1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42#define KMSG_COMPONENT "IPVS"
43#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
44
45#include <linux/ip.h>
46#include <linux/slab.h>
47#include <linux/module.h>
48#include <linux/kernel.h>
49#include <linux/skbuff.h>
50#include <linux/jiffies.h>
51
52
53#include <linux/fs.h>
54#include <linux/sysctl.h>
55
56#include <net/ip_vs.h>
57
58
59
60
61
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
70
71
72
73
74#define COUNT_FOR_FULL_EXPIRATION 30
75
76
77
78
79
80#ifndef CONFIG_IP_VS_LBLC_TAB_BITS
81#define CONFIG_IP_VS_LBLC_TAB_BITS 10
82#endif
83#define IP_VS_LBLC_TAB_BITS CONFIG_IP_VS_LBLC_TAB_BITS
84#define IP_VS_LBLC_TAB_SIZE (1 << IP_VS_LBLC_TAB_BITS)
85#define IP_VS_LBLC_TAB_MASK (IP_VS_LBLC_TAB_SIZE - 1)
86
87
88
89
90
91
92struct ip_vs_lblc_entry {
93 struct hlist_node list;
94 int af;
95 union nf_inet_addr addr;
96 struct ip_vs_dest __rcu *dest;
97 unsigned long lastuse;
98 struct rcu_head rcu_head;
99};
100
101
102
103
104
105struct ip_vs_lblc_table {
106 struct rcu_head rcu_head;
107 struct hlist_head bucket[IP_VS_LBLC_TAB_SIZE];
108 struct timer_list periodic_timer;
109 atomic_t entries;
110 int max_size;
111 int rover;
112 int counter;
113 bool dead;
114};
115
116
117
118
119
120#ifdef CONFIG_SYSCTL
121static ctl_table vs_vars_table[] = {
122 {
123 .procname = "lblc_expiration",
124 .data = NULL,
125 .maxlen = sizeof(int),
126 .mode = 0644,
127 .proc_handler = proc_dointvec_jiffies,
128 },
129 { }
130};
131#endif
132
133static inline void ip_vs_lblc_free(struct ip_vs_lblc_entry *en)
134{
135 struct ip_vs_dest *dest;
136
137 hlist_del_rcu(&en->list);
138
139
140
141
142 dest = rcu_dereference_protected(en->dest, 1);
143 ip_vs_dest_put(dest);
144 kfree_rcu(en, rcu_head);
145}
146
147
148
149
150
151static inline unsigned int
152ip_vs_lblc_hashkey(int af, const union nf_inet_addr *addr)
153{
154 __be32 addr_fold = addr->ip;
155
156#ifdef CONFIG_IP_VS_IPV6
157 if (af == AF_INET6)
158 addr_fold = addr->ip6[0]^addr->ip6[1]^
159 addr->ip6[2]^addr->ip6[3];
160#endif
161 return (ntohl(addr_fold)*2654435761UL) & IP_VS_LBLC_TAB_MASK;
162}
163
164
165
166
167
168
169static void
170ip_vs_lblc_hash(struct ip_vs_lblc_table *tbl, struct ip_vs_lblc_entry *en)
171{
172 unsigned int hash = ip_vs_lblc_hashkey(en->af, &en->addr);
173
174 hlist_add_head_rcu(&en->list, &tbl->bucket[hash]);
175 atomic_inc(&tbl->entries);
176}
177
178
179
180static inline struct ip_vs_lblc_entry *
181ip_vs_lblc_get(int af, struct ip_vs_lblc_table *tbl,
182 const union nf_inet_addr *addr)
183{
184 unsigned int hash = ip_vs_lblc_hashkey(af, addr);
185 struct ip_vs_lblc_entry *en;
186
187 hlist_for_each_entry_rcu(en, &tbl->bucket[hash], list)
188 if (ip_vs_addr_equal(af, &en->addr, addr))
189 return en;
190
191 return NULL;
192}
193
194
195
196
197
198
199static inline struct ip_vs_lblc_entry *
200ip_vs_lblc_new(struct ip_vs_lblc_table *tbl, const union nf_inet_addr *daddr,
201 struct ip_vs_dest *dest)
202{
203 struct ip_vs_lblc_entry *en;
204
205 en = ip_vs_lblc_get(dest->af, tbl, daddr);
206 if (!en) {
207 en = kmalloc(sizeof(*en), GFP_ATOMIC);
208 if (!en)
209 return NULL;
210
211 en->af = dest->af;
212 ip_vs_addr_copy(dest->af, &en->addr, daddr);
213 en->lastuse = jiffies;
214
215 ip_vs_dest_hold(dest);
216 RCU_INIT_POINTER(en->dest, dest);
217
218 ip_vs_lblc_hash(tbl, en);
219 } else {
220 struct ip_vs_dest *old_dest;
221
222 old_dest = rcu_dereference_protected(en->dest, 1);
223 if (old_dest != dest) {
224 ip_vs_dest_put(old_dest);
225 ip_vs_dest_hold(dest);
226
227 RCU_INIT_POINTER(en->dest, dest);
228 }
229 }
230
231 return en;
232}
233
234
235
236
237
238static void ip_vs_lblc_flush(struct ip_vs_service *svc)
239{
240 struct ip_vs_lblc_table *tbl = svc->sched_data;
241 struct ip_vs_lblc_entry *en;
242 struct hlist_node *next;
243 int i;
244
245 spin_lock_bh(&svc->sched_lock);
246 tbl->dead = 1;
247 for (i=0; i<IP_VS_LBLC_TAB_SIZE; i++) {
248 hlist_for_each_entry_safe(en, next, &tbl->bucket[i], list) {
249 ip_vs_lblc_free(en);
250 atomic_dec(&tbl->entries);
251 }
252 }
253 spin_unlock_bh(&svc->sched_lock);
254}
255
256static int sysctl_lblc_expiration(struct ip_vs_service *svc)
257{
258#ifdef CONFIG_SYSCTL
259 struct netns_ipvs *ipvs = net_ipvs(svc->net);
260 return ipvs->sysctl_lblc_expiration;
261#else
262 return DEFAULT_EXPIRATION;
263#endif
264}
265
266static inline void ip_vs_lblc_full_check(struct ip_vs_service *svc)
267{
268 struct ip_vs_lblc_table *tbl = svc->sched_data;
269 struct ip_vs_lblc_entry *en;
270 struct hlist_node *next;
271 unsigned long now = jiffies;
272 int i, j;
273
274 for (i=0, j=tbl->rover; i<IP_VS_LBLC_TAB_SIZE; i++) {
275 j = (j + 1) & IP_VS_LBLC_TAB_MASK;
276
277 spin_lock(&svc->sched_lock);
278 hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) {
279 if (time_before(now,
280 en->lastuse +
281 sysctl_lblc_expiration(svc)))
282 continue;
283
284 ip_vs_lblc_free(en);
285 atomic_dec(&tbl->entries);
286 }
287 spin_unlock(&svc->sched_lock);
288 }
289 tbl->rover = j;
290}
291
292
293
294
295
296
297
298
299
300
301
302
303
304static void ip_vs_lblc_check_expire(unsigned long data)
305{
306 struct ip_vs_service *svc = (struct ip_vs_service *) data;
307 struct ip_vs_lblc_table *tbl = svc->sched_data;
308 unsigned long now = jiffies;
309 int goal;
310 int i, j;
311 struct ip_vs_lblc_entry *en;
312 struct hlist_node *next;
313
314 if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) {
315
316 ip_vs_lblc_full_check(svc);
317 tbl->counter = 1;
318 goto out;
319 }
320
321 if (atomic_read(&tbl->entries) <= tbl->max_size) {
322 tbl->counter++;
323 goto out;
324 }
325
326 goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3;
327 if (goal > tbl->max_size/2)
328 goal = tbl->max_size/2;
329
330 for (i=0, j=tbl->rover; i<IP_VS_LBLC_TAB_SIZE; i++) {
331 j = (j + 1) & IP_VS_LBLC_TAB_MASK;
332
333 spin_lock(&svc->sched_lock);
334 hlist_for_each_entry_safe(en, next, &tbl->bucket[j], list) {
335 if (time_before(now, en->lastuse + ENTRY_TIMEOUT))
336 continue;
337
338 ip_vs_lblc_free(en);
339 atomic_dec(&tbl->entries);
340 goal--;
341 }
342 spin_unlock(&svc->sched_lock);
343 if (goal <= 0)
344 break;
345 }
346 tbl->rover = j;
347
348 out:
349 mod_timer(&tbl->periodic_timer, jiffies+CHECK_EXPIRE_INTERVAL);
350}
351
352
353static int ip_vs_lblc_init_svc(struct ip_vs_service *svc)
354{
355 int i;
356 struct ip_vs_lblc_table *tbl;
357
358
359
360
361 tbl = kmalloc(sizeof(*tbl), GFP_KERNEL);
362 if (tbl == NULL)
363 return -ENOMEM;
364
365 svc->sched_data = tbl;
366 IP_VS_DBG(6, "LBLC hash table (memory=%Zdbytes) allocated for "
367 "current service\n", sizeof(*tbl));
368
369
370
371
372 for (i=0; i<IP_VS_LBLC_TAB_SIZE; i++) {
373 INIT_HLIST_HEAD(&tbl->bucket[i]);
374 }
375 tbl->max_size = IP_VS_LBLC_TAB_SIZE*16;
376 tbl->rover = 0;
377 tbl->counter = 1;
378 tbl->dead = 0;
379
380
381
382
383 setup_timer(&tbl->periodic_timer, ip_vs_lblc_check_expire,
384 (unsigned long)svc);
385 mod_timer(&tbl->periodic_timer, jiffies + CHECK_EXPIRE_INTERVAL);
386
387 return 0;
388}
389
390
391static void ip_vs_lblc_done_svc(struct ip_vs_service *svc)
392{
393 struct ip_vs_lblc_table *tbl = svc->sched_data;
394
395
396 del_timer_sync(&tbl->periodic_timer);
397
398
399 ip_vs_lblc_flush(svc);
400
401
402 kfree_rcu(tbl, rcu_head);
403 IP_VS_DBG(6, "LBLC hash table (memory=%Zdbytes) released\n",
404 sizeof(*tbl));
405}
406
407
408static inline struct ip_vs_dest *
409__ip_vs_lblc_schedule(struct ip_vs_service *svc)
410{
411 struct ip_vs_dest *dest, *least;
412 int loh, doh;
413
414
415
416
417
418
419
420
421
422
423
424
425
426 list_for_each_entry_rcu(dest, &svc->destinations, n_list) {
427 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
428 continue;
429 if (atomic_read(&dest->weight) > 0) {
430 least = dest;
431 loh = ip_vs_dest_conn_overhead(least);
432 goto nextstage;
433 }
434 }
435 return NULL;
436
437
438
439
440 nextstage:
441 list_for_each_entry_continue_rcu(dest, &svc->destinations, n_list) {
442 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
443 continue;
444
445 doh = ip_vs_dest_conn_overhead(dest);
446 if (loh * atomic_read(&dest->weight) >
447 doh * atomic_read(&least->weight)) {
448 least = dest;
449 loh = doh;
450 }
451 }
452
453 IP_VS_DBG_BUF(6, "LBLC: server %s:%d "
454 "activeconns %d refcnt %d weight %d overhead %d\n",
455 IP_VS_DBG_ADDR(least->af, &least->addr),
456 ntohs(least->port),
457 atomic_read(&least->activeconns),
458 atomic_read(&least->refcnt),
459 atomic_read(&least->weight), loh);
460
461 return least;
462}
463
464
465
466
467
468
469static inline int
470is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc)
471{
472 if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) {
473 struct ip_vs_dest *d;
474
475 list_for_each_entry_rcu(d, &svc->destinations, n_list) {
476 if (atomic_read(&d->activeconns)*2
477 < atomic_read(&d->weight)) {
478 return 1;
479 }
480 }
481 }
482 return 0;
483}
484
485
486
487
488
489static struct ip_vs_dest *
490ip_vs_lblc_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
491{
492 struct ip_vs_lblc_table *tbl = svc->sched_data;
493 struct ip_vs_iphdr iph;
494 struct ip_vs_dest *dest = NULL;
495 struct ip_vs_lblc_entry *en;
496
497 ip_vs_fill_iph_addr_only(svc->af, skb, &iph);
498
499 IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
500
501
502 en = ip_vs_lblc_get(svc->af, tbl, &iph.daddr);
503 if (en) {
504
505 en->lastuse = jiffies;
506
507
508
509
510
511
512
513
514
515
516 dest = rcu_dereference(en->dest);
517 if ((dest->flags & IP_VS_DEST_F_AVAILABLE) &&
518 atomic_read(&dest->weight) > 0 && !is_overloaded(dest, svc))
519 goto out;
520 }
521
522
523 dest = __ip_vs_lblc_schedule(svc);
524 if (!dest) {
525 ip_vs_scheduler_err(svc, "no destination available");
526 return NULL;
527 }
528
529
530 spin_lock_bh(&svc->sched_lock);
531 if (!tbl->dead)
532 ip_vs_lblc_new(tbl, &iph.daddr, dest);
533 spin_unlock_bh(&svc->sched_lock);
534
535out:
536 IP_VS_DBG_BUF(6, "LBLC: destination IP address %s --> server %s:%d\n",
537 IP_VS_DBG_ADDR(svc->af, &iph.daddr),
538 IP_VS_DBG_ADDR(svc->af, &dest->addr), ntohs(dest->port));
539
540 return dest;
541}
542
543
544
545
546
547static struct ip_vs_scheduler ip_vs_lblc_scheduler =
548{
549 .name = "lblc",
550 .refcnt = ATOMIC_INIT(0),
551 .module = THIS_MODULE,
552 .n_list = LIST_HEAD_INIT(ip_vs_lblc_scheduler.n_list),
553 .init_service = ip_vs_lblc_init_svc,
554 .done_service = ip_vs_lblc_done_svc,
555 .schedule = ip_vs_lblc_schedule,
556};
557
558
559
560
561#ifdef CONFIG_SYSCTL
562static int __net_init __ip_vs_lblc_init(struct net *net)
563{
564 struct netns_ipvs *ipvs = net_ipvs(net);
565
566 if (!ipvs)
567 return -ENOENT;
568
569 if (!net_eq(net, &init_net)) {
570 ipvs->lblc_ctl_table = kmemdup(vs_vars_table,
571 sizeof(vs_vars_table),
572 GFP_KERNEL);
573 if (ipvs->lblc_ctl_table == NULL)
574 return -ENOMEM;
575
576
577 if (net->user_ns != &init_user_ns)
578 ipvs->lblc_ctl_table[0].procname = NULL;
579
580 } else
581 ipvs->lblc_ctl_table = vs_vars_table;
582 ipvs->sysctl_lblc_expiration = DEFAULT_EXPIRATION;
583 ipvs->lblc_ctl_table[0].data = &ipvs->sysctl_lblc_expiration;
584
585 ipvs->lblc_ctl_header =
586 register_net_sysctl(net, "net/ipv4/vs", ipvs->lblc_ctl_table);
587 if (!ipvs->lblc_ctl_header) {
588 if (!net_eq(net, &init_net))
589 kfree(ipvs->lblc_ctl_table);
590 return -ENOMEM;
591 }
592
593 return 0;
594}
595
596static void __net_exit __ip_vs_lblc_exit(struct net *net)
597{
598 struct netns_ipvs *ipvs = net_ipvs(net);
599
600 unregister_net_sysctl_table(ipvs->lblc_ctl_header);
601
602 if (!net_eq(net, &init_net))
603 kfree(ipvs->lblc_ctl_table);
604}
605
606#else
607
608static int __net_init __ip_vs_lblc_init(struct net *net) { return 0; }
609static void __net_exit __ip_vs_lblc_exit(struct net *net) { }
610
611#endif
612
613static struct pernet_operations ip_vs_lblc_ops = {
614 .init = __ip_vs_lblc_init,
615 .exit = __ip_vs_lblc_exit,
616};
617
618static int __init ip_vs_lblc_init(void)
619{
620 int ret;
621
622 ret = register_pernet_subsys(&ip_vs_lblc_ops);
623 if (ret)
624 return ret;
625
626 ret = register_ip_vs_scheduler(&ip_vs_lblc_scheduler);
627 if (ret)
628 unregister_pernet_subsys(&ip_vs_lblc_ops);
629 return ret;
630}
631
632static void __exit ip_vs_lblc_cleanup(void)
633{
634 unregister_ip_vs_scheduler(&ip_vs_lblc_scheduler);
635 unregister_pernet_subsys(&ip_vs_lblc_ops);
636 synchronize_rcu();
637}
638
639
640module_init(ip_vs_lblc_init);
641module_exit(ip_vs_lblc_cleanup);
642MODULE_LICENSE("GPL");
643