This source file includes following definitions.
- ip_vs_lblc_rcu_free
- ip_vs_lblc_del
- ip_vs_lblc_hashkey
- ip_vs_lblc_hash
- ip_vs_lblc_get
- ip_vs_lblc_new
- ip_vs_lblc_flush
- sysctl_lblc_expiration
- ip_vs_lblc_full_check
- ip_vs_lblc_check_expire
- ip_vs_lblc_init_svc
- ip_vs_lblc_done_svc
- __ip_vs_lblc_schedule
- is_overloaded
- ip_vs_lblc_schedule
- __ip_vs_lblc_init
- __ip_vs_lblc_exit
- __ip_vs_lblc_init
- __ip_vs_lblc_exit
- ip_vs_lblc_init
- ip_vs_lblc_cleanup
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 #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
49 #include <linux/fs.h>
50 #include <linux/sysctl.h>
51
52 #include <net/ip_vs.h>
53
54
55
56
57
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
66
67
68
69
70 #define COUNT_FOR_FULL_EXPIRATION 30
71
72
73
74
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
86
87
88 struct ip_vs_lblc_entry {
89 struct hlist_node list;
90 int af;
91 union nf_inet_addr addr;
92 struct ip_vs_dest *dest;
93 unsigned long lastuse;
94 struct rcu_head rcu_head;
95 };
96
97
98
99
100
101 struct ip_vs_lblc_table {
102 struct rcu_head rcu_head;
103 struct hlist_head bucket[IP_VS_LBLC_TAB_SIZE];
104 struct timer_list periodic_timer;
105 struct ip_vs_service *svc;
106 atomic_t entries;
107 int max_size;
108 int rover;
109 int counter;
110 bool dead;
111 };
112
113
114
115
116
117 #ifdef CONFIG_SYSCTL
118 static 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
130 static 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
140 static 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
148
149 static inline unsigned int
150 ip_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
165
166
167 static void
168 ip_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
178 static inline struct ip_vs_lblc_entry *
179 ip_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
195
196
197 static inline struct ip_vs_lblc_entry *
198 ip_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
228
229 static 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
247 static 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
256 static 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
285
286
287
288
289
290
291
292
293
294 static 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
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
343 static 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
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
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
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
382 static void ip_vs_lblc_done_svc(struct ip_vs_service *svc)
383 {
384 struct ip_vs_lblc_table *tbl = svc->sched_data;
385
386
387 del_timer_sync(&tbl->periodic_timer);
388
389
390 ip_vs_lblc_flush(svc);
391
392
393 kfree_rcu(tbl, rcu_head);
394 IP_VS_DBG(6, "LBLC hash table (memory=%zdbytes) released\n",
395 sizeof(*tbl));
396 }
397
398
399 static 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
407
408
409
410
411
412
413
414
415
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
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
458
459
460 static inline int
461 is_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
479
480 static struct ip_vs_dest *
481 ip_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
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 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
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
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
524 out:
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
535
536 static 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
548
549 #ifdef CONFIG_SYSCTL
550 static int __net_init __ip_vs_lblc_init(struct net *net)
551 {
552 struct netns_ipvs *ipvs = net_ipvs(net);
553
554 if (!ipvs)
555 return -ENOENT;
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
564
565 if (net->user_ns != &init_user_ns)
566 ipvs->lblc_ctl_table[0].procname = NULL;
567
568 } else
569 ipvs->lblc_ctl_table = vs_vars_table;
570 ipvs->sysctl_lblc_expiration = DEFAULT_EXPIRATION;
571 ipvs->lblc_ctl_table[0].data = &ipvs->sysctl_lblc_expiration;
572
573 ipvs->lblc_ctl_header =
574 register_net_sysctl(net, "net/ipv4/vs", ipvs->lblc_ctl_table);
575 if (!ipvs->lblc_ctl_header) {
576 if (!net_eq(net, &init_net))
577 kfree(ipvs->lblc_ctl_table);
578 return -ENOMEM;
579 }
580
581 return 0;
582 }
583
584 static void __net_exit __ip_vs_lblc_exit(struct net *net)
585 {
586 struct netns_ipvs *ipvs = net_ipvs(net);
587
588 unregister_net_sysctl_table(ipvs->lblc_ctl_header);
589
590 if (!net_eq(net, &init_net))
591 kfree(ipvs->lblc_ctl_table);
592 }
593
594 #else
595
596 static int __net_init __ip_vs_lblc_init(struct net *net) { return 0; }
597 static void __net_exit __ip_vs_lblc_exit(struct net *net) { }
598
599 #endif
600
601 static struct pernet_operations ip_vs_lblc_ops = {
602 .init = __ip_vs_lblc_init,
603 .exit = __ip_vs_lblc_exit,
604 };
605
606 static int __init ip_vs_lblc_init(void)
607 {
608 int ret;
609
610 ret = register_pernet_subsys(&ip_vs_lblc_ops);
611 if (ret)
612 return ret;
613
614 ret = register_ip_vs_scheduler(&ip_vs_lblc_scheduler);
615 if (ret)
616 unregister_pernet_subsys(&ip_vs_lblc_ops);
617 return ret;
618 }
619
620 static void __exit ip_vs_lblc_cleanup(void)
621 {
622 unregister_ip_vs_scheduler(&ip_vs_lblc_scheduler);
623 unregister_pernet_subsys(&ip_vs_lblc_ops);
624 rcu_barrier();
625 }
626
627
628 module_init(ip_vs_lblc_init);
629 module_exit(ip_vs_lblc_cleanup);
630 MODULE_LICENSE("GPL");