diff options
author | Julius Volz <juliusv@google.com> | 2008-09-19 06:32:57 -0400 |
---|---|---|
committer | Simon Horman <horms@verge.net.au> | 2008-10-06 17:38:24 -0400 |
commit | cb7f6a7b716e801097b564dec3ccb58d330aef56 (patch) | |
tree | 92fa8fa5381e04576c43eab88874ab54ea670767 /net/ipv4/ipvs/ip_vs_lblcr.c | |
parent | 8d5803bf6fbe5264000afc8c34bff08e8ecc023b (diff) |
IPVS: Move IPVS to net/netfilter/ipvs
Since IPVS now has partial IPv6 support, this patch moves IPVS from
net/ipv4/ipvs to net/netfilter/ipvs. It's a result of:
$ git mv net/ipv4/ipvs net/netfilter
and adapting the relevant Kconfigs/Makefiles to the new path.
Signed-off-by: Julius Volz <juliusv@google.com>
Signed-off-by: Simon Horman <horms@verge.net.au>
Diffstat (limited to 'net/ipv4/ipvs/ip_vs_lblcr.c')
-rw-r--r-- | net/ipv4/ipvs/ip_vs_lblcr.c | 755 |
1 files changed, 0 insertions, 755 deletions
diff --git a/net/ipv4/ipvs/ip_vs_lblcr.c b/net/ipv4/ipvs/ip_vs_lblcr.c deleted file mode 100644 index 1f75ea83bcf8..000000000000 --- a/net/ipv4/ipvs/ip_vs_lblcr.c +++ /dev/null | |||
@@ -1,755 +0,0 @@ | |||
1 | /* | ||
2 | * IPVS: Locality-Based Least-Connection with Replication scheduler | ||
3 | * | ||
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 | |||
40 | #include <linux/ip.h> | ||
41 | #include <linux/module.h> | ||
42 | #include <linux/kernel.h> | ||
43 | #include <linux/skbuff.h> | ||
44 | #include <linux/jiffies.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 | /* | ||
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 | |||
109 | e = kmalloc(sizeof(*e), GFP_ATOMIC); | ||
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 */ | ||
119 | e->next = set->list; | ||
120 | set->list = e; | ||
121 | atomic_inc(&set->size); | ||
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 | |||
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 | } | ||
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 | |||
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 | } | ||
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 | } | ||
204 | |||
205 | IP_VS_DBG(6, "ip_vs_dest_set_min: server %d.%d.%d.%d:%d " | ||
206 | "activeconns %d refcnt %d weight %d overhead %d\n", | ||
207 | NIPQUAD(least->addr.ip), ntohs(least->port), | ||
208 | atomic_read(&least->activeconns), | ||
209 | atomic_read(&least->refcnt), | ||
210 | atomic_read(&least->weight), loh); | ||
211 | return least; | ||
212 | } | ||
213 | |||
214 | |||
215 | /* get weighted most-connection node in the destination set */ | ||
216 | static inline struct ip_vs_dest *ip_vs_dest_set_max(struct ip_vs_dest_set *set) | ||
217 | { | ||
218 | register struct ip_vs_dest_list *e; | ||
219 | struct ip_vs_dest *dest, *most; | ||
220 | int moh, doh; | ||
221 | |||
222 | if (set == NULL) | ||
223 | return NULL; | ||
224 | |||
225 | /* select the first destination server, whose weight > 0 */ | ||
226 | for (e=set->list; e!=NULL; e=e->next) { | ||
227 | most = e->dest; | ||
228 | if (atomic_read(&most->weight) > 0) { | ||
229 | moh = atomic_read(&most->activeconns) * 50 | ||
230 | + atomic_read(&most->inactconns); | ||
231 | goto nextstage; | ||
232 | } | ||
233 | } | ||
234 | return NULL; | ||
235 | |||
236 | /* find the destination with the weighted most load */ | ||
237 | nextstage: | ||
238 | for (e=e->next; e!=NULL; e=e->next) { | ||
239 | dest = e->dest; | ||
240 | doh = atomic_read(&dest->activeconns) * 50 | ||
241 | + atomic_read(&dest->inactconns); | ||
242 | /* moh/mw < doh/dw ==> moh*dw < doh*mw, where mw,dw>0 */ | ||
243 | if ((moh * atomic_read(&dest->weight) < | ||
244 | doh * atomic_read(&most->weight)) | ||
245 | && (atomic_read(&dest->weight) > 0)) { | ||
246 | most = dest; | ||
247 | moh = doh; | ||
248 | } | ||
249 | } | ||
250 | |||
251 | IP_VS_DBG(6, "ip_vs_dest_set_max: server %d.%d.%d.%d:%d " | ||
252 | "activeconns %d refcnt %d weight %d overhead %d\n", | ||
253 | NIPQUAD(most->addr.ip), ntohs(most->port), | ||
254 | atomic_read(&most->activeconns), | ||
255 | atomic_read(&most->refcnt), | ||
256 | atomic_read(&most->weight), moh); | ||
257 | return most; | ||
258 | } | ||
259 | |||
260 | |||
261 | /* | ||
262 | * IPVS lblcr entry represents an association between destination | ||
263 | * IP address and its destination server set | ||
264 | */ | ||
265 | struct ip_vs_lblcr_entry { | ||
266 | struct list_head list; | ||
267 | __be32 addr; /* destination IP address */ | ||
268 | struct ip_vs_dest_set set; /* destination server set */ | ||
269 | unsigned long lastuse; /* last used time */ | ||
270 | }; | ||
271 | |||
272 | |||
273 | /* | ||
274 | * IPVS lblcr hash table | ||
275 | */ | ||
276 | struct ip_vs_lblcr_table { | ||
277 | struct list_head bucket[IP_VS_LBLCR_TAB_SIZE]; /* hash bucket */ | ||
278 | atomic_t entries; /* number of entries */ | ||
279 | int max_size; /* maximum size of entries */ | ||
280 | struct timer_list periodic_timer; /* collect stale entries */ | ||
281 | int rover; /* rover for expire check */ | ||
282 | int counter; /* counter for no expire */ | ||
283 | }; | ||
284 | |||
285 | |||
286 | /* | ||
287 | * IPVS LBLCR sysctl table | ||
288 | */ | ||
289 | |||
290 | static ctl_table vs_vars_table[] = { | ||
291 | { | ||
292 | .procname = "lblcr_expiration", | ||
293 | .data = &sysctl_ip_vs_lblcr_expiration, | ||
294 | .maxlen = sizeof(int), | ||
295 | .mode = 0644, | ||
296 | .proc_handler = &proc_dointvec_jiffies, | ||
297 | }, | ||
298 | { .ctl_name = 0 } | ||
299 | }; | ||
300 | |||
301 | static struct ctl_table_header * sysctl_header; | ||
302 | |||
303 | static inline void ip_vs_lblcr_free(struct ip_vs_lblcr_entry *en) | ||
304 | { | ||
305 | list_del(&en->list); | ||
306 | ip_vs_dest_set_eraseall(&en->set); | ||
307 | kfree(en); | ||
308 | } | ||
309 | |||
310 | |||
311 | /* | ||
312 | * Returns hash value for IPVS LBLCR entry | ||
313 | */ | ||
314 | static inline unsigned ip_vs_lblcr_hashkey(__be32 addr) | ||
315 | { | ||
316 | return (ntohl(addr)*2654435761UL) & IP_VS_LBLCR_TAB_MASK; | ||
317 | } | ||
318 | |||
319 | |||
320 | /* | ||
321 | * Hash an entry in the ip_vs_lblcr_table. | ||
322 | * returns bool success. | ||
323 | */ | ||
324 | static void | ||
325 | ip_vs_lblcr_hash(struct ip_vs_lblcr_table *tbl, struct ip_vs_lblcr_entry *en) | ||
326 | { | ||
327 | unsigned hash = ip_vs_lblcr_hashkey(en->addr); | ||
328 | |||
329 | list_add(&en->list, &tbl->bucket[hash]); | ||
330 | atomic_inc(&tbl->entries); | ||
331 | } | ||
332 | |||
333 | |||
334 | /* | ||
335 | * Get ip_vs_lblcr_entry associated with supplied parameters. Called under | ||
336 | * read lock. | ||
337 | */ | ||
338 | static inline struct ip_vs_lblcr_entry * | ||
339 | ip_vs_lblcr_get(struct ip_vs_lblcr_table *tbl, __be32 addr) | ||
340 | { | ||
341 | unsigned hash = ip_vs_lblcr_hashkey(addr); | ||
342 | struct ip_vs_lblcr_entry *en; | ||
343 | |||
344 | list_for_each_entry(en, &tbl->bucket[hash], list) | ||
345 | if (en->addr == addr) | ||
346 | return en; | ||
347 | |||
348 | return NULL; | ||
349 | } | ||
350 | |||
351 | |||
352 | /* | ||
353 | * Create or update an ip_vs_lblcr_entry, which is a mapping of a destination | ||
354 | * IP address to a server. Called under write lock. | ||
355 | */ | ||
356 | static inline struct ip_vs_lblcr_entry * | ||
357 | ip_vs_lblcr_new(struct ip_vs_lblcr_table *tbl, __be32 daddr, | ||
358 | struct ip_vs_dest *dest) | ||
359 | { | ||
360 | struct ip_vs_lblcr_entry *en; | ||
361 | |||
362 | en = ip_vs_lblcr_get(tbl, daddr); | ||
363 | if (!en) { | ||
364 | en = kmalloc(sizeof(*en), GFP_ATOMIC); | ||
365 | if (!en) { | ||
366 | IP_VS_ERR("ip_vs_lblcr_new(): no memory\n"); | ||
367 | return NULL; | ||
368 | } | ||
369 | |||
370 | en->addr = daddr; | ||
371 | en->lastuse = jiffies; | ||
372 | |||
373 | /* initilize its dest set */ | ||
374 | atomic_set(&(en->set.size), 0); | ||
375 | en->set.list = NULL; | ||
376 | rwlock_init(&en->set.lock); | ||
377 | |||
378 | ip_vs_lblcr_hash(tbl, en); | ||
379 | } | ||
380 | |||
381 | write_lock(&en->set.lock); | ||
382 | ip_vs_dest_set_insert(&en->set, dest); | ||
383 | write_unlock(&en->set.lock); | ||
384 | |||
385 | return en; | ||
386 | } | ||
387 | |||
388 | |||
389 | /* | ||
390 | * Flush all the entries of the specified table. | ||
391 | */ | ||
392 | static void ip_vs_lblcr_flush(struct ip_vs_lblcr_table *tbl) | ||
393 | { | ||
394 | int i; | ||
395 | struct ip_vs_lblcr_entry *en, *nxt; | ||
396 | |||
397 | /* No locking required, only called during cleanup. */ | ||
398 | for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) { | ||
399 | list_for_each_entry_safe(en, nxt, &tbl->bucket[i], list) { | ||
400 | ip_vs_lblcr_free(en); | ||
401 | } | ||
402 | } | ||
403 | } | ||
404 | |||
405 | |||
406 | static inline void ip_vs_lblcr_full_check(struct ip_vs_service *svc) | ||
407 | { | ||
408 | struct ip_vs_lblcr_table *tbl = svc->sched_data; | ||
409 | unsigned long now = jiffies; | ||
410 | int i, j; | ||
411 | struct ip_vs_lblcr_entry *en, *nxt; | ||
412 | |||
413 | for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) { | ||
414 | j = (j + 1) & IP_VS_LBLCR_TAB_MASK; | ||
415 | |||
416 | write_lock(&svc->sched_lock); | ||
417 | list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) { | ||
418 | if (time_after(en->lastuse+sysctl_ip_vs_lblcr_expiration, | ||
419 | now)) | ||
420 | continue; | ||
421 | |||
422 | ip_vs_lblcr_free(en); | ||
423 | atomic_dec(&tbl->entries); | ||
424 | } | ||
425 | write_unlock(&svc->sched_lock); | ||
426 | } | ||
427 | tbl->rover = j; | ||
428 | } | ||
429 | |||
430 | |||
431 | /* | ||
432 | * Periodical timer handler for IPVS lblcr table | ||
433 | * It is used to collect stale entries when the number of entries | ||
434 | * exceeds the maximum size of the table. | ||
435 | * | ||
436 | * Fixme: we probably need more complicated algorithm to collect | ||
437 | * entries that have not been used for a long time even | ||
438 | * if the number of entries doesn't exceed the maximum size | ||
439 | * of the table. | ||
440 | * The full expiration check is for this purpose now. | ||
441 | */ | ||
442 | static void ip_vs_lblcr_check_expire(unsigned long data) | ||
443 | { | ||
444 | struct ip_vs_service *svc = (struct ip_vs_service *) data; | ||
445 | struct ip_vs_lblcr_table *tbl = svc->sched_data; | ||
446 | unsigned long now = jiffies; | ||
447 | int goal; | ||
448 | int i, j; | ||
449 | struct ip_vs_lblcr_entry *en, *nxt; | ||
450 | |||
451 | if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) { | ||
452 | /* do full expiration check */ | ||
453 | ip_vs_lblcr_full_check(svc); | ||
454 | tbl->counter = 1; | ||
455 | goto out; | ||
456 | } | ||
457 | |||
458 | if (atomic_read(&tbl->entries) <= tbl->max_size) { | ||
459 | tbl->counter++; | ||
460 | goto out; | ||
461 | } | ||
462 | |||
463 | goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3; | ||
464 | if (goal > tbl->max_size/2) | ||
465 | goal = tbl->max_size/2; | ||
466 | |||
467 | for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) { | ||
468 | j = (j + 1) & IP_VS_LBLCR_TAB_MASK; | ||
469 | |||
470 | write_lock(&svc->sched_lock); | ||
471 | list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) { | ||
472 | if (time_before(now, en->lastuse+ENTRY_TIMEOUT)) | ||
473 | continue; | ||
474 | |||
475 | ip_vs_lblcr_free(en); | ||
476 | atomic_dec(&tbl->entries); | ||
477 | goal--; | ||
478 | } | ||
479 | write_unlock(&svc->sched_lock); | ||
480 | if (goal <= 0) | ||
481 | break; | ||
482 | } | ||
483 | tbl->rover = j; | ||
484 | |||
485 | out: | ||
486 | mod_timer(&tbl->periodic_timer, jiffies+CHECK_EXPIRE_INTERVAL); | ||
487 | } | ||
488 | |||
489 | static int ip_vs_lblcr_init_svc(struct ip_vs_service *svc) | ||
490 | { | ||
491 | int i; | ||
492 | struct ip_vs_lblcr_table *tbl; | ||
493 | |||
494 | /* | ||
495 | * Allocate the ip_vs_lblcr_table for this service | ||
496 | */ | ||
497 | tbl = kmalloc(sizeof(*tbl), GFP_ATOMIC); | ||
498 | if (tbl == NULL) { | ||
499 | IP_VS_ERR("ip_vs_lblcr_init_svc(): no memory\n"); | ||
500 | return -ENOMEM; | ||
501 | } | ||
502 | svc->sched_data = tbl; | ||
503 | IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) allocated for " | ||
504 | "current service\n", sizeof(*tbl)); | ||
505 | |||
506 | /* | ||
507 | * Initialize the hash buckets | ||
508 | */ | ||
509 | for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) { | ||
510 | INIT_LIST_HEAD(&tbl->bucket[i]); | ||
511 | } | ||
512 | tbl->max_size = IP_VS_LBLCR_TAB_SIZE*16; | ||
513 | tbl->rover = 0; | ||
514 | tbl->counter = 1; | ||
515 | |||
516 | /* | ||
517 | * Hook periodic timer for garbage collection | ||
518 | */ | ||
519 | setup_timer(&tbl->periodic_timer, ip_vs_lblcr_check_expire, | ||
520 | (unsigned long)svc); | ||
521 | mod_timer(&tbl->periodic_timer, jiffies + CHECK_EXPIRE_INTERVAL); | ||
522 | |||
523 | return 0; | ||
524 | } | ||
525 | |||
526 | |||
527 | static int ip_vs_lblcr_done_svc(struct ip_vs_service *svc) | ||
528 | { | ||
529 | struct ip_vs_lblcr_table *tbl = svc->sched_data; | ||
530 | |||
531 | /* remove periodic timer */ | ||
532 | del_timer_sync(&tbl->periodic_timer); | ||
533 | |||
534 | /* got to clean up table entries here */ | ||
535 | ip_vs_lblcr_flush(tbl); | ||
536 | |||
537 | /* release the table itself */ | ||
538 | kfree(tbl); | ||
539 | IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) released\n", | ||
540 | sizeof(*tbl)); | ||
541 | |||
542 | return 0; | ||
543 | } | ||
544 | |||
545 | |||
546 | static inline struct ip_vs_dest * | ||
547 | __ip_vs_lblcr_schedule(struct ip_vs_service *svc, struct iphdr *iph) | ||
548 | { | ||
549 | struct ip_vs_dest *dest, *least; | ||
550 | int loh, doh; | ||
551 | |||
552 | /* | ||
553 | * We think the overhead of processing active connections is fifty | ||
554 | * times higher than that of inactive connections in average. (This | ||
555 | * fifty times might not be accurate, we will change it later.) We | ||
556 | * use the following formula to estimate the overhead: | ||
557 | * dest->activeconns*50 + dest->inactconns | ||
558 | * and the load: | ||
559 | * (dest overhead) / dest->weight | ||
560 | * | ||
561 | * Remember -- no floats in kernel mode!!! | ||
562 | * The comparison of h1*w2 > h2*w1 is equivalent to that of | ||
563 | * h1/w1 > h2/w2 | ||
564 | * if every weight is larger than zero. | ||
565 | * | ||
566 | * The server with weight=0 is quiesced and will not receive any | ||
567 | * new connection. | ||
568 | */ | ||
569 | list_for_each_entry(dest, &svc->destinations, n_list) { | ||
570 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | ||
571 | continue; | ||
572 | |||
573 | if (atomic_read(&dest->weight) > 0) { | ||
574 | least = dest; | ||
575 | loh = atomic_read(&least->activeconns) * 50 | ||
576 | + atomic_read(&least->inactconns); | ||
577 | goto nextstage; | ||
578 | } | ||
579 | } | ||
580 | return NULL; | ||
581 | |||
582 | /* | ||
583 | * Find the destination with the least load. | ||
584 | */ | ||
585 | nextstage: | ||
586 | list_for_each_entry_continue(dest, &svc->destinations, n_list) { | ||
587 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | ||
588 | continue; | ||
589 | |||
590 | doh = atomic_read(&dest->activeconns) * 50 | ||
591 | + atomic_read(&dest->inactconns); | ||
592 | if (loh * atomic_read(&dest->weight) > | ||
593 | doh * atomic_read(&least->weight)) { | ||
594 | least = dest; | ||
595 | loh = doh; | ||
596 | } | ||
597 | } | ||
598 | |||
599 | IP_VS_DBG(6, "LBLCR: server %d.%d.%d.%d:%d " | ||
600 | "activeconns %d refcnt %d weight %d overhead %d\n", | ||
601 | NIPQUAD(least->addr.ip), ntohs(least->port), | ||
602 | atomic_read(&least->activeconns), | ||
603 | atomic_read(&least->refcnt), | ||
604 | atomic_read(&least->weight), loh); | ||
605 | |||
606 | return least; | ||
607 | } | ||
608 | |||
609 | |||
610 | /* | ||
611 | * If this destination server is overloaded and there is a less loaded | ||
612 | * server, then return true. | ||
613 | */ | ||
614 | static inline int | ||
615 | is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc) | ||
616 | { | ||
617 | if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) { | ||
618 | struct ip_vs_dest *d; | ||
619 | |||
620 | list_for_each_entry(d, &svc->destinations, n_list) { | ||
621 | if (atomic_read(&d->activeconns)*2 | ||
622 | < atomic_read(&d->weight)) { | ||
623 | return 1; | ||
624 | } | ||
625 | } | ||
626 | } | ||
627 | return 0; | ||
628 | } | ||
629 | |||
630 | |||
631 | /* | ||
632 | * Locality-Based (weighted) Least-Connection scheduling | ||
633 | */ | ||
634 | static struct ip_vs_dest * | ||
635 | ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb) | ||
636 | { | ||
637 | struct ip_vs_lblcr_table *tbl = svc->sched_data; | ||
638 | struct iphdr *iph = ip_hdr(skb); | ||
639 | struct ip_vs_dest *dest = NULL; | ||
640 | struct ip_vs_lblcr_entry *en; | ||
641 | |||
642 | IP_VS_DBG(6, "ip_vs_lblcr_schedule(): Scheduling...\n"); | ||
643 | |||
644 | /* First look in our cache */ | ||
645 | read_lock(&svc->sched_lock); | ||
646 | en = ip_vs_lblcr_get(tbl, iph->daddr); | ||
647 | if (en) { | ||
648 | /* We only hold a read lock, but this is atomic */ | ||
649 | en->lastuse = jiffies; | ||
650 | |||
651 | /* Get the least loaded destination */ | ||
652 | read_lock(&en->set.lock); | ||
653 | dest = ip_vs_dest_set_min(&en->set); | ||
654 | read_unlock(&en->set.lock); | ||
655 | |||
656 | /* More than one destination + enough time passed by, cleanup */ | ||
657 | if (atomic_read(&en->set.size) > 1 && | ||
658 | time_after(jiffies, en->set.lastmod + | ||
659 | sysctl_ip_vs_lblcr_expiration)) { | ||
660 | struct ip_vs_dest *m; | ||
661 | |||
662 | write_lock(&en->set.lock); | ||
663 | m = ip_vs_dest_set_max(&en->set); | ||
664 | if (m) | ||
665 | ip_vs_dest_set_erase(&en->set, m); | ||
666 | write_unlock(&en->set.lock); | ||
667 | } | ||
668 | |||
669 | /* If the destination is not overloaded, use it */ | ||
670 | if (dest && !is_overloaded(dest, svc)) { | ||
671 | read_unlock(&svc->sched_lock); | ||
672 | goto out; | ||
673 | } | ||
674 | |||
675 | /* The cache entry is invalid, time to schedule */ | ||
676 | dest = __ip_vs_lblcr_schedule(svc, iph); | ||
677 | if (!dest) { | ||
678 | IP_VS_DBG(1, "no destination available\n"); | ||
679 | read_unlock(&svc->sched_lock); | ||
680 | return NULL; | ||
681 | } | ||
682 | |||
683 | /* Update our cache entry */ | ||
684 | write_lock(&en->set.lock); | ||
685 | ip_vs_dest_set_insert(&en->set, dest); | ||
686 | write_unlock(&en->set.lock); | ||
687 | } | ||
688 | read_unlock(&svc->sched_lock); | ||
689 | |||
690 | if (dest) | ||
691 | goto out; | ||
692 | |||
693 | /* No cache entry, time to schedule */ | ||
694 | dest = __ip_vs_lblcr_schedule(svc, iph); | ||
695 | if (!dest) { | ||
696 | IP_VS_DBG(1, "no destination available\n"); | ||
697 | return NULL; | ||
698 | } | ||
699 | |||
700 | /* If we fail to create a cache entry, we'll just use the valid dest */ | ||
701 | write_lock(&svc->sched_lock); | ||
702 | ip_vs_lblcr_new(tbl, iph->daddr, dest); | ||
703 | write_unlock(&svc->sched_lock); | ||
704 | |||
705 | out: | ||
706 | IP_VS_DBG(6, "LBLCR: destination IP address %u.%u.%u.%u " | ||
707 | "--> server %u.%u.%u.%u:%d\n", | ||
708 | NIPQUAD(iph->daddr), | ||
709 | NIPQUAD(dest->addr.ip), | ||
710 | ntohs(dest->port)); | ||
711 | |||
712 | return dest; | ||
713 | } | ||
714 | |||
715 | |||
716 | /* | ||
717 | * IPVS LBLCR Scheduler structure | ||
718 | */ | ||
719 | static struct ip_vs_scheduler ip_vs_lblcr_scheduler = | ||
720 | { | ||
721 | .name = "lblcr", | ||
722 | .refcnt = ATOMIC_INIT(0), | ||
723 | .module = THIS_MODULE, | ||
724 | .n_list = LIST_HEAD_INIT(ip_vs_lblcr_scheduler.n_list), | ||
725 | #ifdef CONFIG_IP_VS_IPV6 | ||
726 | .supports_ipv6 = 0, | ||
727 | #endif | ||
728 | .init_service = ip_vs_lblcr_init_svc, | ||
729 | .done_service = ip_vs_lblcr_done_svc, | ||
730 | .schedule = ip_vs_lblcr_schedule, | ||
731 | }; | ||
732 | |||
733 | |||
734 | static int __init ip_vs_lblcr_init(void) | ||
735 | { | ||
736 | int ret; | ||
737 | |||
738 | sysctl_header = register_sysctl_paths(net_vs_ctl_path, vs_vars_table); | ||
739 | ret = register_ip_vs_scheduler(&ip_vs_lblcr_scheduler); | ||
740 | if (ret) | ||
741 | unregister_sysctl_table(sysctl_header); | ||
742 | return ret; | ||
743 | } | ||
744 | |||
745 | |||
746 | static void __exit ip_vs_lblcr_cleanup(void) | ||
747 | { | ||
748 | unregister_sysctl_table(sysctl_header); | ||
749 | unregister_ip_vs_scheduler(&ip_vs_lblcr_scheduler); | ||
750 | } | ||
751 | |||
752 | |||
753 | module_init(ip_vs_lblcr_init); | ||
754 | module_exit(ip_vs_lblcr_cleanup); | ||
755 | MODULE_LICENSE("GPL"); | ||