(tbl, rcu_head); 394 IP_VS_DBG(6, "LBLC hash table (memory=%zdbytes) released\n", 395 sizeof(*tbl)); 396} 397 398 399static inline struct ip_vs_dest * 400__ip_vs_lblc_schedule(struct ip_vs_service *svc) 401{ 402 struct ip_vs_dest *dest, *least; 403 int loh, doh; 404 405 /* 406 * We use the following formula to estimate the load: 407 * (dest overhead) / dest->weight 408 * 409 * Remember -- no floats in kernel mode!!! 410 * The comparison of h1*w2 > h2*w1 is equivalent to that of 411 * h1/w1 > h2/w2 412 * if every weight is larger than zero. 413 * 414 * The server with weight=0 is quiesced and will not receive any 415 * new connection. 416 */ 417 list_for_each_entry_rcu(dest, &svc->destinations, n_list) { 418 if (dest->flags & IP_VS_DEST_F_OVERLOAD) 419 continue; 420 if (atomic_read(&dest->weight) > 0) { 421 least = dest; 422 loh = ip_vs_dest_conn_overhead(least); 423 goto nextstage; 424 } 425 } 426 return NULL; 427 428 /* 429 * Find the destination with the least load. 430 */ 431 nextstage: 432 list_for_each_entry_continue_rcu(dest, &svc->destinations, n_list) { 433 if (dest->flags & IP_VS_DEST_F_OVERLOAD) 434 continue; 435 436 doh = ip_vs_dest_conn_overhead(dest); 437 if ((__s64)loh * atomic_read(&dest->weight) > 438 (__s64)doh * atomic_read(&least->weight)) { 439 least = dest; 440 loh = doh; 441 } 442 } 443 444 IP_VS_DBG_BUF(6, "LBLC: server %s:%d " 445 "activeconns %d refcnt %d weight %d overhead %d\n", 446 IP_VS_DBG_ADDR(least->af, &least->addr), 447 ntohs(least->port), 448 atomic_read(&least->activeconns), 449 refcount_read(&least->refcnt), 450 atomic_read(&least->weight), loh); 451 452 return least; 453} 454 455 456/* 457 * If this destination server is overloaded and there is a less loaded 458 * server, then return true. 459 */ 460static inline int 461is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc) 462{ 463 if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) { 464 struct ip_vs_dest *d; 465 466 list_for_each_entry_rcu(d, &svc->destinations, n_list) { 467 if (atomic_read(&d->activeconns)*2 468 < atomic_read(&d->weight)) { 469 return 1; 470 } 471 } 472 } 473 return 0; 474} 475 476 477/* 478 * Locality-Based (weighted) Least-Connection scheduling 479 */ 480static struct ip_vs_dest * 481ip_vs_lblc_schedule(struct ip_vs_service *svc, const struct sk_buff *skb, 482 struct ip_vs_iphdr *iph) 483{ 484 struct ip_vs_lblc_table *tbl = svc->sched_data; 485 struct ip_vs_dest *dest = NULL; 486 struct ip_vs_lblc_entry *en; 487 488 IP_VS_DBG(6, "%s(): Scheduling...\n", __func__); 489 490 /* First look in our cache */ 491 en = ip_vs_lblc_get(svc->af, tbl, &iph->daddr); 492 if (en) { 493 /* We only hold a read lock, but this is atomic */ 494 en->lastuse = jiffies; 495 496 /* 497 * If the destination is not available, i.e. it's in the trash, 498 * we must ignore it, as it may be removed from under our feet, 499 * if someone drops our reference count. Our caller only makes 500 * sure that destinations, that are not in the trash, are not 501 * moved to the trash, while we are scheduling. But anyone can 502 * free up entries from the trash at any time. 503 */ 504 505 dest = en->dest; 506 if ((dest->flags & IP_VS_DEST_F_AVAILABLE) && 507 atomic_read(&dest->weight) > 0 && !is_overloaded(dest, svc)) 508 goto out; 509 } 510 511 /* No cache entry or it is invalid, time to schedule */ 512 dest = __ip_vs_lblc_schedule(svc); 513 if (!dest) { 514 ip_vs_scheduler_err(svc, "no destination available"); 515 return NULL; 516 } 517 518 /* If we fail to create a cache entry, we'll just use the valid dest */ 519 spin_lock_bh(&svc->sched_lock); 520 if (!tbl->dead) 521 ip_vs_lblc_new(tbl, &iph->daddr, svc->af, dest); 522 spin_unlock_bh(&svc->sched_lock); 523 524out: 525 IP_VS_DBG_BUF(6, "LBLC: destination IP address %s --> server %s:%d\n", 526 IP_VS_DBG_ADDR(svc->af, &iph->daddr), 527 IP_VS_DBG_ADDR(dest->af, &dest->addr), ntohs(dest->port)); 528 529 return dest; 530} 531 532 533/* 534 * IPVS LBLC Scheduler structure 535 */ 536static struct ip_vs_scheduler ip_vs_lblc_scheduler = { 537 .name = "lblc", 538 .refcnt = ATOMIC_INIT(0), 539 .module = THIS_MODULE, 540 .n_list = LIST_HEAD_INIT(ip_vs_lblc_scheduler.n_list), 541 .init_service = ip_vs_lblc_init_svc, 542 .done_service = ip_vs_lblc_done_svc, 543 .schedule = ip_vs_lblc_schedule, 544}; 545 546/* 547 * per netns init. 548 */ 549#ifdef CONFIG_SYSCTL 550static int __net_init __ip_vs_lblc_init(struct net *net) 551{ 552 struct netns_ipvs *ipvs = net_ipvs(net); 553 size_t vars_table_size = ARRAY_SIZE(vs_vars_table); 554 555 if (!ipvs) 556 return -ENOENT; 557 558 if (!net_eq(net, &init_net)) { 559 ipvs->lblc_ctl_table = kmemdup(vs_vars_table, 560 sizeof(vs_vars_table), 561 GFP_KERNEL); 562 if (ipvs->lblc_ctl_table == NULL) 563 return -ENOMEM; 564 565 /* Don't export sysctls to unprivileged users */ 566 if (net->user_ns != &init_user_ns) { 567 ipvs->lblc_ctl_table[0].procname = NULL; 568 vars_table_size = 0; 569 } 570 571 } else 572 ipvs->lblc_ctl_table = vs_vars_table; 573 ipvs->sysctl_lblc_expiration = DEFAULT_EXPIRATION; 574 ipvs->lblc_ctl_table[0].data = &ipvs->sysctl_lblc_expiration; 575 576 ipvs->lblc_ctl_header = register_net_sysctl_sz(net, "net/ipv4/vs", 577 ipvs->lblc_ctl_table, 578 vars_table_size); 579 if (!ipvs->lblc_ctl_header) { 580 if (!net_eq(net, &init_net)) 581 kfree(ipvs->lblc_ctl_table); 582 return -ENOMEM; 583 } 584 585 return 0; 586} 587 588static void __net_exit __ip_vs_lblc_exit(struct net *net) 589{ 590 struct netns_ipvs *ipvs = net_ipvs(net); 591 592 unregister_net_sysctl_table(ipvs->lblc_ctl_header); 593 594 if (!net_eq(net, &init_net)) 595 kfree(ipvs->lblc_ctl_table); 596} 597 598#else 599 600static int __net_init __ip_vs_lblc_init(struct net *net) { return 0; } 601static void __net_exit __ip_vs_lblc_exit(struct net *net) { } 602 603#endif 604 605static struct pernet_operations ip_vs_lblc_ops = { 606 .init = __ip_vs_lblc_init, 607 .exit = __ip_vs_lblc_exit, 608}; 609 610static int __init ip_vs_lblc_init(void) 611{ 612 int ret; 613 614 ret = register_pernet_subsys(&ip_vs_lblc_ops); 615 if (ret) 616 return ret; 617 618 ret = register_ip_vs_scheduler(&ip_vs_lblc_scheduler); 619 if (ret) 620 unregister_pernet_subsys(&ip_vs_lblc_ops); 621 return ret; 622} 623 624static void __exit ip_vs_lblc_cleanup(void) 625{ 626 unregister_ip_vs_scheduler(&ip_vs_lblc_scheduler); 627 unregister_pernet_subsys(&ip_vs_lblc_ops); 628 rcu_barrier(); 629} 630 631 632module_init(ip_vs_lblc_init); 633module_exit(ip_vs_lblc_cleanup); 634MODULE_LICENSE("GPL"); 635MODULE_DESCRIPTION("ipvs locality-based least-connection scheduler"); 636