]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * IPVS: Locality-Based Least-Connection with Replication scheduler | |
3 | * | |
4 | * Version: $Id: ip_vs_lblcr.c,v 1.11 2002/09/15 08:14:08 wensong Exp $ | |
5 | * | |
6 | * Authors: Wensong Zhang <wensong@gnuchina.org> | |
7 | * | |
8 | * This program is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU General Public License | |
10 | * as published by the Free Software Foundation; either version | |
11 | * 2 of the License, or (at your option) any later version. | |
12 | * | |
13 | * Changes: | |
14 | * Julian Anastasov : Added the missing (dest->weight>0) | |
15 | * condition in the ip_vs_dest_set_max. | |
16 | * | |
17 | */ | |
18 | ||
19 | /* | |
20 | * The lblc/r algorithm is as follows (pseudo code): | |
21 | * | |
22 | * if serverSet[dest_ip] is null then | |
23 | * n, serverSet[dest_ip] <- {weighted least-conn node}; | |
24 | * else | |
25 | * n <- {least-conn (alive) node in serverSet[dest_ip]}; | |
26 | * if (n is null) OR | |
27 | * (n.conns>n.weight AND | |
28 | * there is a node m with m.conns<m.weight/2) then | |
29 | * n <- {weighted least-conn node}; | |
30 | * add n to serverSet[dest_ip]; | |
31 | * if |serverSet[dest_ip]| > 1 AND | |
32 | * now - serverSet[dest_ip].lastMod > T then | |
33 | * m <- {most conn node in serverSet[dest_ip]}; | |
34 | * remove m from serverSet[dest_ip]; | |
35 | * if serverSet[dest_ip] changed then | |
36 | * serverSet[dest_ip].lastMod <- now; | |
37 | * | |
38 | * return n; | |
39 | * | |
40 | */ | |
41 | ||
14c85021 | 42 | #include <linux/ip.h> |
1da177e4 LT |
43 | #include <linux/module.h> |
44 | #include <linux/kernel.h> | |
14c85021 | 45 | #include <linux/skbuff.h> |
d7fe0f24 | 46 | #include <linux/jiffies.h> |
1da177e4 LT |
47 | |
48 | /* for sysctl */ | |
49 | #include <linux/fs.h> | |
50 | #include <linux/sysctl.h> | |
51 | /* for proc_net_create/proc_net_remove */ | |
52 | #include <linux/proc_fs.h> | |
53 | ||
54 | #include <net/ip_vs.h> | |
55 | ||
56 | ||
57 | /* | |
58 | * It is for garbage collection of stale IPVS lblcr entries, | |
59 | * when the table is full. | |
60 | */ | |
61 | #define CHECK_EXPIRE_INTERVAL (60*HZ) | |
62 | #define ENTRY_TIMEOUT (6*60*HZ) | |
63 | ||
64 | /* | |
65 | * It is for full expiration check. | |
66 | * When there is no partial expiration check (garbage collection) | |
67 | * in a half hour, do a full expiration check to collect stale | |
68 | * entries that haven't been touched for a day. | |
69 | */ | |
70 | #define COUNT_FOR_FULL_EXPIRATION 30 | |
71 | static int sysctl_ip_vs_lblcr_expiration = 24*60*60*HZ; | |
72 | ||
73 | ||
74 | /* | |
75 | * for IPVS lblcr entry hash table | |
76 | */ | |
77 | #ifndef CONFIG_IP_VS_LBLCR_TAB_BITS | |
78 | #define CONFIG_IP_VS_LBLCR_TAB_BITS 10 | |
79 | #endif | |
80 | #define IP_VS_LBLCR_TAB_BITS CONFIG_IP_VS_LBLCR_TAB_BITS | |
81 | #define IP_VS_LBLCR_TAB_SIZE (1 << IP_VS_LBLCR_TAB_BITS) | |
82 | #define IP_VS_LBLCR_TAB_MASK (IP_VS_LBLCR_TAB_SIZE - 1) | |
83 | ||
84 | ||
85 | /* | |
86 | * IPVS destination set structure and operations | |
87 | */ | |
88 | struct ip_vs_dest_list { | |
89 | struct ip_vs_dest_list *next; /* list link */ | |
90 | struct ip_vs_dest *dest; /* destination server */ | |
91 | }; | |
92 | ||
93 | struct ip_vs_dest_set { | |
94 | atomic_t size; /* set size */ | |
95 | unsigned long lastmod; /* last modified time */ | |
96 | struct ip_vs_dest_list *list; /* destination list */ | |
97 | rwlock_t lock; /* lock for this list */ | |
98 | }; | |
99 | ||
100 | ||
101 | static struct ip_vs_dest_list * | |
102 | ip_vs_dest_set_insert(struct ip_vs_dest_set *set, struct ip_vs_dest *dest) | |
103 | { | |
104 | struct ip_vs_dest_list *e; | |
105 | ||
106 | for (e=set->list; e!=NULL; e=e->next) { | |
107 | if (e->dest == dest) | |
108 | /* already existed */ | |
109 | return NULL; | |
110 | } | |
111 | ||
112 | e = kmalloc(sizeof(struct ip_vs_dest_list), GFP_ATOMIC); | |
113 | if (e == NULL) { | |
114 | IP_VS_ERR("ip_vs_dest_set_insert(): no memory\n"); | |
115 | return NULL; | |
116 | } | |
117 | ||
118 | atomic_inc(&dest->refcnt); | |
119 | e->dest = dest; | |
120 | ||
121 | /* link it to the list */ | |
122 | write_lock(&set->lock); | |
123 | e->next = set->list; | |
124 | set->list = e; | |
125 | atomic_inc(&set->size); | |
126 | write_unlock(&set->lock); | |
127 | ||
128 | set->lastmod = jiffies; | |
129 | return e; | |
130 | } | |
131 | ||
132 | static void | |
133 | ip_vs_dest_set_erase(struct ip_vs_dest_set *set, struct ip_vs_dest *dest) | |
134 | { | |
135 | struct ip_vs_dest_list *e, **ep; | |
136 | ||
137 | write_lock(&set->lock); | |
138 | for (ep=&set->list, e=*ep; e!=NULL; e=*ep) { | |
139 | if (e->dest == dest) { | |
140 | /* HIT */ | |
141 | *ep = e->next; | |
142 | atomic_dec(&set->size); | |
143 | set->lastmod = jiffies; | |
144 | atomic_dec(&e->dest->refcnt); | |
145 | kfree(e); | |
146 | break; | |
147 | } | |
148 | ep = &e->next; | |
149 | } | |
150 | write_unlock(&set->lock); | |
151 | } | |
152 | ||
153 | static void ip_vs_dest_set_eraseall(struct ip_vs_dest_set *set) | |
154 | { | |
155 | struct ip_vs_dest_list *e, **ep; | |
156 | ||
157 | write_lock(&set->lock); | |
158 | for (ep=&set->list, e=*ep; e!=NULL; e=*ep) { | |
159 | *ep = e->next; | |
160 | /* | |
161 | * We don't kfree dest because it is refered either | |
162 | * by its service or by the trash dest list. | |
163 | */ | |
164 | atomic_dec(&e->dest->refcnt); | |
165 | kfree(e); | |
166 | } | |
167 | write_unlock(&set->lock); | |
168 | } | |
169 | ||
170 | /* get weighted least-connection node in the destination set */ | |
171 | static inline struct ip_vs_dest *ip_vs_dest_set_min(struct ip_vs_dest_set *set) | |
172 | { | |
173 | register struct ip_vs_dest_list *e; | |
174 | struct ip_vs_dest *dest, *least; | |
175 | int loh, doh; | |
176 | ||
177 | if (set == NULL) | |
178 | return NULL; | |
179 | ||
180 | read_lock(&set->lock); | |
181 | /* select the first destination server, whose weight > 0 */ | |
182 | for (e=set->list; e!=NULL; e=e->next) { | |
183 | least = e->dest; | |
184 | if (least->flags & IP_VS_DEST_F_OVERLOAD) | |
185 | continue; | |
186 | ||
187 | if ((atomic_read(&least->weight) > 0) | |
188 | && (least->flags & IP_VS_DEST_F_AVAILABLE)) { | |
189 | loh = atomic_read(&least->activeconns) * 50 | |
190 | + atomic_read(&least->inactconns); | |
191 | goto nextstage; | |
192 | } | |
193 | } | |
194 | read_unlock(&set->lock); | |
195 | return NULL; | |
196 | ||
197 | /* find the destination with the weighted least load */ | |
198 | nextstage: | |
199 | for (e=e->next; e!=NULL; e=e->next) { | |
200 | dest = e->dest; | |
201 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
202 | continue; | |
203 | ||
204 | doh = atomic_read(&dest->activeconns) * 50 | |
205 | + atomic_read(&dest->inactconns); | |
206 | if ((loh * atomic_read(&dest->weight) > | |
207 | doh * atomic_read(&least->weight)) | |
208 | && (dest->flags & IP_VS_DEST_F_AVAILABLE)) { | |
209 | least = dest; | |
210 | loh = doh; | |
211 | } | |
212 | } | |
213 | read_unlock(&set->lock); | |
214 | ||
215 | IP_VS_DBG(6, "ip_vs_dest_set_min: server %d.%d.%d.%d:%d " | |
216 | "activeconns %d refcnt %d weight %d overhead %d\n", | |
217 | NIPQUAD(least->addr), ntohs(least->port), | |
218 | atomic_read(&least->activeconns), | |
219 | atomic_read(&least->refcnt), | |
220 | atomic_read(&least->weight), loh); | |
221 | return least; | |
222 | } | |
223 | ||
224 | ||
225 | /* get weighted most-connection node in the destination set */ | |
226 | static inline struct ip_vs_dest *ip_vs_dest_set_max(struct ip_vs_dest_set *set) | |
227 | { | |
228 | register struct ip_vs_dest_list *e; | |
229 | struct ip_vs_dest *dest, *most; | |
230 | int moh, doh; | |
231 | ||
232 | if (set == NULL) | |
233 | return NULL; | |
234 | ||
235 | read_lock(&set->lock); | |
236 | /* select the first destination server, whose weight > 0 */ | |
237 | for (e=set->list; e!=NULL; e=e->next) { | |
238 | most = e->dest; | |
239 | if (atomic_read(&most->weight) > 0) { | |
240 | moh = atomic_read(&most->activeconns) * 50 | |
241 | + atomic_read(&most->inactconns); | |
242 | goto nextstage; | |
243 | } | |
244 | } | |
245 | read_unlock(&set->lock); | |
246 | return NULL; | |
247 | ||
248 | /* find the destination with the weighted most load */ | |
249 | nextstage: | |
250 | for (e=e->next; e!=NULL; e=e->next) { | |
251 | dest = e->dest; | |
252 | doh = atomic_read(&dest->activeconns) * 50 | |
253 | + atomic_read(&dest->inactconns); | |
254 | /* moh/mw < doh/dw ==> moh*dw < doh*mw, where mw,dw>0 */ | |
255 | if ((moh * atomic_read(&dest->weight) < | |
256 | doh * atomic_read(&most->weight)) | |
257 | && (atomic_read(&dest->weight) > 0)) { | |
258 | most = dest; | |
259 | moh = doh; | |
260 | } | |
261 | } | |
262 | read_unlock(&set->lock); | |
263 | ||
264 | IP_VS_DBG(6, "ip_vs_dest_set_max: server %d.%d.%d.%d:%d " | |
265 | "activeconns %d refcnt %d weight %d overhead %d\n", | |
266 | NIPQUAD(most->addr), ntohs(most->port), | |
267 | atomic_read(&most->activeconns), | |
268 | atomic_read(&most->refcnt), | |
269 | atomic_read(&most->weight), moh); | |
270 | return most; | |
271 | } | |
272 | ||
273 | ||
274 | /* | |
275 | * IPVS lblcr entry represents an association between destination | |
276 | * IP address and its destination server set | |
277 | */ | |
278 | struct ip_vs_lblcr_entry { | |
279 | struct list_head list; | |
014d730d | 280 | __be32 addr; /* destination IP address */ |
1da177e4 LT |
281 | struct ip_vs_dest_set set; /* destination server set */ |
282 | unsigned long lastuse; /* last used time */ | |
283 | }; | |
284 | ||
285 | ||
286 | /* | |
287 | * IPVS lblcr hash table | |
288 | */ | |
289 | struct ip_vs_lblcr_table { | |
290 | rwlock_t lock; /* lock for this table */ | |
291 | struct list_head bucket[IP_VS_LBLCR_TAB_SIZE]; /* hash bucket */ | |
292 | atomic_t entries; /* number of entries */ | |
293 | int max_size; /* maximum size of entries */ | |
294 | struct timer_list periodic_timer; /* collect stale entries */ | |
295 | int rover; /* rover for expire check */ | |
296 | int counter; /* counter for no expire */ | |
297 | }; | |
298 | ||
299 | ||
300 | /* | |
301 | * IPVS LBLCR sysctl table | |
302 | */ | |
303 | ||
304 | static ctl_table vs_vars_table[] = { | |
305 | { | |
306 | .ctl_name = NET_IPV4_VS_LBLCR_EXPIRE, | |
307 | .procname = "lblcr_expiration", | |
308 | .data = &sysctl_ip_vs_lblcr_expiration, | |
309 | .maxlen = sizeof(int), | |
e905a9ed | 310 | .mode = 0644, |
1da177e4 LT |
311 | .proc_handler = &proc_dointvec_jiffies, |
312 | }, | |
313 | { .ctl_name = 0 } | |
314 | }; | |
315 | ||
316 | static ctl_table vs_table[] = { | |
317 | { | |
318 | .ctl_name = NET_IPV4_VS, | |
319 | .procname = "vs", | |
320 | .mode = 0555, | |
321 | .child = vs_vars_table | |
322 | }, | |
323 | { .ctl_name = 0 } | |
324 | }; | |
325 | ||
bf0ff9e5 | 326 | static ctl_table ipvs_ipv4_table[] = { |
1da177e4 LT |
327 | { |
328 | .ctl_name = NET_IPV4, | |
e905a9ed | 329 | .procname = "ipv4", |
1da177e4 LT |
330 | .mode = 0555, |
331 | .child = vs_table | |
332 | }, | |
333 | { .ctl_name = 0 } | |
334 | }; | |
335 | ||
336 | static ctl_table lblcr_root_table[] = { | |
337 | { | |
338 | .ctl_name = CTL_NET, | |
e905a9ed YH |
339 | .procname = "net", |
340 | .mode = 0555, | |
bf0ff9e5 | 341 | .child = ipvs_ipv4_table |
1da177e4 LT |
342 | }, |
343 | { .ctl_name = 0 } | |
344 | }; | |
345 | ||
346 | static struct ctl_table_header * sysctl_header; | |
347 | ||
348 | /* | |
349 | * new/free a ip_vs_lblcr_entry, which is a mapping of a destination | |
350 | * IP address to a server. | |
351 | */ | |
014d730d | 352 | static inline struct ip_vs_lblcr_entry *ip_vs_lblcr_new(__be32 daddr) |
1da177e4 LT |
353 | { |
354 | struct ip_vs_lblcr_entry *en; | |
355 | ||
356 | en = kmalloc(sizeof(struct ip_vs_lblcr_entry), GFP_ATOMIC); | |
357 | if (en == NULL) { | |
358 | IP_VS_ERR("ip_vs_lblcr_new(): no memory\n"); | |
359 | return NULL; | |
360 | } | |
361 | ||
362 | INIT_LIST_HEAD(&en->list); | |
363 | en->addr = daddr; | |
364 | ||
365 | /* initilize its dest set */ | |
366 | atomic_set(&(en->set.size), 0); | |
367 | en->set.list = NULL; | |
368 | rwlock_init(&en->set.lock); | |
369 | ||
370 | return en; | |
371 | } | |
372 | ||
373 | ||
374 | static inline void ip_vs_lblcr_free(struct ip_vs_lblcr_entry *en) | |
375 | { | |
376 | list_del(&en->list); | |
377 | ip_vs_dest_set_eraseall(&en->set); | |
378 | kfree(en); | |
379 | } | |
380 | ||
381 | ||
382 | /* | |
383 | * Returns hash value for IPVS LBLCR entry | |
384 | */ | |
014d730d | 385 | static inline unsigned ip_vs_lblcr_hashkey(__be32 addr) |
1da177e4 LT |
386 | { |
387 | return (ntohl(addr)*2654435761UL) & IP_VS_LBLCR_TAB_MASK; | |
388 | } | |
389 | ||
390 | ||
391 | /* | |
392 | * Hash an entry in the ip_vs_lblcr_table. | |
393 | * returns bool success. | |
394 | */ | |
395 | static int | |
396 | ip_vs_lblcr_hash(struct ip_vs_lblcr_table *tbl, struct ip_vs_lblcr_entry *en) | |
397 | { | |
398 | unsigned hash; | |
399 | ||
400 | if (!list_empty(&en->list)) { | |
401 | IP_VS_ERR("ip_vs_lblcr_hash(): request for already hashed, " | |
402 | "called from %p\n", __builtin_return_address(0)); | |
403 | return 0; | |
404 | } | |
405 | ||
406 | /* | |
407 | * Hash by destination IP address | |
408 | */ | |
409 | hash = ip_vs_lblcr_hashkey(en->addr); | |
410 | ||
411 | write_lock(&tbl->lock); | |
412 | list_add(&en->list, &tbl->bucket[hash]); | |
413 | atomic_inc(&tbl->entries); | |
414 | write_unlock(&tbl->lock); | |
415 | ||
416 | return 1; | |
417 | } | |
418 | ||
419 | ||
1da177e4 LT |
420 | /* |
421 | * Get ip_vs_lblcr_entry associated with supplied parameters. | |
422 | */ | |
423 | static inline struct ip_vs_lblcr_entry * | |
014d730d | 424 | ip_vs_lblcr_get(struct ip_vs_lblcr_table *tbl, __be32 addr) |
1da177e4 LT |
425 | { |
426 | unsigned hash; | |
427 | struct ip_vs_lblcr_entry *en; | |
428 | ||
429 | hash = ip_vs_lblcr_hashkey(addr); | |
430 | ||
431 | read_lock(&tbl->lock); | |
432 | ||
433 | list_for_each_entry(en, &tbl->bucket[hash], list) { | |
434 | if (en->addr == addr) { | |
435 | /* HIT */ | |
436 | read_unlock(&tbl->lock); | |
437 | return en; | |
438 | } | |
439 | } | |
440 | ||
441 | read_unlock(&tbl->lock); | |
442 | ||
443 | return NULL; | |
444 | } | |
445 | ||
446 | ||
447 | /* | |
448 | * Flush all the entries of the specified table. | |
449 | */ | |
450 | static void ip_vs_lblcr_flush(struct ip_vs_lblcr_table *tbl) | |
451 | { | |
452 | int i; | |
453 | struct ip_vs_lblcr_entry *en, *nxt; | |
454 | ||
455 | for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) { | |
456 | write_lock(&tbl->lock); | |
457 | list_for_each_entry_safe(en, nxt, &tbl->bucket[i], list) { | |
458 | ip_vs_lblcr_free(en); | |
459 | atomic_dec(&tbl->entries); | |
460 | } | |
461 | write_unlock(&tbl->lock); | |
462 | } | |
463 | } | |
464 | ||
465 | ||
466 | static inline void ip_vs_lblcr_full_check(struct ip_vs_lblcr_table *tbl) | |
467 | { | |
468 | unsigned long now = jiffies; | |
469 | int i, j; | |
470 | struct ip_vs_lblcr_entry *en, *nxt; | |
471 | ||
472 | for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) { | |
473 | j = (j + 1) & IP_VS_LBLCR_TAB_MASK; | |
474 | ||
475 | write_lock(&tbl->lock); | |
476 | list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) { | |
477 | if (time_after(en->lastuse+sysctl_ip_vs_lblcr_expiration, | |
478 | now)) | |
479 | continue; | |
480 | ||
481 | ip_vs_lblcr_free(en); | |
482 | atomic_dec(&tbl->entries); | |
483 | } | |
484 | write_unlock(&tbl->lock); | |
485 | } | |
486 | tbl->rover = j; | |
487 | } | |
488 | ||
489 | ||
490 | /* | |
491 | * Periodical timer handler for IPVS lblcr table | |
492 | * It is used to collect stale entries when the number of entries | |
493 | * exceeds the maximum size of the table. | |
494 | * | |
495 | * Fixme: we probably need more complicated algorithm to collect | |
496 | * entries that have not been used for a long time even | |
497 | * if the number of entries doesn't exceed the maximum size | |
498 | * of the table. | |
499 | * The full expiration check is for this purpose now. | |
500 | */ | |
501 | static void ip_vs_lblcr_check_expire(unsigned long data) | |
502 | { | |
503 | struct ip_vs_lblcr_table *tbl; | |
504 | unsigned long now = jiffies; | |
505 | int goal; | |
506 | int i, j; | |
507 | struct ip_vs_lblcr_entry *en, *nxt; | |
508 | ||
509 | tbl = (struct ip_vs_lblcr_table *)data; | |
510 | ||
511 | if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) { | |
512 | /* do full expiration check */ | |
513 | ip_vs_lblcr_full_check(tbl); | |
514 | tbl->counter = 1; | |
515 | goto out; | |
516 | } | |
517 | ||
518 | if (atomic_read(&tbl->entries) <= tbl->max_size) { | |
519 | tbl->counter++; | |
520 | goto out; | |
521 | } | |
522 | ||
523 | goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3; | |
524 | if (goal > tbl->max_size/2) | |
525 | goal = tbl->max_size/2; | |
526 | ||
527 | for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) { | |
528 | j = (j + 1) & IP_VS_LBLCR_TAB_MASK; | |
529 | ||
530 | write_lock(&tbl->lock); | |
531 | list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) { | |
532 | if (time_before(now, en->lastuse+ENTRY_TIMEOUT)) | |
533 | continue; | |
534 | ||
535 | ip_vs_lblcr_free(en); | |
536 | atomic_dec(&tbl->entries); | |
537 | goal--; | |
538 | } | |
539 | write_unlock(&tbl->lock); | |
540 | if (goal <= 0) | |
541 | break; | |
542 | } | |
543 | tbl->rover = j; | |
544 | ||
545 | out: | |
546 | mod_timer(&tbl->periodic_timer, jiffies+CHECK_EXPIRE_INTERVAL); | |
547 | } | |
548 | ||
549 | ||
550 | #ifdef CONFIG_IP_VS_LBLCR_DEBUG | |
551 | static struct ip_vs_lblcr_table *lblcr_table_list; | |
552 | ||
553 | /* | |
554 | * /proc/net/ip_vs_lblcr to display the mappings of | |
555 | * destination IP address <==> its serverSet | |
556 | */ | |
557 | static int | |
558 | ip_vs_lblcr_getinfo(char *buffer, char **start, off_t offset, int length) | |
559 | { | |
560 | off_t pos=0, begin; | |
561 | int len=0, size; | |
562 | struct ip_vs_lblcr_table *tbl; | |
563 | unsigned long now = jiffies; | |
564 | int i; | |
565 | struct ip_vs_lblcr_entry *en; | |
566 | ||
567 | tbl = lblcr_table_list; | |
568 | ||
569 | size = sprintf(buffer, "LastTime Dest IP address Server set\n"); | |
570 | pos += size; | |
571 | len += size; | |
572 | ||
573 | for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) { | |
574 | read_lock_bh(&tbl->lock); | |
575 | list_for_each_entry(en, &tbl->bucket[i], list) { | |
576 | char tbuf[16]; | |
577 | struct ip_vs_dest_list *d; | |
578 | ||
579 | sprintf(tbuf, "%u.%u.%u.%u", NIPQUAD(en->addr)); | |
580 | size = sprintf(buffer+len, "%8lu %-16s ", | |
581 | now-en->lastuse, tbuf); | |
582 | ||
583 | read_lock(&en->set.lock); | |
584 | for (d=en->set.list; d!=NULL; d=d->next) { | |
585 | size += sprintf(buffer+len+size, | |
586 | "%u.%u.%u.%u ", | |
587 | NIPQUAD(d->dest->addr)); | |
588 | } | |
589 | read_unlock(&en->set.lock); | |
590 | size += sprintf(buffer+len+size, "\n"); | |
591 | len += size; | |
592 | pos += size; | |
593 | if (pos <= offset) | |
594 | len=0; | |
595 | if (pos >= offset+length) { | |
596 | read_unlock_bh(&tbl->lock); | |
597 | goto done; | |
598 | } | |
599 | } | |
600 | read_unlock_bh(&tbl->lock); | |
601 | } | |
602 | ||
603 | done: | |
604 | begin = len - (pos - offset); | |
605 | *start = buffer + begin; | |
606 | len -= begin; | |
607 | if(len>length) | |
608 | len = length; | |
609 | return len; | |
610 | } | |
611 | #endif | |
612 | ||
613 | ||
614 | static int ip_vs_lblcr_init_svc(struct ip_vs_service *svc) | |
615 | { | |
616 | int i; | |
617 | struct ip_vs_lblcr_table *tbl; | |
618 | ||
619 | /* | |
620 | * Allocate the ip_vs_lblcr_table for this service | |
621 | */ | |
622 | tbl = kmalloc(sizeof(struct ip_vs_lblcr_table), GFP_ATOMIC); | |
623 | if (tbl == NULL) { | |
624 | IP_VS_ERR("ip_vs_lblcr_init_svc(): no memory\n"); | |
625 | return -ENOMEM; | |
626 | } | |
627 | svc->sched_data = tbl; | |
628 | IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) allocated for " | |
629 | "current service\n", | |
630 | sizeof(struct ip_vs_lblcr_table)); | |
631 | ||
632 | /* | |
633 | * Initialize the hash buckets | |
634 | */ | |
635 | for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) { | |
636 | INIT_LIST_HEAD(&tbl->bucket[i]); | |
637 | } | |
638 | rwlock_init(&tbl->lock); | |
639 | tbl->max_size = IP_VS_LBLCR_TAB_SIZE*16; | |
640 | tbl->rover = 0; | |
641 | tbl->counter = 1; | |
642 | ||
643 | /* | |
644 | * Hook periodic timer for garbage collection | |
645 | */ | |
646 | init_timer(&tbl->periodic_timer); | |
647 | tbl->periodic_timer.data = (unsigned long)tbl; | |
648 | tbl->periodic_timer.function = ip_vs_lblcr_check_expire; | |
649 | tbl->periodic_timer.expires = jiffies+CHECK_EXPIRE_INTERVAL; | |
650 | add_timer(&tbl->periodic_timer); | |
651 | ||
652 | #ifdef CONFIG_IP_VS_LBLCR_DEBUG | |
653 | lblcr_table_list = tbl; | |
654 | #endif | |
655 | return 0; | |
656 | } | |
657 | ||
658 | ||
659 | static int ip_vs_lblcr_done_svc(struct ip_vs_service *svc) | |
660 | { | |
661 | struct ip_vs_lblcr_table *tbl = svc->sched_data; | |
662 | ||
663 | /* remove periodic timer */ | |
664 | del_timer_sync(&tbl->periodic_timer); | |
665 | ||
666 | /* got to clean up table entries here */ | |
667 | ip_vs_lblcr_flush(tbl); | |
668 | ||
669 | /* release the table itself */ | |
670 | kfree(svc->sched_data); | |
671 | IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) released\n", | |
672 | sizeof(struct ip_vs_lblcr_table)); | |
673 | ||
674 | return 0; | |
675 | } | |
676 | ||
677 | ||
678 | static int ip_vs_lblcr_update_svc(struct ip_vs_service *svc) | |
679 | { | |
680 | return 0; | |
681 | } | |
682 | ||
683 | ||
684 | static inline struct ip_vs_dest * | |
685 | __ip_vs_wlc_schedule(struct ip_vs_service *svc, struct iphdr *iph) | |
686 | { | |
687 | struct ip_vs_dest *dest, *least; | |
688 | int loh, doh; | |
689 | ||
690 | /* | |
691 | * We think the overhead of processing active connections is fifty | |
692 | * times higher than that of inactive connections in average. (This | |
693 | * fifty times might not be accurate, we will change it later.) We | |
694 | * use the following formula to estimate the overhead: | |
695 | * dest->activeconns*50 + dest->inactconns | |
696 | * and the load: | |
697 | * (dest overhead) / dest->weight | |
698 | * | |
699 | * Remember -- no floats in kernel mode!!! | |
700 | * The comparison of h1*w2 > h2*w1 is equivalent to that of | |
701 | * h1/w1 > h2/w2 | |
702 | * if every weight is larger than zero. | |
703 | * | |
704 | * The server with weight=0 is quiesced and will not receive any | |
705 | * new connection. | |
706 | */ | |
707 | list_for_each_entry(dest, &svc->destinations, n_list) { | |
708 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
709 | continue; | |
710 | ||
711 | if (atomic_read(&dest->weight) > 0) { | |
712 | least = dest; | |
713 | loh = atomic_read(&least->activeconns) * 50 | |
714 | + atomic_read(&least->inactconns); | |
715 | goto nextstage; | |
716 | } | |
717 | } | |
718 | return NULL; | |
719 | ||
720 | /* | |
721 | * Find the destination with the least load. | |
722 | */ | |
723 | nextstage: | |
724 | list_for_each_entry_continue(dest, &svc->destinations, n_list) { | |
725 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
726 | continue; | |
727 | ||
728 | doh = atomic_read(&dest->activeconns) * 50 | |
729 | + atomic_read(&dest->inactconns); | |
730 | if (loh * atomic_read(&dest->weight) > | |
731 | doh * atomic_read(&least->weight)) { | |
732 | least = dest; | |
733 | loh = doh; | |
734 | } | |
735 | } | |
736 | ||
737 | IP_VS_DBG(6, "LBLCR: server %d.%d.%d.%d:%d " | |
738 | "activeconns %d refcnt %d weight %d overhead %d\n", | |
739 | NIPQUAD(least->addr), ntohs(least->port), | |
740 | atomic_read(&least->activeconns), | |
741 | atomic_read(&least->refcnt), | |
742 | atomic_read(&least->weight), loh); | |
743 | ||
744 | return least; | |
745 | } | |
746 | ||
747 | ||
748 | /* | |
749 | * If this destination server is overloaded and there is a less loaded | |
750 | * server, then return true. | |
751 | */ | |
752 | static inline int | |
753 | is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc) | |
754 | { | |
755 | if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) { | |
756 | struct ip_vs_dest *d; | |
757 | ||
758 | list_for_each_entry(d, &svc->destinations, n_list) { | |
759 | if (atomic_read(&d->activeconns)*2 | |
760 | < atomic_read(&d->weight)) { | |
761 | return 1; | |
762 | } | |
763 | } | |
764 | } | |
765 | return 0; | |
766 | } | |
767 | ||
768 | ||
769 | /* | |
770 | * Locality-Based (weighted) Least-Connection scheduling | |
771 | */ | |
772 | static struct ip_vs_dest * | |
773 | ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb) | |
774 | { | |
775 | struct ip_vs_dest *dest; | |
776 | struct ip_vs_lblcr_table *tbl; | |
777 | struct ip_vs_lblcr_entry *en; | |
eddc9ec5 | 778 | struct iphdr *iph = ip_hdr(skb); |
1da177e4 LT |
779 | |
780 | IP_VS_DBG(6, "ip_vs_lblcr_schedule(): Scheduling...\n"); | |
781 | ||
782 | tbl = (struct ip_vs_lblcr_table *)svc->sched_data; | |
783 | en = ip_vs_lblcr_get(tbl, iph->daddr); | |
784 | if (en == NULL) { | |
785 | dest = __ip_vs_wlc_schedule(svc, iph); | |
786 | if (dest == NULL) { | |
787 | IP_VS_DBG(1, "no destination available\n"); | |
788 | return NULL; | |
789 | } | |
790 | en = ip_vs_lblcr_new(iph->daddr); | |
791 | if (en == NULL) { | |
792 | return NULL; | |
793 | } | |
794 | ip_vs_dest_set_insert(&en->set, dest); | |
795 | ip_vs_lblcr_hash(tbl, en); | |
796 | } else { | |
797 | dest = ip_vs_dest_set_min(&en->set); | |
798 | if (!dest || is_overloaded(dest, svc)) { | |
799 | dest = __ip_vs_wlc_schedule(svc, iph); | |
800 | if (dest == NULL) { | |
801 | IP_VS_DBG(1, "no destination available\n"); | |
802 | return NULL; | |
803 | } | |
804 | ip_vs_dest_set_insert(&en->set, dest); | |
805 | } | |
806 | if (atomic_read(&en->set.size) > 1 && | |
807 | jiffies-en->set.lastmod > sysctl_ip_vs_lblcr_expiration) { | |
808 | struct ip_vs_dest *m; | |
809 | m = ip_vs_dest_set_max(&en->set); | |
810 | if (m) | |
811 | ip_vs_dest_set_erase(&en->set, m); | |
812 | } | |
813 | } | |
814 | en->lastuse = jiffies; | |
815 | ||
816 | IP_VS_DBG(6, "LBLCR: destination IP address %u.%u.%u.%u " | |
817 | "--> server %u.%u.%u.%u:%d\n", | |
818 | NIPQUAD(en->addr), | |
819 | NIPQUAD(dest->addr), | |
820 | ntohs(dest->port)); | |
821 | ||
822 | return dest; | |
823 | } | |
824 | ||
825 | ||
826 | /* | |
827 | * IPVS LBLCR Scheduler structure | |
828 | */ | |
829 | static struct ip_vs_scheduler ip_vs_lblcr_scheduler = | |
830 | { | |
831 | .name = "lblcr", | |
832 | .refcnt = ATOMIC_INIT(0), | |
833 | .module = THIS_MODULE, | |
834 | .init_service = ip_vs_lblcr_init_svc, | |
835 | .done_service = ip_vs_lblcr_done_svc, | |
836 | .update_service = ip_vs_lblcr_update_svc, | |
837 | .schedule = ip_vs_lblcr_schedule, | |
838 | }; | |
839 | ||
840 | ||
841 | static int __init ip_vs_lblcr_init(void) | |
842 | { | |
843 | INIT_LIST_HEAD(&ip_vs_lblcr_scheduler.n_list); | |
0b4d4147 | 844 | sysctl_header = register_sysctl_table(lblcr_root_table); |
1da177e4 LT |
845 | #ifdef CONFIG_IP_VS_LBLCR_DEBUG |
846 | proc_net_create("ip_vs_lblcr", 0, ip_vs_lblcr_getinfo); | |
847 | #endif | |
848 | return register_ip_vs_scheduler(&ip_vs_lblcr_scheduler); | |
849 | } | |
850 | ||
851 | ||
852 | static void __exit ip_vs_lblcr_cleanup(void) | |
853 | { | |
854 | #ifdef CONFIG_IP_VS_LBLCR_DEBUG | |
855 | proc_net_remove("ip_vs_lblcr"); | |
856 | #endif | |
857 | unregister_sysctl_table(sysctl_header); | |
858 | unregister_ip_vs_scheduler(&ip_vs_lblcr_scheduler); | |
859 | } | |
860 | ||
861 | ||
862 | module_init(ip_vs_lblcr_init); | |
863 | module_exit(ip_vs_lblcr_cleanup); | |
864 | MODULE_LICENSE("GPL"); |