]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * IPVS: Locality-Based Least-Connection with Replication scheduler | |
3 | * | |
1da177e4 LT |
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 | ||
14c85021 | 40 | #include <linux/ip.h> |
1da177e4 LT |
41 | #include <linux/module.h> |
42 | #include <linux/kernel.h> | |
14c85021 | 43 | #include <linux/skbuff.h> |
d7fe0f24 | 44 | #include <linux/jiffies.h> |
1da177e4 LT |
45 | |
46 | /* for sysctl */ | |
47 | #include <linux/fs.h> | |
48 | #include <linux/sysctl.h> | |
457c4cbc | 49 | #include <net/net_namespace.h> |
1da177e4 LT |
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 | /* | |
62 | * It is for full expiration check. | |
63 | * When there is no partial expiration check (garbage collection) | |
64 | * in a half hour, do a full expiration check to collect stale | |
65 | * entries that haven't been touched for a day. | |
66 | */ | |
67 | #define COUNT_FOR_FULL_EXPIRATION 30 | |
68 | static int sysctl_ip_vs_lblcr_expiration = 24*60*60*HZ; | |
69 | ||
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 | */ | |
85 | struct ip_vs_dest_list { | |
86 | struct ip_vs_dest_list *next; /* list link */ | |
87 | struct ip_vs_dest *dest; /* destination server */ | |
88 | }; | |
89 | ||
90 | struct ip_vs_dest_set { | |
91 | atomic_t size; /* set size */ | |
92 | unsigned long lastmod; /* last modified time */ | |
93 | struct ip_vs_dest_list *list; /* destination list */ | |
94 | rwlock_t lock; /* lock for this list */ | |
95 | }; | |
96 | ||
97 | ||
98 | static struct ip_vs_dest_list * | |
99 | ip_vs_dest_set_insert(struct ip_vs_dest_set *set, struct ip_vs_dest *dest) | |
100 | { | |
101 | struct ip_vs_dest_list *e; | |
102 | ||
103 | for (e=set->list; e!=NULL; e=e->next) { | |
104 | if (e->dest == dest) | |
105 | /* already existed */ | |
106 | return NULL; | |
107 | } | |
108 | ||
f728bafb | 109 | e = kmalloc(sizeof(*e), GFP_ATOMIC); |
1da177e4 LT |
110 | if (e == NULL) { |
111 | IP_VS_ERR("ip_vs_dest_set_insert(): no memory\n"); | |
112 | return NULL; | |
113 | } | |
114 | ||
115 | atomic_inc(&dest->refcnt); | |
116 | e->dest = dest; | |
117 | ||
118 | /* link it to the list */ | |
1da177e4 LT |
119 | e->next = set->list; |
120 | set->list = e; | |
121 | atomic_inc(&set->size); | |
1da177e4 LT |
122 | |
123 | set->lastmod = jiffies; | |
124 | return e; | |
125 | } | |
126 | ||
127 | static void | |
128 | ip_vs_dest_set_erase(struct ip_vs_dest_set *set, struct ip_vs_dest *dest) | |
129 | { | |
130 | struct ip_vs_dest_list *e, **ep; | |
131 | ||
1da177e4 LT |
132 | for (ep=&set->list, e=*ep; e!=NULL; e=*ep) { |
133 | if (e->dest == dest) { | |
134 | /* HIT */ | |
135 | *ep = e->next; | |
136 | atomic_dec(&set->size); | |
137 | set->lastmod = jiffies; | |
138 | atomic_dec(&e->dest->refcnt); | |
139 | kfree(e); | |
140 | break; | |
141 | } | |
142 | ep = &e->next; | |
143 | } | |
1da177e4 LT |
144 | } |
145 | ||
146 | static void ip_vs_dest_set_eraseall(struct ip_vs_dest_set *set) | |
147 | { | |
148 | struct ip_vs_dest_list *e, **ep; | |
149 | ||
150 | write_lock(&set->lock); | |
151 | for (ep=&set->list, e=*ep; e!=NULL; e=*ep) { | |
152 | *ep = e->next; | |
153 | /* | |
154 | * We don't kfree dest because it is refered either | |
155 | * by its service or by the trash dest list. | |
156 | */ | |
157 | atomic_dec(&e->dest->refcnt); | |
158 | kfree(e); | |
159 | } | |
160 | write_unlock(&set->lock); | |
161 | } | |
162 | ||
163 | /* get weighted least-connection node in the destination set */ | |
164 | static inline struct ip_vs_dest *ip_vs_dest_set_min(struct ip_vs_dest_set *set) | |
165 | { | |
166 | register struct ip_vs_dest_list *e; | |
167 | struct ip_vs_dest *dest, *least; | |
168 | int loh, doh; | |
169 | ||
170 | if (set == NULL) | |
171 | return NULL; | |
172 | ||
1da177e4 LT |
173 | /* select the first destination server, whose weight > 0 */ |
174 | for (e=set->list; e!=NULL; e=e->next) { | |
175 | least = e->dest; | |
176 | if (least->flags & IP_VS_DEST_F_OVERLOAD) | |
177 | continue; | |
178 | ||
179 | if ((atomic_read(&least->weight) > 0) | |
180 | && (least->flags & IP_VS_DEST_F_AVAILABLE)) { | |
181 | loh = atomic_read(&least->activeconns) * 50 | |
182 | + atomic_read(&least->inactconns); | |
183 | goto nextstage; | |
184 | } | |
185 | } | |
1da177e4 LT |
186 | return NULL; |
187 | ||
188 | /* find the destination with the weighted least load */ | |
189 | nextstage: | |
190 | for (e=e->next; e!=NULL; e=e->next) { | |
191 | dest = e->dest; | |
192 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
193 | continue; | |
194 | ||
195 | doh = atomic_read(&dest->activeconns) * 50 | |
196 | + atomic_read(&dest->inactconns); | |
197 | if ((loh * atomic_read(&dest->weight) > | |
198 | doh * atomic_read(&least->weight)) | |
199 | && (dest->flags & IP_VS_DEST_F_AVAILABLE)) { | |
200 | least = dest; | |
201 | loh = doh; | |
202 | } | |
203 | } | |
1da177e4 | 204 | |
44548375 JV |
205 | IP_VS_DBG_BUF(6, "ip_vs_dest_set_min: server %s:%d " |
206 | "activeconns %d refcnt %d weight %d overhead %d\n", | |
207 | IP_VS_DBG_ADDR(least->af, &least->addr), | |
208 | ntohs(least->port), | |
209 | atomic_read(&least->activeconns), | |
210 | atomic_read(&least->refcnt), | |
211 | atomic_read(&least->weight), loh); | |
1da177e4 LT |
212 | return least; |
213 | } | |
214 | ||
215 | ||
216 | /* get weighted most-connection node in the destination set */ | |
217 | static inline struct ip_vs_dest *ip_vs_dest_set_max(struct ip_vs_dest_set *set) | |
218 | { | |
219 | register struct ip_vs_dest_list *e; | |
220 | struct ip_vs_dest *dest, *most; | |
221 | int moh, doh; | |
222 | ||
223 | if (set == NULL) | |
224 | return NULL; | |
225 | ||
1da177e4 LT |
226 | /* select the first destination server, whose weight > 0 */ |
227 | for (e=set->list; e!=NULL; e=e->next) { | |
228 | most = e->dest; | |
229 | if (atomic_read(&most->weight) > 0) { | |
230 | moh = atomic_read(&most->activeconns) * 50 | |
231 | + atomic_read(&most->inactconns); | |
232 | goto nextstage; | |
233 | } | |
234 | } | |
1da177e4 LT |
235 | return NULL; |
236 | ||
237 | /* find the destination with the weighted most load */ | |
238 | nextstage: | |
239 | for (e=e->next; e!=NULL; e=e->next) { | |
240 | dest = e->dest; | |
241 | doh = atomic_read(&dest->activeconns) * 50 | |
242 | + atomic_read(&dest->inactconns); | |
243 | /* moh/mw < doh/dw ==> moh*dw < doh*mw, where mw,dw>0 */ | |
244 | if ((moh * atomic_read(&dest->weight) < | |
245 | doh * atomic_read(&most->weight)) | |
246 | && (atomic_read(&dest->weight) > 0)) { | |
247 | most = dest; | |
248 | moh = doh; | |
249 | } | |
250 | } | |
1da177e4 | 251 | |
44548375 JV |
252 | IP_VS_DBG_BUF(6, "ip_vs_dest_set_max: server %s:%d " |
253 | "activeconns %d refcnt %d weight %d overhead %d\n", | |
254 | IP_VS_DBG_ADDR(most->af, &most->addr), ntohs(most->port), | |
255 | atomic_read(&most->activeconns), | |
256 | atomic_read(&most->refcnt), | |
257 | atomic_read(&most->weight), moh); | |
1da177e4 LT |
258 | return most; |
259 | } | |
260 | ||
261 | ||
262 | /* | |
263 | * IPVS lblcr entry represents an association between destination | |
264 | * IP address and its destination server set | |
265 | */ | |
266 | struct ip_vs_lblcr_entry { | |
267 | struct list_head list; | |
44548375 JV |
268 | int af; /* address family */ |
269 | union nf_inet_addr addr; /* destination IP address */ | |
1da177e4 LT |
270 | struct ip_vs_dest_set set; /* destination server set */ |
271 | unsigned long lastuse; /* last used time */ | |
272 | }; | |
273 | ||
274 | ||
275 | /* | |
276 | * IPVS lblcr hash table | |
277 | */ | |
278 | struct ip_vs_lblcr_table { | |
1da177e4 LT |
279 | struct list_head bucket[IP_VS_LBLCR_TAB_SIZE]; /* hash bucket */ |
280 | atomic_t entries; /* number of entries */ | |
281 | int max_size; /* maximum size of entries */ | |
282 | struct timer_list periodic_timer; /* collect stale entries */ | |
283 | int rover; /* rover for expire check */ | |
284 | int counter; /* counter for no expire */ | |
285 | }; | |
286 | ||
287 | ||
288 | /* | |
289 | * IPVS LBLCR sysctl table | |
290 | */ | |
291 | ||
292 | static ctl_table vs_vars_table[] = { | |
293 | { | |
1da177e4 LT |
294 | .procname = "lblcr_expiration", |
295 | .data = &sysctl_ip_vs_lblcr_expiration, | |
296 | .maxlen = sizeof(int), | |
e905a9ed | 297 | .mode = 0644, |
1da177e4 LT |
298 | .proc_handler = &proc_dointvec_jiffies, |
299 | }, | |
300 | { .ctl_name = 0 } | |
301 | }; | |
302 | ||
1da177e4 LT |
303 | static struct ctl_table_header * sysctl_header; |
304 | ||
1da177e4 LT |
305 | static inline void ip_vs_lblcr_free(struct ip_vs_lblcr_entry *en) |
306 | { | |
307 | list_del(&en->list); | |
308 | ip_vs_dest_set_eraseall(&en->set); | |
309 | kfree(en); | |
310 | } | |
311 | ||
312 | ||
313 | /* | |
314 | * Returns hash value for IPVS LBLCR entry | |
315 | */ | |
44548375 JV |
316 | static inline unsigned |
317 | ip_vs_lblcr_hashkey(int af, const union nf_inet_addr *addr) | |
1da177e4 | 318 | { |
44548375 JV |
319 | __be32 addr_fold = addr->ip; |
320 | ||
321 | #ifdef CONFIG_IP_VS_IPV6 | |
322 | if (af == AF_INET6) | |
323 | addr_fold = addr->ip6[0]^addr->ip6[1]^ | |
324 | addr->ip6[2]^addr->ip6[3]; | |
325 | #endif | |
326 | return (ntohl(addr_fold)*2654435761UL) & IP_VS_LBLCR_TAB_MASK; | |
1da177e4 LT |
327 | } |
328 | ||
329 | ||
330 | /* | |
331 | * Hash an entry in the ip_vs_lblcr_table. | |
332 | * returns bool success. | |
333 | */ | |
f728bafb | 334 | static void |
1da177e4 LT |
335 | ip_vs_lblcr_hash(struct ip_vs_lblcr_table *tbl, struct ip_vs_lblcr_entry *en) |
336 | { | |
44548375 | 337 | unsigned hash = ip_vs_lblcr_hashkey(en->af, &en->addr); |
1da177e4 | 338 | |
1da177e4 LT |
339 | list_add(&en->list, &tbl->bucket[hash]); |
340 | atomic_inc(&tbl->entries); | |
1da177e4 LT |
341 | } |
342 | ||
343 | ||
1da177e4 | 344 | /* |
f728bafb SW |
345 | * Get ip_vs_lblcr_entry associated with supplied parameters. Called under |
346 | * read lock. | |
1da177e4 LT |
347 | */ |
348 | static inline struct ip_vs_lblcr_entry * | |
44548375 JV |
349 | ip_vs_lblcr_get(int af, struct ip_vs_lblcr_table *tbl, |
350 | const union nf_inet_addr *addr) | |
1da177e4 | 351 | { |
44548375 | 352 | unsigned hash = ip_vs_lblcr_hashkey(af, addr); |
1da177e4 LT |
353 | struct ip_vs_lblcr_entry *en; |
354 | ||
f728bafb | 355 | list_for_each_entry(en, &tbl->bucket[hash], list) |
44548375 | 356 | if (ip_vs_addr_equal(af, &en->addr, addr)) |
f728bafb SW |
357 | return en; |
358 | ||
359 | return NULL; | |
360 | } | |
1da177e4 | 361 | |
1da177e4 | 362 | |
f728bafb SW |
363 | /* |
364 | * Create or update an ip_vs_lblcr_entry, which is a mapping of a destination | |
365 | * IP address to a server. Called under write lock. | |
366 | */ | |
367 | static inline struct ip_vs_lblcr_entry * | |
44548375 | 368 | ip_vs_lblcr_new(struct ip_vs_lblcr_table *tbl, const union nf_inet_addr *daddr, |
f728bafb SW |
369 | struct ip_vs_dest *dest) |
370 | { | |
371 | struct ip_vs_lblcr_entry *en; | |
372 | ||
44548375 | 373 | en = ip_vs_lblcr_get(dest->af, tbl, daddr); |
f728bafb SW |
374 | if (!en) { |
375 | en = kmalloc(sizeof(*en), GFP_ATOMIC); | |
376 | if (!en) { | |
377 | IP_VS_ERR("ip_vs_lblcr_new(): no memory\n"); | |
378 | return NULL; | |
1da177e4 | 379 | } |
f728bafb | 380 | |
44548375 JV |
381 | en->af = dest->af; |
382 | ip_vs_addr_copy(dest->af, &en->addr, daddr); | |
f728bafb SW |
383 | en->lastuse = jiffies; |
384 | ||
385 | /* initilize its dest set */ | |
386 | atomic_set(&(en->set.size), 0); | |
387 | en->set.list = NULL; | |
388 | rwlock_init(&en->set.lock); | |
389 | ||
390 | ip_vs_lblcr_hash(tbl, en); | |
1da177e4 LT |
391 | } |
392 | ||
f728bafb SW |
393 | write_lock(&en->set.lock); |
394 | ip_vs_dest_set_insert(&en->set, dest); | |
395 | write_unlock(&en->set.lock); | |
1da177e4 | 396 | |
f728bafb | 397 | return en; |
1da177e4 LT |
398 | } |
399 | ||
400 | ||
401 | /* | |
402 | * Flush all the entries of the specified table. | |
403 | */ | |
404 | static void ip_vs_lblcr_flush(struct ip_vs_lblcr_table *tbl) | |
405 | { | |
406 | int i; | |
407 | struct ip_vs_lblcr_entry *en, *nxt; | |
408 | ||
f728bafb | 409 | /* No locking required, only called during cleanup. */ |
1da177e4 | 410 | for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) { |
1da177e4 LT |
411 | list_for_each_entry_safe(en, nxt, &tbl->bucket[i], list) { |
412 | ip_vs_lblcr_free(en); | |
1da177e4 | 413 | } |
1da177e4 LT |
414 | } |
415 | } | |
416 | ||
417 | ||
f728bafb | 418 | static inline void ip_vs_lblcr_full_check(struct ip_vs_service *svc) |
1da177e4 | 419 | { |
f728bafb | 420 | struct ip_vs_lblcr_table *tbl = svc->sched_data; |
1da177e4 LT |
421 | unsigned long now = jiffies; |
422 | int i, j; | |
423 | struct ip_vs_lblcr_entry *en, *nxt; | |
424 | ||
425 | for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) { | |
426 | j = (j + 1) & IP_VS_LBLCR_TAB_MASK; | |
427 | ||
f728bafb | 428 | write_lock(&svc->sched_lock); |
1da177e4 LT |
429 | list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) { |
430 | if (time_after(en->lastuse+sysctl_ip_vs_lblcr_expiration, | |
431 | now)) | |
432 | continue; | |
433 | ||
434 | ip_vs_lblcr_free(en); | |
435 | atomic_dec(&tbl->entries); | |
436 | } | |
f728bafb | 437 | write_unlock(&svc->sched_lock); |
1da177e4 LT |
438 | } |
439 | tbl->rover = j; | |
440 | } | |
441 | ||
442 | ||
443 | /* | |
444 | * Periodical timer handler for IPVS lblcr table | |
445 | * It is used to collect stale entries when the number of entries | |
446 | * exceeds the maximum size of the table. | |
447 | * | |
448 | * Fixme: we probably need more complicated algorithm to collect | |
449 | * entries that have not been used for a long time even | |
450 | * if the number of entries doesn't exceed the maximum size | |
451 | * of the table. | |
452 | * The full expiration check is for this purpose now. | |
453 | */ | |
454 | static void ip_vs_lblcr_check_expire(unsigned long data) | |
455 | { | |
f728bafb SW |
456 | struct ip_vs_service *svc = (struct ip_vs_service *) data; |
457 | struct ip_vs_lblcr_table *tbl = svc->sched_data; | |
1da177e4 LT |
458 | unsigned long now = jiffies; |
459 | int goal; | |
460 | int i, j; | |
461 | struct ip_vs_lblcr_entry *en, *nxt; | |
462 | ||
1da177e4 LT |
463 | if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) { |
464 | /* do full expiration check */ | |
f728bafb | 465 | ip_vs_lblcr_full_check(svc); |
1da177e4 LT |
466 | tbl->counter = 1; |
467 | goto out; | |
468 | } | |
469 | ||
470 | if (atomic_read(&tbl->entries) <= tbl->max_size) { | |
471 | tbl->counter++; | |
472 | goto out; | |
473 | } | |
474 | ||
475 | goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3; | |
476 | if (goal > tbl->max_size/2) | |
477 | goal = tbl->max_size/2; | |
478 | ||
479 | for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) { | |
480 | j = (j + 1) & IP_VS_LBLCR_TAB_MASK; | |
481 | ||
f728bafb | 482 | write_lock(&svc->sched_lock); |
1da177e4 LT |
483 | list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) { |
484 | if (time_before(now, en->lastuse+ENTRY_TIMEOUT)) | |
485 | continue; | |
486 | ||
487 | ip_vs_lblcr_free(en); | |
488 | atomic_dec(&tbl->entries); | |
489 | goal--; | |
490 | } | |
f728bafb | 491 | write_unlock(&svc->sched_lock); |
1da177e4 LT |
492 | if (goal <= 0) |
493 | break; | |
494 | } | |
495 | tbl->rover = j; | |
496 | ||
497 | out: | |
498 | mod_timer(&tbl->periodic_timer, jiffies+CHECK_EXPIRE_INTERVAL); | |
499 | } | |
500 | ||
1da177e4 LT |
501 | static int ip_vs_lblcr_init_svc(struct ip_vs_service *svc) |
502 | { | |
503 | int i; | |
504 | struct ip_vs_lblcr_table *tbl; | |
505 | ||
506 | /* | |
507 | * Allocate the ip_vs_lblcr_table for this service | |
508 | */ | |
f728bafb | 509 | tbl = kmalloc(sizeof(*tbl), GFP_ATOMIC); |
1da177e4 LT |
510 | if (tbl == NULL) { |
511 | IP_VS_ERR("ip_vs_lblcr_init_svc(): no memory\n"); | |
512 | return -ENOMEM; | |
513 | } | |
514 | svc->sched_data = tbl; | |
515 | IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) allocated for " | |
f728bafb | 516 | "current service\n", sizeof(*tbl)); |
1da177e4 LT |
517 | |
518 | /* | |
519 | * Initialize the hash buckets | |
520 | */ | |
521 | for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) { | |
522 | INIT_LIST_HEAD(&tbl->bucket[i]); | |
523 | } | |
1da177e4 LT |
524 | tbl->max_size = IP_VS_LBLCR_TAB_SIZE*16; |
525 | tbl->rover = 0; | |
526 | tbl->counter = 1; | |
527 | ||
528 | /* | |
529 | * Hook periodic timer for garbage collection | |
530 | */ | |
b24b8a24 | 531 | setup_timer(&tbl->periodic_timer, ip_vs_lblcr_check_expire, |
f728bafb SW |
532 | (unsigned long)svc); |
533 | mod_timer(&tbl->periodic_timer, jiffies + CHECK_EXPIRE_INTERVAL); | |
1da177e4 | 534 | |
1da177e4 LT |
535 | return 0; |
536 | } | |
537 | ||
538 | ||
539 | static int ip_vs_lblcr_done_svc(struct ip_vs_service *svc) | |
540 | { | |
541 | struct ip_vs_lblcr_table *tbl = svc->sched_data; | |
542 | ||
543 | /* remove periodic timer */ | |
544 | del_timer_sync(&tbl->periodic_timer); | |
545 | ||
546 | /* got to clean up table entries here */ | |
547 | ip_vs_lblcr_flush(tbl); | |
548 | ||
549 | /* release the table itself */ | |
f728bafb | 550 | kfree(tbl); |
1da177e4 | 551 | IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) released\n", |
f728bafb | 552 | sizeof(*tbl)); |
1da177e4 LT |
553 | |
554 | return 0; | |
555 | } | |
556 | ||
557 | ||
1da177e4 | 558 | static inline struct ip_vs_dest * |
44548375 | 559 | __ip_vs_lblcr_schedule(struct ip_vs_service *svc) |
1da177e4 LT |
560 | { |
561 | struct ip_vs_dest *dest, *least; | |
562 | int loh, doh; | |
563 | ||
564 | /* | |
565 | * We think the overhead of processing active connections is fifty | |
566 | * times higher than that of inactive connections in average. (This | |
567 | * fifty times might not be accurate, we will change it later.) We | |
568 | * use the following formula to estimate the overhead: | |
569 | * dest->activeconns*50 + dest->inactconns | |
570 | * and the load: | |
571 | * (dest overhead) / dest->weight | |
572 | * | |
573 | * Remember -- no floats in kernel mode!!! | |
574 | * The comparison of h1*w2 > h2*w1 is equivalent to that of | |
575 | * h1/w1 > h2/w2 | |
576 | * if every weight is larger than zero. | |
577 | * | |
578 | * The server with weight=0 is quiesced and will not receive any | |
579 | * new connection. | |
580 | */ | |
581 | list_for_each_entry(dest, &svc->destinations, n_list) { | |
582 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
583 | continue; | |
584 | ||
585 | if (atomic_read(&dest->weight) > 0) { | |
586 | least = dest; | |
587 | loh = atomic_read(&least->activeconns) * 50 | |
588 | + atomic_read(&least->inactconns); | |
589 | goto nextstage; | |
590 | } | |
591 | } | |
592 | return NULL; | |
593 | ||
594 | /* | |
595 | * Find the destination with the least load. | |
596 | */ | |
597 | nextstage: | |
598 | list_for_each_entry_continue(dest, &svc->destinations, n_list) { | |
599 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
600 | continue; | |
601 | ||
602 | doh = atomic_read(&dest->activeconns) * 50 | |
603 | + atomic_read(&dest->inactconns); | |
604 | if (loh * atomic_read(&dest->weight) > | |
605 | doh * atomic_read(&least->weight)) { | |
606 | least = dest; | |
607 | loh = doh; | |
608 | } | |
609 | } | |
610 | ||
44548375 JV |
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); | |
1da177e4 LT |
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 | |
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(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 * | |
648 | ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb) | |
649 | { | |
f728bafb | 650 | struct ip_vs_lblcr_table *tbl = svc->sched_data; |
44548375 | 651 | struct ip_vs_iphdr iph; |
f728bafb SW |
652 | struct ip_vs_dest *dest = NULL; |
653 | struct ip_vs_lblcr_entry *en; | |
1da177e4 | 654 | |
44548375 JV |
655 | ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph); |
656 | ||
1da177e4 LT |
657 | IP_VS_DBG(6, "ip_vs_lblcr_schedule(): Scheduling...\n"); |
658 | ||
f728bafb SW |
659 | /* First look in our cache */ |
660 | read_lock(&svc->sched_lock); | |
44548375 | 661 | en = ip_vs_lblcr_get(svc->af, tbl, &iph.daddr); |
f728bafb SW |
662 | if (en) { |
663 | /* We only hold a read lock, but this is atomic */ | |
664 | en->lastuse = jiffies; | |
665 | ||
666 | /* Get the least loaded destination */ | |
667 | read_lock(&en->set.lock); | |
1da177e4 | 668 | dest = ip_vs_dest_set_min(&en->set); |
f728bafb SW |
669 | read_unlock(&en->set.lock); |
670 | ||
671 | /* More than one destination + enough time passed by, cleanup */ | |
1da177e4 | 672 | if (atomic_read(&en->set.size) > 1 && |
f728bafb SW |
673 | time_after(jiffies, en->set.lastmod + |
674 | sysctl_ip_vs_lblcr_expiration)) { | |
1da177e4 | 675 | struct ip_vs_dest *m; |
f728bafb SW |
676 | |
677 | write_lock(&en->set.lock); | |
1da177e4 LT |
678 | m = ip_vs_dest_set_max(&en->set); |
679 | if (m) | |
680 | ip_vs_dest_set_erase(&en->set, m); | |
f728bafb SW |
681 | write_unlock(&en->set.lock); |
682 | } | |
683 | ||
684 | /* If the destination is not overloaded, use it */ | |
685 | if (dest && !is_overloaded(dest, svc)) { | |
686 | read_unlock(&svc->sched_lock); | |
687 | goto out; | |
688 | } | |
689 | ||
690 | /* The cache entry is invalid, time to schedule */ | |
44548375 | 691 | dest = __ip_vs_lblcr_schedule(svc); |
f728bafb SW |
692 | if (!dest) { |
693 | IP_VS_DBG(1, "no destination available\n"); | |
694 | read_unlock(&svc->sched_lock); | |
695 | return NULL; | |
1da177e4 | 696 | } |
f728bafb SW |
697 | |
698 | /* Update our cache entry */ | |
699 | write_lock(&en->set.lock); | |
700 | ip_vs_dest_set_insert(&en->set, dest); | |
701 | write_unlock(&en->set.lock); | |
702 | } | |
703 | read_unlock(&svc->sched_lock); | |
704 | ||
705 | if (dest) | |
706 | goto out; | |
707 | ||
708 | /* No cache entry, time to schedule */ | |
44548375 | 709 | dest = __ip_vs_lblcr_schedule(svc); |
f728bafb SW |
710 | if (!dest) { |
711 | IP_VS_DBG(1, "no destination available\n"); | |
712 | return NULL; | |
1da177e4 | 713 | } |
1da177e4 | 714 | |
f728bafb SW |
715 | /* If we fail to create a cache entry, we'll just use the valid dest */ |
716 | write_lock(&svc->sched_lock); | |
44548375 | 717 | ip_vs_lblcr_new(tbl, &iph.daddr, dest); |
f728bafb SW |
718 | write_unlock(&svc->sched_lock); |
719 | ||
720 | out: | |
44548375 JV |
721 | IP_VS_DBG_BUF(6, "LBLCR: destination IP address %s --> server %s:%d\n", |
722 | IP_VS_DBG_ADDR(svc->af, &iph.daddr), | |
723 | IP_VS_DBG_ADDR(svc->af, &dest->addr), ntohs(dest->port)); | |
1da177e4 LT |
724 | |
725 | return dest; | |
726 | } | |
727 | ||
728 | ||
729 | /* | |
730 | * IPVS LBLCR Scheduler structure | |
731 | */ | |
732 | static struct ip_vs_scheduler ip_vs_lblcr_scheduler = | |
733 | { | |
734 | .name = "lblcr", | |
735 | .refcnt = ATOMIC_INIT(0), | |
736 | .module = THIS_MODULE, | |
d149ccc9 | 737 | .n_list = LIST_HEAD_INIT(ip_vs_lblcr_scheduler.n_list), |
1da177e4 LT |
738 | .init_service = ip_vs_lblcr_init_svc, |
739 | .done_service = ip_vs_lblcr_done_svc, | |
1da177e4 LT |
740 | .schedule = ip_vs_lblcr_schedule, |
741 | }; | |
742 | ||
743 | ||
744 | static int __init ip_vs_lblcr_init(void) | |
745 | { | |
a014bc8f PE |
746 | int ret; |
747 | ||
90754f8e | 748 | sysctl_header = register_sysctl_paths(net_vs_ctl_path, vs_vars_table); |
a014bc8f PE |
749 | ret = register_ip_vs_scheduler(&ip_vs_lblcr_scheduler); |
750 | if (ret) | |
751 | unregister_sysctl_table(sysctl_header); | |
752 | return ret; | |
1da177e4 LT |
753 | } |
754 | ||
755 | ||
756 | static void __exit ip_vs_lblcr_cleanup(void) | |
757 | { | |
1da177e4 LT |
758 | unregister_sysctl_table(sysctl_header); |
759 | unregister_ip_vs_scheduler(&ip_vs_lblcr_scheduler); | |
760 | } | |
761 | ||
762 | ||
763 | module_init(ip_vs_lblcr_init); | |
764 | module_exit(ip_vs_lblcr_cleanup); | |
765 | MODULE_LICENSE("GPL"); |