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 struct rcu_head rcu_head;
94 };
95
96 struct ip_vs_dest_set {
97 atomic_t size; /* set size */
98 unsigned long lastmod; /* last modified time */
99 struct list_head list; /* destination list */
100 };
101
102
ip_vs_dest_set_insert(struct ip_vs_dest_set * set,struct ip_vs_dest * dest,bool check)103 static void ip_vs_dest_set_insert(struct ip_vs_dest_set *set,
104 struct ip_vs_dest *dest, bool check)
105 {
106 struct ip_vs_dest_set_elem *e;
107
108 if (check) {
109 list_for_each_entry(e, &set->list, list) {
110 if (e->dest == dest)
111 return;
112 }
113 }
114
115 e = kmalloc(sizeof(*e), GFP_ATOMIC);
116 if (e == NULL)
117 return;
118
119 ip_vs_dest_hold(dest);
120 e->dest = dest;
121
122 list_add_rcu(&e->list, &set->list);
123 atomic_inc(&set->size);
124
125 set->lastmod = jiffies;
126 }
127
ip_vs_lblcr_elem_rcu_free(struct rcu_head * head)128 static void ip_vs_lblcr_elem_rcu_free(struct rcu_head *head)
129 {
130 struct ip_vs_dest_set_elem *e;
131
132 e = container_of(head, struct ip_vs_dest_set_elem, rcu_head);
133 ip_vs_dest_put_and_free(e->dest);
134 kfree(e);
135 }
136
137 static void
ip_vs_dest_set_erase(struct ip_vs_dest_set * set,struct ip_vs_dest * dest)138 ip_vs_dest_set_erase(struct ip_vs_dest_set *set, struct ip_vs_dest *dest)
139 {
140 struct ip_vs_dest_set_elem *e;
141
142 list_for_each_entry(e, &set->list, list) {
143 if (e->dest == dest) {
144 /* HIT */
145 atomic_dec(&set->size);
146 set->lastmod = jiffies;
147 list_del_rcu(&e->list);
148 call_rcu(&e->rcu_head, ip_vs_lblcr_elem_rcu_free);
149 break;
150 }
151 }
152 }
153
ip_vs_dest_set_eraseall(struct ip_vs_dest_set * set)154 static void ip_vs_dest_set_eraseall(struct ip_vs_dest_set *set)
155 {
156 struct ip_vs_dest_set_elem *e, *ep;
157
158 list_for_each_entry_safe(e, ep, &set->list, list) {
159 list_del_rcu(&e->list);
160 call_rcu(&e->rcu_head, ip_vs_lblcr_elem_rcu_free);
161 }
162 }
163
164 /* get weighted least-connection node in the destination set */
ip_vs_dest_set_min(struct ip_vs_dest_set * set)165 static inline struct ip_vs_dest *ip_vs_dest_set_min(struct ip_vs_dest_set *set)
166 {
167 register struct ip_vs_dest_set_elem *e;
168 struct ip_vs_dest *dest, *least;
169 int loh, doh;
170
171 /* select the first destination server, whose weight > 0 */
172 list_for_each_entry_rcu(e, &set->list, list) {
173 least = e->dest;
174 if (least->flags & IP_VS_DEST_F_OVERLOAD)
175 continue;
176
177 if ((atomic_read(&least->weight) > 0)
178 && (least->flags & IP_VS_DEST_F_AVAILABLE)) {
179 loh = ip_vs_dest_conn_overhead(least);
180 goto nextstage;
181 }
182 }
183 return NULL;
184
185 /* find the destination with the weighted least load */
186 nextstage:
187 list_for_each_entry_continue_rcu(e, &set->list, list) {
188 dest = e->dest;
189 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
190 continue;
191
192 doh = ip_vs_dest_conn_overhead(dest);
193 if (((__s64)loh * atomic_read(&dest->weight) >
194 (__s64)doh * atomic_read(&least->weight))
195 && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
196 least = dest;
197 loh = doh;
198 }
199 }
200
201 IP_VS_DBG_BUF(6, "%s(): server %s:%d "
202 "activeconns %d refcnt %d weight %d overhead %d\n",
203 __func__,
204 IP_VS_DBG_ADDR(least->af, &least->addr),
205 ntohs(least->port),
206 atomic_read(&least->activeconns),
207 atomic_read(&least->refcnt),
208 atomic_read(&least->weight), loh);
209 return least;
210 }
211
212
213 /* get weighted most-connection node in the destination set */
ip_vs_dest_set_max(struct ip_vs_dest_set * set)214 static inline struct ip_vs_dest *ip_vs_dest_set_max(struct ip_vs_dest_set *set)
215 {
216 register struct ip_vs_dest_set_elem *e;
217 struct ip_vs_dest *dest, *most;
218 int moh, doh;
219
220 if (set == NULL)
221 return NULL;
222
223 /* select the first destination server, whose weight > 0 */
224 list_for_each_entry(e, &set->list, list) {
225 most = e->dest;
226 if (atomic_read(&most->weight) > 0) {
227 moh = ip_vs_dest_conn_overhead(most);
228 goto nextstage;
229 }
230 }
231 return NULL;
232
233 /* find the destination with the weighted most load */
234 nextstage:
235 list_for_each_entry_continue(e, &set->list, list) {
236 dest = e->dest;
237 doh = ip_vs_dest_conn_overhead(dest);
238 /* moh/mw < doh/dw ==> moh*dw < doh*mw, where mw,dw>0 */
239 if (((__s64)moh * atomic_read(&dest->weight) <
240 (__s64)doh * atomic_read(&most->weight))
241 && (atomic_read(&dest->weight) > 0)) {
242 most = dest;
243 moh = doh;
244 }
245 }
246
247 IP_VS_DBG_BUF(6, "%s(): server %s:%d "
248 "activeconns %d refcnt %d weight %d overhead %d\n",
249 __func__,
250 IP_VS_DBG_ADDR(most->af, &most->addr), ntohs(most->port),
251 atomic_read(&most->activeconns),
252 atomic_read(&most->refcnt),
253 atomic_read(&most->weight), moh);
254 return most;
255 }
256
257
258 /*
259 * IPVS lblcr entry represents an association between destination
260 * IP address and its destination server set
261 */
262 struct ip_vs_lblcr_entry {
263 struct hlist_node list;
264 int af; /* address family */
265 union nf_inet_addr addr; /* destination IP address */
266 struct ip_vs_dest_set set; /* destination server set */
267 unsigned long lastuse; /* last used time */
268 struct rcu_head rcu_head;
269 };
270
271
272 /*
273 * IPVS lblcr hash table
274 */
275 struct ip_vs_lblcr_table {
276 struct rcu_head rcu_head;
277 struct hlist_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 bool dead;
284 };
285
286
287 #ifdef CONFIG_SYSCTL
288 /*
289 * IPVS LBLCR sysctl table
290 */
291
292 static struct ctl_table vs_vars_table[] = {
293 {
294 .procname = "lblcr_expiration",
295 .data = NULL,
296 .maxlen = sizeof(int),
297 .mode = 0644,
298 .proc_handler = proc_dointvec_jiffies,
299 },
300 { }
301 };
302 #endif
303
ip_vs_lblcr_free(struct ip_vs_lblcr_entry * en)304 static inline void ip_vs_lblcr_free(struct ip_vs_lblcr_entry *en)
305 {
306 hlist_del_rcu(&en->list);
307 ip_vs_dest_set_eraseall(&en->set);
308 kfree_rcu(en, rcu_head);
309 }
310
311
312 /*
313 * Returns hash value for IPVS LBLCR entry
314 */
315 static inline unsigned int
ip_vs_lblcr_hashkey(int af,const union nf_inet_addr * addr)316 ip_vs_lblcr_hashkey(int af, const union nf_inet_addr *addr)
317 {
318 __be32 addr_fold = addr->ip;
319
320 #ifdef CONFIG_IP_VS_IPV6
321 if (af == AF_INET6)
322 addr_fold = addr->ip6[0]^addr->ip6[1]^
323 addr->ip6[2]^addr->ip6[3];
324 #endif
325 return (ntohl(addr_fold)*2654435761UL) & IP_VS_LBLCR_TAB_MASK;
326 }
327
328
329 /*
330 * Hash an entry in the ip_vs_lblcr_table.
331 * returns bool success.
332 */
333 static void
ip_vs_lblcr_hash(struct ip_vs_lblcr_table * tbl,struct ip_vs_lblcr_entry * en)334 ip_vs_lblcr_hash(struct ip_vs_lblcr_table *tbl, struct ip_vs_lblcr_entry *en)
335 {
336 unsigned int hash = ip_vs_lblcr_hashkey(en->af, &en->addr);
337
338 hlist_add_head_rcu(&en->list, &tbl->bucket[hash]);
339 atomic_inc(&tbl->entries);
340 }
341
342
343 /* Get ip_vs_lblcr_entry associated with supplied parameters. */
344 static inline struct ip_vs_lblcr_entry *
ip_vs_lblcr_get(int af,struct ip_vs_lblcr_table * tbl,const union nf_inet_addr * addr)345 ip_vs_lblcr_get(int af, struct ip_vs_lblcr_table *tbl,
346 const union nf_inet_addr *addr)
347 {
348 unsigned int hash = ip_vs_lblcr_hashkey(af, addr);
349 struct ip_vs_lblcr_entry *en;
350
351 hlist_for_each_entry_rcu(en, &tbl->bucket[hash], list)
352 if (ip_vs_addr_equal(af, &en->addr, addr))
353 return en;
354
355 return NULL;
356 }
357
358
359 /*
360 * Create or update an ip_vs_lblcr_entry, which is a mapping of a destination
361 * IP address to a server. Called under spin lock.
362 */
363 static inline struct ip_vs_lblcr_entry *
ip_vs_lblcr_new(struct ip_vs_lblcr_table * tbl,const union nf_inet_addr * daddr,u16 af,struct ip_vs_dest * dest)364 ip_vs_lblcr_new(struct ip_vs_lblcr_table *tbl, const union nf_inet_addr *daddr,
365 u16 af, struct ip_vs_dest *dest)
366 {
367 struct ip_vs_lblcr_entry *en;
368
369 en = ip_vs_lblcr_get(af, tbl, daddr);
370 if (!en) {
371 en = kmalloc(sizeof(*en), GFP_ATOMIC);
372 if (!en)
373 return NULL;
374
375 en->af = af;
376 ip_vs_addr_copy(af, &en->addr, daddr);
377 en->lastuse = jiffies;
378
379 /* initialize its dest set */
380 atomic_set(&(en->set.size), 0);
381 INIT_LIST_HEAD(&en->set.list);
382
383 ip_vs_dest_set_insert(&en->set, dest, false);
384
385 ip_vs_lblcr_hash(tbl, en);
386 return en;
387 }
388
389 ip_vs_dest_set_insert(&en->set, dest, true);
390
391 return en;
392 }
393
394
395 /*
396 * Flush all the entries of the specified table.
397 */
ip_vs_lblcr_flush(struct ip_vs_service * svc)398 static void ip_vs_lblcr_flush(struct ip_vs_service *svc)
399 {
400 struct ip_vs_lblcr_table *tbl = svc->sched_data;
401 int i;
402 struct ip_vs_lblcr_entry *en;
403 struct hlist_node *next;
404
405 spin_lock_bh(&svc->sched_lock);
406 tbl->dead = 1;
407 for (i = 0; i < IP_VS_LBLCR_TAB_SIZE; i++) {
408 hlist_for_each_entry_safe(en, next, &tbl->bucket[i], list) {
409 ip_vs_lblcr_free(en);
410 }
411 }
412 spin_unlock_bh(&svc->sched_lock);
413 }
414
sysctl_lblcr_expiration(struct ip_vs_service * svc)415 static int sysctl_lblcr_expiration(struct ip_vs_service *svc)
416 {
417 #ifdef CONFIG_SYSCTL
418 struct netns_ipvs *ipvs = net_ipvs(svc->net);
419 return ipvs->sysctl_lblcr_expiration;
420 #else
421 return DEFAULT_EXPIRATION;
422 #endif
423 }
424
ip_vs_lblcr_full_check(struct ip_vs_service * svc)425 static inline void ip_vs_lblcr_full_check(struct ip_vs_service *svc)
426 {
427 struct ip_vs_lblcr_table *tbl = svc->sched_data;
428 unsigned long now = jiffies;
429 int i, j;
430 struct ip_vs_lblcr_entry *en;
431 struct hlist_node *next;
432
433 for (i = 0, j = tbl->rover; i < IP_VS_LBLCR_TAB_SIZE; i++) {
434 j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
435
436 spin_lock(&svc->sched_lock);
437 hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) {
438 if (time_after(en->lastuse +
439 sysctl_lblcr_expiration(svc), now))
440 continue;
441
442 ip_vs_lblcr_free(en);
443 atomic_dec(&tbl->entries);
444 }
445 spin_unlock(&svc->sched_lock);
446 }
447 tbl->rover = j;
448 }
449
450
451 /*
452 * Periodical timer handler for IPVS lblcr table
453 * It is used to collect stale entries when the number of entries
454 * exceeds the maximum size of the table.
455 *
456 * Fixme: we probably need more complicated algorithm to collect
457 * entries that have not been used for a long time even
458 * if the number of entries doesn't exceed the maximum size
459 * of the table.
460 * The full expiration check is for this purpose now.
461 */
ip_vs_lblcr_check_expire(unsigned long data)462 static void ip_vs_lblcr_check_expire(unsigned long data)
463 {
464 struct ip_vs_service *svc = (struct ip_vs_service *) data;
465 struct ip_vs_lblcr_table *tbl = svc->sched_data;
466 unsigned long now = jiffies;
467 int goal;
468 int i, j;
469 struct ip_vs_lblcr_entry *en;
470 struct hlist_node *next;
471
472 if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) {
473 /* do full expiration check */
474 ip_vs_lblcr_full_check(svc);
475 tbl->counter = 1;
476 goto out;
477 }
478
479 if (atomic_read(&tbl->entries) <= tbl->max_size) {
480 tbl->counter++;
481 goto out;
482 }
483
484 goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3;
485 if (goal > tbl->max_size/2)
486 goal = tbl->max_size/2;
487
488 for (i = 0, j = tbl->rover; i < IP_VS_LBLCR_TAB_SIZE; i++) {
489 j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
490
491 spin_lock(&svc->sched_lock);
492 hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) {
493 if (time_before(now, en->lastuse+ENTRY_TIMEOUT))
494 continue;
495
496 ip_vs_lblcr_free(en);
497 atomic_dec(&tbl->entries);
498 goal--;
499 }
500 spin_unlock(&svc->sched_lock);
501 if (goal <= 0)
502 break;
503 }
504 tbl->rover = j;
505
506 out:
507 mod_timer(&tbl->periodic_timer, jiffies+CHECK_EXPIRE_INTERVAL);
508 }
509
ip_vs_lblcr_init_svc(struct ip_vs_service * svc)510 static int ip_vs_lblcr_init_svc(struct ip_vs_service *svc)
511 {
512 int i;
513 struct ip_vs_lblcr_table *tbl;
514
515 /*
516 * Allocate the ip_vs_lblcr_table for this service
517 */
518 tbl = kmalloc(sizeof(*tbl), GFP_KERNEL);
519 if (tbl == NULL)
520 return -ENOMEM;
521
522 svc->sched_data = tbl;
523 IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) allocated for "
524 "current service\n", sizeof(*tbl));
525
526 /*
527 * Initialize the hash buckets
528 */
529 for (i = 0; i < IP_VS_LBLCR_TAB_SIZE; i++) {
530 INIT_HLIST_HEAD(&tbl->bucket[i]);
531 }
532 tbl->max_size = IP_VS_LBLCR_TAB_SIZE*16;
533 tbl->rover = 0;
534 tbl->counter = 1;
535 tbl->dead = 0;
536
537 /*
538 * Hook periodic timer for garbage collection
539 */
540 setup_timer(&tbl->periodic_timer, ip_vs_lblcr_check_expire,
541 (unsigned long)svc);
542 mod_timer(&tbl->periodic_timer, jiffies + CHECK_EXPIRE_INTERVAL);
543
544 return 0;
545 }
546
547
ip_vs_lblcr_done_svc(struct ip_vs_service * svc)548 static void ip_vs_lblcr_done_svc(struct ip_vs_service *svc)
549 {
550 struct ip_vs_lblcr_table *tbl = svc->sched_data;
551
552 /* remove periodic timer */
553 del_timer_sync(&tbl->periodic_timer);
554
555 /* got to clean up table entries here */
556 ip_vs_lblcr_flush(svc);
557
558 /* release the table itself */
559 kfree_rcu(tbl, rcu_head);
560 IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) released\n",
561 sizeof(*tbl));
562 }
563
564
565 static inline struct ip_vs_dest *
__ip_vs_lblcr_schedule(struct ip_vs_service * svc)566 __ip_vs_lblcr_schedule(struct ip_vs_service *svc)
567 {
568 struct ip_vs_dest *dest, *least;
569 int loh, doh;
570
571 /*
572 * We use the following formula to estimate the load:
573 * (dest overhead) / dest->weight
574 *
575 * Remember -- no floats in kernel mode!!!
576 * The comparison of h1*w2 > h2*w1 is equivalent to that of
577 * h1/w1 > h2/w2
578 * if every weight is larger than zero.
579 *
580 * The server with weight=0 is quiesced and will not receive any
581 * new connection.
582 */
583 list_for_each_entry_rcu(dest, &svc->destinations, n_list) {
584 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
585 continue;
586
587 if (atomic_read(&dest->weight) > 0) {
588 least = dest;
589 loh = ip_vs_dest_conn_overhead(least);
590 goto nextstage;
591 }
592 }
593 return NULL;
594
595 /*
596 * Find the destination with the least load.
597 */
598 nextstage:
599 list_for_each_entry_continue_rcu(dest, &svc->destinations, n_list) {
600 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
601 continue;
602
603 doh = ip_vs_dest_conn_overhead(dest);
604 if ((__s64)loh * atomic_read(&dest->weight) >
605 (__s64)doh * atomic_read(&least->weight)) {
606 least = dest;
607 loh = doh;
608 }
609 }
610
611 IP_VS_DBG_BUF(6, "LBLCR: server %s:%d "
612 "activeconns %d refcnt %d weight %d overhead %d\n",
613 IP_VS_DBG_ADDR(least->af, &least->addr),
614 ntohs(least->port),
615 atomic_read(&least->activeconns),
616 atomic_read(&least->refcnt),
617 atomic_read(&least->weight), loh);
618
619 return least;
620 }
621
622
623 /*
624 * If this destination server is overloaded and there is a less loaded
625 * server, then return true.
626 */
627 static inline int
is_overloaded(struct ip_vs_dest * dest,struct ip_vs_service * svc)628 is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc)
629 {
630 if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) {
631 struct ip_vs_dest *d;
632
633 list_for_each_entry_rcu(d, &svc->destinations, n_list) {
634 if (atomic_read(&d->activeconns)*2
635 < atomic_read(&d->weight)) {
636 return 1;
637 }
638 }
639 }
640 return 0;
641 }
642
643
644 /*
645 * Locality-Based (weighted) Least-Connection scheduling
646 */
647 static struct ip_vs_dest *
ip_vs_lblcr_schedule(struct ip_vs_service * svc,const struct sk_buff * skb,struct ip_vs_iphdr * iph)648 ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb,
649 struct ip_vs_iphdr *iph)
650 {
651 struct ip_vs_lblcr_table *tbl = svc->sched_data;
652 struct ip_vs_dest *dest;
653 struct ip_vs_lblcr_entry *en;
654
655 IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
656
657 /* First look in our cache */
658 en = ip_vs_lblcr_get(svc->af, tbl, &iph->daddr);
659 if (en) {
660 en->lastuse = jiffies;
661
662 /* Get the least loaded destination */
663 dest = ip_vs_dest_set_min(&en->set);
664
665 /* More than one destination + enough time passed by, cleanup */
666 if (atomic_read(&en->set.size) > 1 &&
667 time_after(jiffies, en->set.lastmod +
668 sysctl_lblcr_expiration(svc))) {
669 spin_lock_bh(&svc->sched_lock);
670 if (atomic_read(&en->set.size) > 1) {
671 struct ip_vs_dest *m;
672
673 m = ip_vs_dest_set_max(&en->set);
674 if (m)
675 ip_vs_dest_set_erase(&en->set, m);
676 }
677 spin_unlock_bh(&svc->sched_lock);
678 }
679
680 /* If the destination is not overloaded, use it */
681 if (dest && !is_overloaded(dest, svc))
682 goto out;
683
684 /* The cache entry is invalid, time to schedule */
685 dest = __ip_vs_lblcr_schedule(svc);
686 if (!dest) {
687 ip_vs_scheduler_err(svc, "no destination available");
688 return NULL;
689 }
690
691 /* Update our cache entry */
692 spin_lock_bh(&svc->sched_lock);
693 if (!tbl->dead)
694 ip_vs_dest_set_insert(&en->set, dest, true);
695 spin_unlock_bh(&svc->sched_lock);
696 goto out;
697 }
698
699 /* No cache entry, time to schedule */
700 dest = __ip_vs_lblcr_schedule(svc);
701 if (!dest) {
702 IP_VS_DBG(1, "no destination available\n");
703 return NULL;
704 }
705
706 /* If we fail to create a cache entry, we'll just use the valid dest */
707 spin_lock_bh(&svc->sched_lock);
708 if (!tbl->dead)
709 ip_vs_lblcr_new(tbl, &iph->daddr, svc->af, dest);
710 spin_unlock_bh(&svc->sched_lock);
711
712 out:
713 IP_VS_DBG_BUF(6, "LBLCR: destination IP address %s --> server %s:%d\n",
714 IP_VS_DBG_ADDR(svc->af, &iph->daddr),
715 IP_VS_DBG_ADDR(dest->af, &dest->addr), ntohs(dest->port));
716
717 return dest;
718 }
719
720
721 /*
722 * IPVS LBLCR Scheduler structure
723 */
724 static struct ip_vs_scheduler ip_vs_lblcr_scheduler =
725 {
726 .name = "lblcr",
727 .refcnt = ATOMIC_INIT(0),
728 .module = THIS_MODULE,
729 .n_list = LIST_HEAD_INIT(ip_vs_lblcr_scheduler.n_list),
730 .init_service = ip_vs_lblcr_init_svc,
731 .done_service = ip_vs_lblcr_done_svc,
732 .schedule = ip_vs_lblcr_schedule,
733 };
734
735 /*
736 * per netns init.
737 */
738 #ifdef CONFIG_SYSCTL
__ip_vs_lblcr_init(struct net * net)739 static int __net_init __ip_vs_lblcr_init(struct net *net)
740 {
741 struct netns_ipvs *ipvs = net_ipvs(net);
742
743 if (!ipvs)
744 return -ENOENT;
745
746 if (!net_eq(net, &init_net)) {
747 ipvs->lblcr_ctl_table = kmemdup(vs_vars_table,
748 sizeof(vs_vars_table),
749 GFP_KERNEL);
750 if (ipvs->lblcr_ctl_table == NULL)
751 return -ENOMEM;
752
753 /* Don't export sysctls to unprivileged users */
754 if (net->user_ns != &init_user_ns)
755 ipvs->lblcr_ctl_table[0].procname = NULL;
756 } else
757 ipvs->lblcr_ctl_table = vs_vars_table;
758 ipvs->sysctl_lblcr_expiration = DEFAULT_EXPIRATION;
759 ipvs->lblcr_ctl_table[0].data = &ipvs->sysctl_lblcr_expiration;
760
761 ipvs->lblcr_ctl_header =
762 register_net_sysctl(net, "net/ipv4/vs", ipvs->lblcr_ctl_table);
763 if (!ipvs->lblcr_ctl_header) {
764 if (!net_eq(net, &init_net))
765 kfree(ipvs->lblcr_ctl_table);
766 return -ENOMEM;
767 }
768
769 return 0;
770 }
771
__ip_vs_lblcr_exit(struct net * net)772 static void __net_exit __ip_vs_lblcr_exit(struct net *net)
773 {
774 struct netns_ipvs *ipvs = net_ipvs(net);
775
776 unregister_net_sysctl_table(ipvs->lblcr_ctl_header);
777
778 if (!net_eq(net, &init_net))
779 kfree(ipvs->lblcr_ctl_table);
780 }
781
782 #else
783
__ip_vs_lblcr_init(struct net * net)784 static int __net_init __ip_vs_lblcr_init(struct net *net) { return 0; }
__ip_vs_lblcr_exit(struct net * net)785 static void __net_exit __ip_vs_lblcr_exit(struct net *net) { }
786
787 #endif
788
789 static struct pernet_operations ip_vs_lblcr_ops = {
790 .init = __ip_vs_lblcr_init,
791 .exit = __ip_vs_lblcr_exit,
792 };
793
ip_vs_lblcr_init(void)794 static int __init ip_vs_lblcr_init(void)
795 {
796 int ret;
797
798 ret = register_pernet_subsys(&ip_vs_lblcr_ops);
799 if (ret)
800 return ret;
801
802 ret = register_ip_vs_scheduler(&ip_vs_lblcr_scheduler);
803 if (ret)
804 unregister_pernet_subsys(&ip_vs_lblcr_ops);
805 return ret;
806 }
807
ip_vs_lblcr_cleanup(void)808 static void __exit ip_vs_lblcr_cleanup(void)
809 {
810 unregister_ip_vs_scheduler(&ip_vs_lblcr_scheduler);
811 unregister_pernet_subsys(&ip_vs_lblcr_ops);
812 rcu_barrier();
813 }
814
815
816 module_init(ip_vs_lblcr_init);
817 module_exit(ip_vs_lblcr_cleanup);
818 MODULE_LICENSE("GPL");
819