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