ip_vs_lblcr.c

  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/*
  3 * IPVS:        Locality-Based Least-Connection with Replication scheduler
  4 *
  5 * Authors:     Wensong Zhang <wensong@gnuchina.org>
  6 *
  7 * Changes:
  8 *     Julian Anastasov        :    Added the missing (dest->weight>0)
  9 *                                  condition in the ip_vs_dest_set_max.
 10 */
 11
 12/*
 13 * The lblc/r algorithm is as follows (pseudo code):
 14 *
 15 *       if serverSet[dest_ip] is null then
 16 *               n, serverSet[dest_ip] <- {weighted least-conn node};
 17 *       else
 18 *               n <- {least-conn (alive) node in serverSet[dest_ip]};
 19 *               if (n is null) OR
 20 *                  (n.conns>n.weight AND
 21 *                   there is a node m with m.conns<m.weight/2) then
 22 *                   n <- {weighted least-conn node};
 23 *                   add n to serverSet[dest_ip];
 24 *               if |serverSet[dest_ip]| > 1 AND
 25 *                   now - serverSet[dest_ip].lastMod > T then
 26 *                   m <- {most conn node in serverSet[dest_ip]};
 27 *                   remove m from serverSet[dest_ip];
 28 *       if serverSet[dest_ip] changed then
 29 *               serverSet[dest_ip].lastMod <- now;
 30 *
 31 *       return n;
 32 *
 33 */
 34
 35#define pr_fmt(fmt) "IPVS: " fmt
 36
 37#include <linux/ip.h>
 38#include <linux/module.h>
 39#include <linux/kernel.h>
 40#include <linux/skbuff.h>
 41#include <linux/jiffies.h>
 42#include <linux/list.h>
 43#include <linux/slab.h>
 44#include <linux/hash.h>
 45
 46/* for sysctl */
 47#include <linux/fs.h>
 48#include <linux/sysctl.h>
 49#include <net/net_namespace.h>
 50
 51#include <net/ip_vs.h>
 52
 53
 54/*
 55 *    It is for garbage collection of stale IPVS lblcr entries,
 56 *    when the table is full.
 57 */
 58#define CHECK_EXPIRE_INTERVAL   (60*HZ)
 59#define ENTRY_TIMEOUT           (6*60*HZ)
 60
 61#define DEFAULT_EXPIRATION	(24*60*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
 71/*
 72 *     for IPVS lblcr entry hash table
 73 */
 74#ifndef CONFIG_IP_VS_LBLCR_TAB_BITS
 75#define CONFIG_IP_VS_LBLCR_TAB_BITS      10
 76#endif
 77#define IP_VS_LBLCR_TAB_BITS     CONFIG_IP_VS_LBLCR_TAB_BITS
 78#define IP_VS_LBLCR_TAB_SIZE     (1 << IP_VS_LBLCR_TAB_BITS)
 79#define IP_VS_LBLCR_TAB_MASK     (IP_VS_LBLCR_TAB_SIZE - 1)
 80
 81
 82/*
 83 *      IPVS destination set structure and operations
 84 */
 85struct ip_vs_dest_set_elem {
 86	struct list_head	list;          /* list link */
 87	struct ip_vs_dest	*dest;		/* destination server */
 88	struct rcu_head		rcu_head;
 89};
 90
 91struct ip_vs_dest_set {
 92	atomic_t                size;           /* set size */
 93	unsigned long           lastmod;        /* last modified time */
 94	struct list_head	list;           /* destination list */
 95};
 96
 97
 98static void ip_vs_dest_set_insert(struct ip_vs_dest_set *set,
 99				  struct ip_vs_dest *dest, bool check)
100{
101	struct ip_vs_dest_set_elem *e;
102
103	if (check) {
104		list_for_each_entry(e, &set->list, list) {
105			if (e->dest == dest)
106				return;
107		}
108	}
109
110	e = kmalloc(sizeof(*e), GFP_ATOMIC);
111	if (e == NULL)
112		return;
113
114	ip_vs_dest_hold(dest);
115	e->dest = dest;
116
117	list_add_rcu(&e->list, &set->list);
118	atomic_inc(&set->size);
119
120	set->lastmod = jiffies;
121}
122
123static void ip_vs_lblcr_elem_rcu_free(struct rcu_head *head)
124{
125	struct ip_vs_dest_set_elem *e;
126
127	e = container_of(head, struct ip_vs_dest_set_elem, rcu_head);
128	ip_vs_dest_put_and_free(e->dest);
129	kfree(e);
130}
131
132static void
133ip_vs_dest_set_erase(struct ip_vs_dest_set *set, struct ip_vs_dest *dest)
134{
135	struct ip_vs_dest_set_elem *e;
136
137	list_for_each_entry(e, &set->list, list) {
138		if (e->dest == dest) {
139			/* HIT */
140			atomic_dec(&set->size);
141			set->lastmod = jiffies;
142			list_del_rcu(&e->list);
143			call_rcu(&e->rcu_head, ip_vs_lblcr_elem_rcu_free);
144			break;
145		}
146	}
147}
148
149static void ip_vs_dest_set_eraseall(struct ip_vs_dest_set *set)
150{
151	struct ip_vs_dest_set_elem *e, *ep;
152
153	list_for_each_entry_safe(e, ep, &set->list, list) {
154		list_del_rcu(&e->list);
155		call_rcu(&e->rcu_head, ip_vs_lblcr_elem_rcu_free);
156	}
157}
158
159/* get weighted least-connection node in the destination set */
160static inline struct ip_vs_dest *ip_vs_dest_set_min(struct ip_vs_dest_set *set)
161{
162	struct ip_vs_dest_set_elem *e;
163	struct ip_vs_dest *dest, *least;
164	int loh, doh;
165
166	/* select the first destination server, whose weight > 0 */
167	list_for_each_entry_rcu(e, &set->list, list) {
168		least = e->dest;
169		if (least->flags & IP_VS_DEST_F_OVERLOAD)
170			continue;
171
172		if ((atomic_read(&least->weight) > 0)
173		    && (least->flags & IP_VS_DEST_F_AVAILABLE)) {
174			loh = ip_vs_dest_conn_overhead(least);
175			goto nextstage;
176		}
177	}
178	return NULL;
179
180	/* find the destination with the weighted least load */
181  nextstage:
182	list_for_each_entry_continue_rcu(e, &set->list, list) {
183		dest = e->dest;
184		if (dest->flags & IP_VS_DEST_F_OVERLOAD)
185			continue;
186
187		doh = ip_vs_dest_conn_overhead(dest);
188		if (((__s64)loh * atomic_read(&dest->weight) >
189		     (__s64)doh * atomic_read(&least->weight))
190		    && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
191			least = dest;
192			loh = doh;
193		}
194	}
195
196	IP_VS_DBG_BUF(6, "%s(): server %s:%d "
197		      "activeconns %d refcnt %d weight %d overhead %d\n",
198		      __func__,
199		      IP_VS_DBG_ADDR(least->af, &least->addr),
200		      ntohs(least->port),
201		      atomic_read(&least->activeconns),
202		      refcount_read(&least->refcnt),
203		      atomic_read(&least->weight), loh);
204	return least;
205}
206
207
208/* get weighted most-connection node in the destination set */
209static inline struct ip_vs_dest *ip_vs_dest_set_max(struct ip_vs_dest_set *set)
210{
211	struct ip_vs_dest_set_elem *e;
212	struct ip_vs_dest *dest, *most;
213	int moh, doh;
214
215	if (set == NULL)
216		return NULL;
217
218	/* select the first destination server, whose weight > 0 */
219	list_for_each_entry(e, &set->list, list) {
220		most = e->dest;
221		if (atomic_read(&most->weight) > 0) {
222			moh = ip_vs_dest_conn_overhead(most);
223			goto nextstage;
224		}
225	}
226	return NULL;
227
228	/* find the destination with the weighted most load */
229  nextstage:
230	list_for_each_entry_continue(e, &set->list, list) {
231		dest = e->dest;
232		doh = ip_vs_dest_conn_overhead(dest);
233		/* moh/mw < doh/dw ==> moh*dw < doh*mw, where mw,dw>0 */
234		if (((__s64)moh * atomic_read(&dest->weight) <
235		     (__s64)doh * atomic_read(&most->weight))
236		    && (atomic_read(&dest->weight) > 0)) {
237			most = dest;
238			moh = doh;
239		}
240	}
241
242	IP_VS_DBG_BUF(6, "%s(): server %s:%d "
243		      "activeconns %d refcnt %d weight %d overhead %d\n",
244		      __func__,
245		      IP_VS_DBG_ADDR(most->af, &most->addr), ntohs(most->port),
246		      atomic_read(&most->activeconns),
247		      refcount_read(&most->refcnt),
248		      atomic_read(&most->weight), moh);
249	return most;
250}
251
252
253/*
254 *      IPVS lblcr entry represents an association between destination
255 *      IP address and its destination server set
256 */
257struct ip_vs_lblcr_entry {
258	struct hlist_node       list;
259	int			af;		/* address family */
260	union nf_inet_addr      addr;           /* destination IP address */
261	struct ip_vs_dest_set   set;            /* destination server set */
262	unsigned long           lastuse;        /* last used time */
263	struct rcu_head		rcu_head;
264};
265
266
267/*
268 *      IPVS lblcr hash table
269 */
270struct ip_vs_lblcr_table {
271	struct rcu_head		rcu_head;
272	struct hlist_head	bucket[IP_VS_LBLCR_TAB_SIZE];  /* hash bucket */
273	atomic_t                entries;        /* number of entries */
274	int                     max_size;       /* maximum size of entries */
275	struct timer_list       periodic_timer; /* collect stale entries */
276	struct ip_vs_service	*svc;		/* pointer back to service */
277	int                     rover;          /* rover for expire check */
278	int                     counter;        /* counter for no expire */
279	bool			dead;
280};
281
282
283#ifdef CONFIG_SYSCTL
284/*
285 *      IPVS LBLCR sysctl table
286 */
287
288static struct ctl_table vs_vars_table[] = {
289	{
290		.procname	= "lblcr_expiration",
291		.data		= NULL,
292		.maxlen		= sizeof(int),
293		.mode		= 0644,
294		.proc_handler	= proc_dointvec_jiffies,
295	},
296};
297#endif
298
299static inline void ip_vs_lblcr_free(struct ip_vs_lblcr_entry *en)
300{
301	hlist_del_rcu(&en->list);
302	ip_vs_dest_set_eraseall(&en->set);
303	kfree_rcu(en, rcu_head);
304}
305
306
307/*
308 *	Returns hash value for IPVS LBLCR entry
309 */
310static inline unsigned int
311ip_vs_lblcr_hashkey(int af, const union nf_inet_addr *addr)
312{
313	__be32 addr_fold = addr->ip;
314
315#ifdef CONFIG_IP_VS_IPV6
316	if (af == AF_INET6)
317		addr_fold = addr->ip6[0]^addr->ip6[1]^
318			    addr->ip6[2]^addr->ip6[3];
319#endif
320	return hash_32(ntohl(addr_fold), IP_VS_LBLCR_TAB_BITS);
321}
322
323
324/*
325 *	Hash an entry in the ip_vs_lblcr_table.
326 *	returns bool success.
327 */
328static void
329ip_vs_lblcr_hash(struct ip_vs_lblcr_table *tbl, struct ip_vs_lblcr_entry *en)
330{
331	unsigned int hash = ip_vs_lblcr_hashkey(en->af, &en->addr);
332
333	hlist_add_head_rcu(&en->list, &tbl->bucket[hash]);
334	atomic_inc(&tbl->entries);
335}
336
337
338/* Get ip_vs_lblcr_entry associated with supplied parameters. */
339static inline struct ip_vs_lblcr_entry *
340ip_vs_lblcr_get(int af, struct ip_vs_lblcr_table *tbl,
341		const union nf_inet_addr *addr)
342{
343	unsigned int hash = ip_vs_lblcr_hashkey(af, addr);
344	struct ip_vs_lblcr_entry *en;
345
346	hlist_for_each_entry_rcu(en, &tbl->bucket[hash], list)
347		if (ip_vs_addr_equal(af, &en->addr, addr))
348			return en;
349
350	return NULL;
351}
352
353
354/*
355 * Create or update an ip_vs_lblcr_entry, which is a mapping of a destination
356 * IP address to a server. Called under spin lock.
357 */
358static inline struct ip_vs_lblcr_entry *
359ip_vs_lblcr_new(struct ip_vs_lblcr_table *tbl, const union nf_inet_addr *daddr,
360		u16 af, struct ip_vs_dest *dest)
361{
362	struct ip_vs_lblcr_entry *en;
363
364	en = ip_vs_lblcr_get(af, tbl, daddr);
365	if (!en) {
366		en = kmalloc(sizeof(*en), GFP_ATOMIC);
367		if (!en)
368			return NULL;
369
370		en->af = af;
371		ip_vs_addr_copy(af, &en->addr, daddr);
372		en->lastuse = jiffies;
373
374		/* initialize its dest set */
375		atomic_set(&(en->set.size), 0);
376		INIT_LIST_HEAD(&en->set.list);
377
378		ip_vs_dest_set_insert(&en->set, dest, false);
379
380		ip_vs_lblcr_hash(tbl, en);
381		return en;
382	}
383
384	ip_vs_dest_set_insert(&en->set, dest, true);
385
386	return en;
387}
388
389
390/*
391 *      Flush all the entries of the specified table.
392 */
393static void ip_vs_lblcr_flush(struct ip_vs_service *svc)
394{
395	struct ip_vs_lblcr_table *tbl = svc->sched_data;
396	int i;
397	struct ip_vs_lblcr_entry *en;
398	struct hlist_node *next;
399
400	spin_lock_bh(&svc->sched_lock);
401	tbl->dead = true;
402	for (i = 0; i < IP_VS_LBLCR_TAB_SIZE; i++) {
403		hlist_for_each_entry_safe(en, next, &tbl->bucket[i], list) {
404			ip_vs_lblcr_free(en);
405		}
406	}
407	spin_unlock_bh(&svc->sched_lock);
408}
409
410static int sysctl_lblcr_expiration(struct ip_vs_service *svc)
411{
412#ifdef CONFIG_SYSCTL
413	return svc->ipvs->sysctl_lblcr_expiration;
414#else
415	return DEFAULT_EXPIRATION;
416#endif
417}
418
419static inline void ip_vs_lblcr_full_check(struct ip_vs_service *svc)
420{
421	struct ip_vs_lblcr_table *tbl = svc->sched_data;
422	unsigned long now = jiffies;
423	int i, j;
424	struct ip_vs_lblcr_entry *en;
425	struct hlist_node *next;
426
427	for (i = 0, j = tbl->rover; i < IP_VS_LBLCR_TAB_SIZE; i++) {
428		j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
429
430		spin_lock(&svc->sched_lock);
431		hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) {
432			if (time_after(en->lastuse +
433				       sysctl_lblcr_expiration(svc), now))
434				continue;
435
436			ip_vs_lblcr_free(en);
437			atomic_dec(&tbl->entries);
438		}
439		spin_unlock(&svc->sched_lock);
440	}
441	tbl->rover = j;
442}
443
444
445/*
446 *      Periodical timer handler for IPVS lblcr table
447 *      It is used to collect stale entries when the number of entries
448 *      exceeds the maximum size of the table.
449 *
450 *      Fixme: we probably need more complicated algorithm to collect
451 *             entries that have not been used for a long time even
452 *             if the number of entries doesn't exceed the maximum size
453 *             of the table.
454 *      The full expiration check is for this purpose now.
455 */
456static void ip_vs_lblcr_check_expire(struct timer_list *t)
457{
458	struct ip_vs_lblcr_table *tbl = timer_container_of(tbl, t,
459							   periodic_timer);
460	struct ip_vs_service *svc = tbl->svc;
461	unsigned long now = jiffies;
462	int goal;
463	int i, j;
464	struct ip_vs_lblcr_entry *en;
465	struct hlist_node *next;
466
467	if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) {
468		/* do full expiration check */
469		ip_vs_lblcr_full_check(svc);
470		tbl->counter = 1;
471		goto out;
472	}
473
474	if (atomic_read(&tbl->entries) <= tbl->max_size) {
475		tbl->counter++;
476		goto out;
477	}
478
479	goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3;
480	if (goal > tbl->max_size/2)
481		goal = tbl->max_size/2;
482
483	for (i = 0, j = tbl->rover; i < IP_VS_LBLCR_TAB_SIZE; i++) {
484		j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
485
486		spin_lock(&svc->sched_lock);
487		hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) {
488			if (time_before(now, en->lastuse+ENTRY_TIMEOUT))
489				continue;
490
491			ip_vs_lblcr_free(en);
492			atomic_dec(&tbl->entries);
493			goal--;
494		}
495		spin_unlock(&svc->sched_lock);
496		if (goal <= 0)
497			break;
498	}
499	tbl->rover = j;
500
501  out:
502	mod_timer(&tbl->periodic_timer, jiffies+CHECK_EXPIRE_INTERVAL);
503}
504
505static int ip_vs_lblcr_init_svc(struct ip_vs_service *svc)
506{
507	int i;
508	struct ip_vs_lblcr_table *tbl;
509
510	/*
511	 *    Allocate the ip_vs_lblcr_table for this service
512	 */
513	tbl = kmalloc(sizeof(*tbl), GFP_KERNEL);
514	if (tbl == NULL)
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_HLIST_HEAD(&tbl->bucket[i]);
526	}
527	tbl->max_size = IP_VS_LBLCR_TAB_SIZE*16;
528	tbl->rover = 0;
529	tbl->counter = 1;
530	tbl->dead = false;
531	tbl->svc = svc;
532	atomic_set(&tbl->entries, 0);
533
534	/*
535	 *    Hook periodic timer for garbage collection
536	 */
537	timer_setup(&tbl->periodic_timer, ip_vs_lblcr_check_expire, 0);
538	mod_timer(&tbl->periodic_timer, jiffies + CHECK_EXPIRE_INTERVAL);
539
540	return 0;
541}
542
543
544static void ip_vs_lblcr_done_svc(struct ip_vs_service *svc)
545{
546	struct ip_vs_lblcr_table *tbl = svc->sched_data;
547
548	/* remove periodic timer */
549	timer_shutdown_sync(&tbl->periodic_timer);
550
551	/* got to clean up table entries here */
552	ip_vs_lblcr_flush(svc);
553
554	/* release the table itself */
555	kfree_rcu(tbl, rcu_head);
556	IP_VS_DBG(6, "LBLCR hash table (memory=%zdbytes) released\n",
557		  sizeof(*tbl));
558}
559
560
561static 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 use the following formula to estimate the load:
569	 *                (dest overhead) / dest->weight
570	 *
571	 * Remember -- no floats in kernel mode!!!
572	 * The comparison of h1*w2 > h2*w1 is equivalent to that of
573	 *                h1/w1 > h2/w2
574	 * if every weight is larger than zero.
575	 *
576	 * The server with weight=0 is quiesced and will not receive any
577	 * new connection.
578	 */
579	list_for_each_entry_rcu(dest, &svc->destinations, n_list) {
580		if (dest->flags & IP_VS_DEST_F_OVERLOAD)
581			continue;
582
583		if (atomic_read(&dest->weight) > 0) {
584			least = dest;
585			loh = ip_vs_dest_conn_overhead(least);
586			goto nextstage;
587		}
588	}
589	return NULL;
590
591	/*
592	 *    Find the destination with the least load.
593	 */
594  nextstage:
595	list_for_each_entry_continue_rcu(dest, &svc->destinations, n_list) {
596		if (dest->flags & IP_VS_DEST_F_OVERLOAD)
597			continue;
598
599		doh = ip_vs_dest_conn_overhead(dest);
600		if ((__s64)loh * atomic_read(&dest->weight) >
601		    (__s64)doh * atomic_read(&least->weight)) {
602			least = dest;
603			loh = doh;
604		}
605	}
606
607	IP_VS_DBG_BUF(6, "LBLCR: server %s:%d "
608		      "activeconns %d refcnt %d weight %d overhead %d\n",
609		      IP_VS_DBG_ADDR(least->af, &least->addr),
610		      ntohs(least->port),
611		      atomic_read(&least->activeconns),
612		      refcount_read(&least->refcnt),
613		      atomic_read(&least->weight), loh);
614
615	return least;
616}
617
618
619/*
620 *   If this destination server is overloaded and there is a less loaded
621 *   server, then return true.
622 */
623static inline int
624is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc)
625{
626	if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) {
627		struct ip_vs_dest *d;
628
629		list_for_each_entry_rcu(d, &svc->destinations, n_list) {
630			if (atomic_read(&d->activeconns)*2
631			    < atomic_read(&d->weight)) {
632				return 1;
633			}
634		}
635	}
636	return 0;
637}
638
639
640/*
641 *    Locality-Based (weighted) Least-Connection scheduling
642 */
643static struct ip_vs_dest *
644ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb,
645		     struct ip_vs_iphdr *iph)
646{
647	struct ip_vs_lblcr_table *tbl = svc->sched_data;
648	struct ip_vs_dest *dest;
649	struct ip_vs_lblcr_entry *en;
650
651	IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
652
653	/* First look in our cache */
654	en = ip_vs_lblcr_get(svc->af, tbl, &iph->daddr);
655	if (en) {
656		en->lastuse = jiffies;
657
658		/* Get the least loaded destination */
659		dest = ip_vs_dest_set_min(&en->set);
660
661		/* More than one destination + enough time passed by, cleanup */
662		if (atomic_read(&en->set.size) > 1 &&
663		    time_after(jiffies, en->set.lastmod +
664				sysctl_lblcr_expiration(svc))) {
665			spin_lock_bh(&svc->sched_lock);
666			if (atomic_read(&en->set.size) > 1) {
667				struct ip_vs_dest *m;
668
669				m = ip_vs_dest_set_max(&en->set);
670				if (m)
671					ip_vs_dest_set_erase(&en->set, m);
672			}
673			spin_unlock_bh(&svc->sched_lock);
674		}
675
676		/* If the destination is not overloaded, use it */
677		if (dest && !is_overloaded(dest, svc))
678			goto out;
679
680		/* The cache entry is invalid, time to schedule */
681		dest = __ip_vs_lblcr_schedule(svc);
682		if (!dest) {
683			ip_vs_scheduler_err(svc, "no destination available");
684			return NULL;
685		}
686
687		/* Update our cache entry */
688		spin_lock_bh(&svc->sched_lock);
689		if (!tbl->dead)
690			ip_vs_dest_set_insert(&en->set, dest, true);
691		spin_unlock_bh(&svc->sched_lock);
692		goto out;
693	}
694
695	/* No cache entry, time to schedule */
696	dest = __ip_vs_lblcr_schedule(svc);
697	if (!dest) {
698		IP_VS_DBG(1, "no destination available\n");
699		return NULL;
700	}
701
702	/* If we fail to create a cache entry, we'll just use the valid dest */
703	spin_lock_bh(&svc->sched_lock);
704	if (!tbl->dead)
705		ip_vs_lblcr_new(tbl, &iph->daddr, svc->af, dest);
706	spin_unlock_bh(&svc->sched_lock);
707
708out:
709	IP_VS_DBG_BUF(6, "LBLCR: destination IP address %s --> server %s:%d\n",
710		      IP_VS_DBG_ADDR(svc->af, &iph->daddr),
711		      IP_VS_DBG_ADDR(dest->af, &dest->addr), ntohs(dest->port));
712
713	return dest;
714}
715
716
717/*
718 *      IPVS LBLCR Scheduler structure
719 */
720static struct ip_vs_scheduler ip_vs_lblcr_scheduler =
721{
722	.name =			"lblcr",
723	.refcnt =		ATOMIC_INIT(0),
724	.module =		THIS_MODULE,
725	.n_list =		LIST_HEAD_INIT(ip_vs_lblcr_scheduler.n_list),
726	.init_service =		ip_vs_lblcr_init_svc,
727	.done_service =		ip_vs_lblcr_done_svc,
728	.schedule =		ip_vs_lblcr_schedule,
729};
730
731/*
732 *  per netns init.
733 */
734#ifdef CONFIG_SYSCTL
735static int __net_init __ip_vs_lblcr_init(struct net *net)
736{
737	struct netns_ipvs *ipvs = net_ipvs(net);
738	size_t vars_table_size = ARRAY_SIZE(vs_vars_table);
739
740	if (!ipvs)
741		return -ENOENT;
742
743	if (!net_eq(net, &init_net)) {
744		ipvs->lblcr_ctl_table = kmemdup(vs_vars_table,
745						sizeof(vs_vars_table),
746						GFP_KERNEL);
747		if (ipvs->lblcr_ctl_table == NULL)
748			return -ENOMEM;
749
750		/* Don't export sysctls to unprivileged users */
751		if (net->user_ns != &init_user_ns)
752			vars_table_size = 0;
753	} else
754		ipvs->lblcr_ctl_table = vs_vars_table;
755	ipvs->sysctl_lblcr_expiration = DEFAULT_EXPIRATION;
756	ipvs->lblcr_ctl_table[0].data = &ipvs->sysctl_lblcr_expiration;
757
758	ipvs->lblcr_ctl_header = register_net_sysctl_sz(net, "net/ipv4/vs",
759							ipvs->lblcr_ctl_table,
760							vars_table_size);
761	if (!ipvs->lblcr_ctl_header) {
762		if (!net_eq(net, &init_net))
763			kfree(ipvs->lblcr_ctl_table);
764		return -ENOMEM;
765	}
766
767	return 0;
768}
769
770static void __net_exit __ip_vs_lblcr_exit(struct net *net)
771{
772	struct netns_ipvs *ipvs = net_ipvs(net);
773
774	unregister_net_sysctl_table(ipvs->lblcr_ctl_header);
775
776	if (!net_eq(net, &init_net))
777		kfree(ipvs->lblcr_ctl_table);
778}
779
780#else
781
782static int __net_init __ip_vs_lblcr_init(struct net *net) { return 0; }
783static void __net_exit __ip_vs_lblcr_exit(struct net *net) { }
784
785#endif
786
787static struct pernet_operations ip_vs_lblcr_ops = {
788	.init = __ip_vs_lblcr_init,
789	.exit = __ip_vs_lblcr_exit,
790};
791
792static int __init ip_vs_lblcr_init(void)
793{
794	int ret;
795
796	ret = register_pernet_subsys(&ip_vs_lblcr_ops);
797	if (ret)
798		return ret;
799
800	ret = register_ip_vs_scheduler(&ip_vs_lblcr_scheduler);
801	if (ret)
802		unregister_pernet_subsys(&ip_vs_lblcr_ops);
803	return ret;
804}
805
806static void __exit ip_vs_lblcr_cleanup(void)
807{
808	unregister_ip_vs_scheduler(&ip_vs_lblcr_scheduler);
809	unregister_pernet_subsys(&ip_vs_lblcr_ops);
810	rcu_barrier();
811}
812
813
814module_init(ip_vs_lblcr_init);
815module_exit(ip_vs_lblcr_cleanup);
816MODULE_LICENSE("GPL");
817MODULE_DESCRIPTION("ipvs locality-based least-connection with replication scheduler");