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