]>
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 | ||
9aada7ac HE |
40 | #define KMSG_COMPONENT "IPVS" |
41 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt | |
42 | ||
14c85021 | 43 | #include <linux/ip.h> |
1da177e4 LT |
44 | #include <linux/module.h> |
45 | #include <linux/kernel.h> | |
14c85021 | 46 | #include <linux/skbuff.h> |
d7fe0f24 | 47 | #include <linux/jiffies.h> |
51f0bc78 | 48 | #include <linux/list.h> |
5a0e3ad6 | 49 | #include <linux/slab.h> |
1da177e4 LT |
50 | |
51 | /* for sysctl */ | |
52 | #include <linux/fs.h> | |
53 | #include <linux/sysctl.h> | |
457c4cbc | 54 | #include <net/net_namespace.h> |
1da177e4 LT |
55 | |
56 | #include <net/ip_vs.h> | |
57 | ||
58 | ||
59 | /* | |
60 | * It is for garbage collection of stale IPVS lblcr entries, | |
61 | * when the table is full. | |
62 | */ | |
63 | #define CHECK_EXPIRE_INTERVAL (60*HZ) | |
64 | #define ENTRY_TIMEOUT (6*60*HZ) | |
65 | ||
66 | /* | |
67 | * It is for full expiration check. | |
68 | * When there is no partial expiration check (garbage collection) | |
69 | * in a half hour, do a full expiration check to collect stale | |
70 | * entries that haven't been touched for a day. | |
71 | */ | |
72 | #define COUNT_FOR_FULL_EXPIRATION 30 | |
73 | static int sysctl_ip_vs_lblcr_expiration = 24*60*60*HZ; | |
74 | ||
75 | ||
76 | /* | |
77 | * for IPVS lblcr entry hash table | |
78 | */ | |
79 | #ifndef CONFIG_IP_VS_LBLCR_TAB_BITS | |
80 | #define CONFIG_IP_VS_LBLCR_TAB_BITS 10 | |
81 | #endif | |
82 | #define IP_VS_LBLCR_TAB_BITS CONFIG_IP_VS_LBLCR_TAB_BITS | |
83 | #define IP_VS_LBLCR_TAB_SIZE (1 << IP_VS_LBLCR_TAB_BITS) | |
84 | #define IP_VS_LBLCR_TAB_MASK (IP_VS_LBLCR_TAB_SIZE - 1) | |
85 | ||
86 | ||
87 | /* | |
88 | * IPVS destination set structure and operations | |
89 | */ | |
51f0bc78 SH |
90 | struct ip_vs_dest_set_elem { |
91 | struct list_head list; /* list link */ | |
1da177e4 LT |
92 | struct ip_vs_dest *dest; /* destination server */ |
93 | }; | |
94 | ||
95 | struct ip_vs_dest_set { | |
96 | atomic_t size; /* set size */ | |
97 | unsigned long lastmod; /* last modified time */ | |
51f0bc78 | 98 | struct list_head list; /* destination list */ |
1da177e4 LT |
99 | rwlock_t lock; /* lock for this list */ |
100 | }; | |
101 | ||
102 | ||
51f0bc78 | 103 | static struct ip_vs_dest_set_elem * |
1da177e4 LT |
104 | ip_vs_dest_set_insert(struct ip_vs_dest_set *set, struct ip_vs_dest *dest) |
105 | { | |
51f0bc78 | 106 | struct ip_vs_dest_set_elem *e; |
1da177e4 | 107 | |
51f0bc78 | 108 | list_for_each_entry(e, &set->list, list) { |
1da177e4 LT |
109 | if (e->dest == dest) |
110 | /* already existed */ | |
111 | return NULL; | |
112 | } | |
113 | ||
f728bafb | 114 | e = kmalloc(sizeof(*e), GFP_ATOMIC); |
1da177e4 | 115 | if (e == NULL) { |
1e3e238e | 116 | pr_err("%s(): no memory\n", __func__); |
1da177e4 LT |
117 | return NULL; |
118 | } | |
119 | ||
120 | atomic_inc(&dest->refcnt); | |
121 | e->dest = dest; | |
122 | ||
51f0bc78 | 123 | list_add(&e->list, &set->list); |
1da177e4 | 124 | atomic_inc(&set->size); |
1da177e4 LT |
125 | |
126 | set->lastmod = jiffies; | |
127 | return e; | |
128 | } | |
129 | ||
130 | static void | |
131 | ip_vs_dest_set_erase(struct ip_vs_dest_set *set, struct ip_vs_dest *dest) | |
132 | { | |
51f0bc78 | 133 | struct ip_vs_dest_set_elem *e; |
1da177e4 | 134 | |
51f0bc78 | 135 | list_for_each_entry(e, &set->list, list) { |
1da177e4 LT |
136 | if (e->dest == dest) { |
137 | /* HIT */ | |
1da177e4 LT |
138 | atomic_dec(&set->size); |
139 | set->lastmod = jiffies; | |
140 | atomic_dec(&e->dest->refcnt); | |
51f0bc78 | 141 | list_del(&e->list); |
1da177e4 LT |
142 | kfree(e); |
143 | break; | |
144 | } | |
1da177e4 | 145 | } |
1da177e4 LT |
146 | } |
147 | ||
148 | static void ip_vs_dest_set_eraseall(struct ip_vs_dest_set *set) | |
149 | { | |
51f0bc78 | 150 | struct ip_vs_dest_set_elem *e, *ep; |
1da177e4 LT |
151 | |
152 | write_lock(&set->lock); | |
51f0bc78 | 153 | list_for_each_entry_safe(e, ep, &set->list, list) { |
1da177e4 LT |
154 | /* |
155 | * We don't kfree dest because it is refered either | |
156 | * by its service or by the trash dest list. | |
157 | */ | |
158 | atomic_dec(&e->dest->refcnt); | |
51f0bc78 | 159 | list_del(&e->list); |
1da177e4 LT |
160 | kfree(e); |
161 | } | |
162 | write_unlock(&set->lock); | |
163 | } | |
164 | ||
165 | /* get weighted least-connection node in the destination set */ | |
166 | static inline struct ip_vs_dest *ip_vs_dest_set_min(struct ip_vs_dest_set *set) | |
167 | { | |
51f0bc78 | 168 | register struct ip_vs_dest_set_elem *e; |
1da177e4 LT |
169 | struct ip_vs_dest *dest, *least; |
170 | int loh, doh; | |
171 | ||
172 | if (set == NULL) | |
173 | return NULL; | |
174 | ||
1da177e4 | 175 | /* select the first destination server, whose weight > 0 */ |
51f0bc78 | 176 | list_for_each_entry(e, &set->list, list) { |
1da177e4 LT |
177 | least = e->dest; |
178 | if (least->flags & IP_VS_DEST_F_OVERLOAD) | |
179 | continue; | |
180 | ||
181 | if ((atomic_read(&least->weight) > 0) | |
182 | && (least->flags & IP_VS_DEST_F_AVAILABLE)) { | |
183 | loh = atomic_read(&least->activeconns) * 50 | |
184 | + atomic_read(&least->inactconns); | |
185 | goto nextstage; | |
186 | } | |
187 | } | |
1da177e4 LT |
188 | return NULL; |
189 | ||
190 | /* find the destination with the weighted least load */ | |
191 | nextstage: | |
51f0bc78 | 192 | list_for_each_entry(e, &set->list, list) { |
1da177e4 LT |
193 | dest = e->dest; |
194 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
195 | continue; | |
196 | ||
197 | doh = atomic_read(&dest->activeconns) * 50 | |
198 | + atomic_read(&dest->inactconns); | |
199 | if ((loh * atomic_read(&dest->weight) > | |
200 | doh * atomic_read(&least->weight)) | |
201 | && (dest->flags & IP_VS_DEST_F_AVAILABLE)) { | |
202 | least = dest; | |
203 | loh = doh; | |
204 | } | |
205 | } | |
1da177e4 | 206 | |
1e3e238e | 207 | IP_VS_DBG_BUF(6, "%s(): server %s:%d " |
44548375 | 208 | "activeconns %d refcnt %d weight %d overhead %d\n", |
1e3e238e | 209 | __func__, |
44548375 JV |
210 | IP_VS_DBG_ADDR(least->af, &least->addr), |
211 | ntohs(least->port), | |
212 | atomic_read(&least->activeconns), | |
213 | atomic_read(&least->refcnt), | |
214 | atomic_read(&least->weight), loh); | |
1da177e4 LT |
215 | return least; |
216 | } | |
217 | ||
218 | ||
219 | /* get weighted most-connection node in the destination set */ | |
220 | static inline struct ip_vs_dest *ip_vs_dest_set_max(struct ip_vs_dest_set *set) | |
221 | { | |
51f0bc78 | 222 | register struct ip_vs_dest_set_elem *e; |
1da177e4 LT |
223 | struct ip_vs_dest *dest, *most; |
224 | int moh, doh; | |
225 | ||
226 | if (set == NULL) | |
227 | return NULL; | |
228 | ||
1da177e4 | 229 | /* select the first destination server, whose weight > 0 */ |
51f0bc78 | 230 | list_for_each_entry(e, &set->list, list) { |
1da177e4 LT |
231 | most = e->dest; |
232 | if (atomic_read(&most->weight) > 0) { | |
233 | moh = atomic_read(&most->activeconns) * 50 | |
234 | + atomic_read(&most->inactconns); | |
235 | goto nextstage; | |
236 | } | |
237 | } | |
1da177e4 LT |
238 | return NULL; |
239 | ||
240 | /* find the destination with the weighted most load */ | |
241 | nextstage: | |
51f0bc78 | 242 | list_for_each_entry(e, &set->list, list) { |
1da177e4 LT |
243 | dest = e->dest; |
244 | doh = atomic_read(&dest->activeconns) * 50 | |
245 | + atomic_read(&dest->inactconns); | |
246 | /* moh/mw < doh/dw ==> moh*dw < doh*mw, where mw,dw>0 */ | |
247 | if ((moh * atomic_read(&dest->weight) < | |
248 | doh * atomic_read(&most->weight)) | |
249 | && (atomic_read(&dest->weight) > 0)) { | |
250 | most = dest; | |
251 | moh = doh; | |
252 | } | |
253 | } | |
1da177e4 | 254 | |
1e3e238e | 255 | IP_VS_DBG_BUF(6, "%s(): server %s:%d " |
44548375 | 256 | "activeconns %d refcnt %d weight %d overhead %d\n", |
1e3e238e | 257 | __func__, |
44548375 JV |
258 | IP_VS_DBG_ADDR(most->af, &most->addr), ntohs(most->port), |
259 | atomic_read(&most->activeconns), | |
260 | atomic_read(&most->refcnt), | |
261 | atomic_read(&most->weight), moh); | |
1da177e4 LT |
262 | return most; |
263 | } | |
264 | ||
265 | ||
266 | /* | |
267 | * IPVS lblcr entry represents an association between destination | |
268 | * IP address and its destination server set | |
269 | */ | |
270 | struct ip_vs_lblcr_entry { | |
271 | struct list_head list; | |
44548375 JV |
272 | int af; /* address family */ |
273 | union nf_inet_addr addr; /* destination IP address */ | |
1da177e4 LT |
274 | struct ip_vs_dest_set set; /* destination server set */ |
275 | unsigned long lastuse; /* last used time */ | |
276 | }; | |
277 | ||
278 | ||
279 | /* | |
280 | * IPVS lblcr hash table | |
281 | */ | |
282 | struct ip_vs_lblcr_table { | |
1da177e4 LT |
283 | struct list_head bucket[IP_VS_LBLCR_TAB_SIZE]; /* hash bucket */ |
284 | atomic_t entries; /* number of entries */ | |
285 | int max_size; /* maximum size of entries */ | |
286 | struct timer_list periodic_timer; /* collect stale entries */ | |
287 | int rover; /* rover for expire check */ | |
288 | int counter; /* counter for no expire */ | |
289 | }; | |
290 | ||
291 | ||
292 | /* | |
293 | * IPVS LBLCR sysctl table | |
294 | */ | |
295 | ||
296 | static ctl_table vs_vars_table[] = { | |
297 | { | |
1da177e4 LT |
298 | .procname = "lblcr_expiration", |
299 | .data = &sysctl_ip_vs_lblcr_expiration, | |
300 | .maxlen = sizeof(int), | |
e905a9ed | 301 | .mode = 0644, |
6d9f239a | 302 | .proc_handler = proc_dointvec_jiffies, |
1da177e4 | 303 | }, |
f8572d8f | 304 | { } |
1da177e4 LT |
305 | }; |
306 | ||
1da177e4 LT |
307 | static struct ctl_table_header * sysctl_header; |
308 | ||
1da177e4 LT |
309 | static inline void ip_vs_lblcr_free(struct ip_vs_lblcr_entry *en) |
310 | { | |
311 | list_del(&en->list); | |
312 | ip_vs_dest_set_eraseall(&en->set); | |
313 | kfree(en); | |
314 | } | |
315 | ||
316 | ||
317 | /* | |
318 | * Returns hash value for IPVS LBLCR entry | |
319 | */ | |
44548375 JV |
320 | static inline unsigned |
321 | ip_vs_lblcr_hashkey(int af, const union nf_inet_addr *addr) | |
1da177e4 | 322 | { |
44548375 JV |
323 | __be32 addr_fold = addr->ip; |
324 | ||
325 | #ifdef CONFIG_IP_VS_IPV6 | |
326 | if (af == AF_INET6) | |
327 | addr_fold = addr->ip6[0]^addr->ip6[1]^ | |
328 | addr->ip6[2]^addr->ip6[3]; | |
329 | #endif | |
330 | return (ntohl(addr_fold)*2654435761UL) & IP_VS_LBLCR_TAB_MASK; | |
1da177e4 LT |
331 | } |
332 | ||
333 | ||
334 | /* | |
335 | * Hash an entry in the ip_vs_lblcr_table. | |
336 | * returns bool success. | |
337 | */ | |
f728bafb | 338 | static void |
1da177e4 LT |
339 | ip_vs_lblcr_hash(struct ip_vs_lblcr_table *tbl, struct ip_vs_lblcr_entry *en) |
340 | { | |
44548375 | 341 | unsigned hash = ip_vs_lblcr_hashkey(en->af, &en->addr); |
1da177e4 | 342 | |
1da177e4 LT |
343 | list_add(&en->list, &tbl->bucket[hash]); |
344 | atomic_inc(&tbl->entries); | |
1da177e4 LT |
345 | } |
346 | ||
347 | ||
1da177e4 | 348 | /* |
f728bafb SW |
349 | * Get ip_vs_lblcr_entry associated with supplied parameters. Called under |
350 | * read lock. | |
1da177e4 LT |
351 | */ |
352 | static inline struct ip_vs_lblcr_entry * | |
44548375 JV |
353 | ip_vs_lblcr_get(int af, struct ip_vs_lblcr_table *tbl, |
354 | const union nf_inet_addr *addr) | |
1da177e4 | 355 | { |
44548375 | 356 | unsigned hash = ip_vs_lblcr_hashkey(af, addr); |
1da177e4 LT |
357 | struct ip_vs_lblcr_entry *en; |
358 | ||
f728bafb | 359 | list_for_each_entry(en, &tbl->bucket[hash], list) |
44548375 | 360 | if (ip_vs_addr_equal(af, &en->addr, addr)) |
f728bafb SW |
361 | return en; |
362 | ||
363 | return NULL; | |
364 | } | |
1da177e4 | 365 | |
1da177e4 | 366 | |
f728bafb SW |
367 | /* |
368 | * Create or update an ip_vs_lblcr_entry, which is a mapping of a destination | |
369 | * IP address to a server. Called under write lock. | |
370 | */ | |
371 | static inline struct ip_vs_lblcr_entry * | |
44548375 | 372 | ip_vs_lblcr_new(struct ip_vs_lblcr_table *tbl, const union nf_inet_addr *daddr, |
f728bafb SW |
373 | struct ip_vs_dest *dest) |
374 | { | |
375 | struct ip_vs_lblcr_entry *en; | |
376 | ||
44548375 | 377 | en = ip_vs_lblcr_get(dest->af, tbl, daddr); |
f728bafb SW |
378 | if (!en) { |
379 | en = kmalloc(sizeof(*en), GFP_ATOMIC); | |
380 | if (!en) { | |
1e3e238e | 381 | pr_err("%s(): no memory\n", __func__); |
f728bafb | 382 | return NULL; |
1da177e4 | 383 | } |
f728bafb | 384 | |
44548375 JV |
385 | en->af = dest->af; |
386 | ip_vs_addr_copy(dest->af, &en->addr, daddr); | |
f728bafb SW |
387 | en->lastuse = jiffies; |
388 | ||
421f91d2 | 389 | /* initialize its dest set */ |
f728bafb | 390 | atomic_set(&(en->set.size), 0); |
51f0bc78 | 391 | INIT_LIST_HEAD(&en->set.list); |
f728bafb SW |
392 | rwlock_init(&en->set.lock); |
393 | ||
394 | ip_vs_lblcr_hash(tbl, en); | |
1da177e4 LT |
395 | } |
396 | ||
f728bafb SW |
397 | write_lock(&en->set.lock); |
398 | ip_vs_dest_set_insert(&en->set, dest); | |
399 | write_unlock(&en->set.lock); | |
1da177e4 | 400 | |
f728bafb | 401 | return en; |
1da177e4 LT |
402 | } |
403 | ||
404 | ||
405 | /* | |
406 | * Flush all the entries of the specified table. | |
407 | */ | |
408 | static void ip_vs_lblcr_flush(struct ip_vs_lblcr_table *tbl) | |
409 | { | |
410 | int i; | |
411 | struct ip_vs_lblcr_entry *en, *nxt; | |
412 | ||
f728bafb | 413 | /* No locking required, only called during cleanup. */ |
1da177e4 | 414 | for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) { |
1da177e4 LT |
415 | list_for_each_entry_safe(en, nxt, &tbl->bucket[i], list) { |
416 | ip_vs_lblcr_free(en); | |
1da177e4 | 417 | } |
1da177e4 LT |
418 | } |
419 | } | |
420 | ||
421 | ||
f728bafb | 422 | static inline void ip_vs_lblcr_full_check(struct ip_vs_service *svc) |
1da177e4 | 423 | { |
f728bafb | 424 | struct ip_vs_lblcr_table *tbl = svc->sched_data; |
1da177e4 LT |
425 | unsigned long now = jiffies; |
426 | int i, j; | |
427 | struct ip_vs_lblcr_entry *en, *nxt; | |
428 | ||
429 | for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) { | |
430 | j = (j + 1) & IP_VS_LBLCR_TAB_MASK; | |
431 | ||
f728bafb | 432 | write_lock(&svc->sched_lock); |
1da177e4 LT |
433 | list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) { |
434 | if (time_after(en->lastuse+sysctl_ip_vs_lblcr_expiration, | |
435 | now)) | |
436 | continue; | |
437 | ||
438 | ip_vs_lblcr_free(en); | |
439 | atomic_dec(&tbl->entries); | |
440 | } | |
f728bafb | 441 | write_unlock(&svc->sched_lock); |
1da177e4 LT |
442 | } |
443 | tbl->rover = j; | |
444 | } | |
445 | ||
446 | ||
447 | /* | |
448 | * Periodical timer handler for IPVS lblcr table | |
449 | * It is used to collect stale entries when the number of entries | |
450 | * exceeds the maximum size of the table. | |
451 | * | |
452 | * Fixme: we probably need more complicated algorithm to collect | |
453 | * entries that have not been used for a long time even | |
454 | * if the number of entries doesn't exceed the maximum size | |
455 | * of the table. | |
456 | * The full expiration check is for this purpose now. | |
457 | */ | |
458 | static void ip_vs_lblcr_check_expire(unsigned long data) | |
459 | { | |
f728bafb SW |
460 | struct ip_vs_service *svc = (struct ip_vs_service *) data; |
461 | struct ip_vs_lblcr_table *tbl = svc->sched_data; | |
1da177e4 LT |
462 | unsigned long now = jiffies; |
463 | int goal; | |
464 | int i, j; | |
465 | struct ip_vs_lblcr_entry *en, *nxt; | |
466 | ||
1da177e4 LT |
467 | if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) { |
468 | /* do full expiration check */ | |
f728bafb | 469 | ip_vs_lblcr_full_check(svc); |
1da177e4 LT |
470 | tbl->counter = 1; |
471 | goto out; | |
472 | } | |
473 | ||
474 | if (atomic_read(&tbl->entries) <= tbl->max_size) { | |
475 | tbl->counter++; | |
476 | goto out; | |
477 | } | |
478 | ||
479 | goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3; | |
480 | if (goal > tbl->max_size/2) | |
481 | goal = tbl->max_size/2; | |
482 | ||
483 | for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) { | |
484 | j = (j + 1) & IP_VS_LBLCR_TAB_MASK; | |
485 | ||
f728bafb | 486 | write_lock(&svc->sched_lock); |
1da177e4 LT |
487 | list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) { |
488 | if (time_before(now, en->lastuse+ENTRY_TIMEOUT)) | |
489 | continue; | |
490 | ||
491 | ip_vs_lblcr_free(en); | |
492 | atomic_dec(&tbl->entries); | |
493 | goal--; | |
494 | } | |
f728bafb | 495 | write_unlock(&svc->sched_lock); |
1da177e4 LT |
496 | if (goal <= 0) |
497 | break; | |
498 | } | |
499 | tbl->rover = j; | |
500 | ||
501 | out: | |
502 | mod_timer(&tbl->periodic_timer, jiffies+CHECK_EXPIRE_INTERVAL); | |
503 | } | |
504 | ||
1da177e4 LT |
505 | static int ip_vs_lblcr_init_svc(struct ip_vs_service *svc) |
506 | { | |
507 | int i; | |
508 | struct ip_vs_lblcr_table *tbl; | |
509 | ||
510 | /* | |
511 | * Allocate the ip_vs_lblcr_table for this service | |
512 | */ | |
f728bafb | 513 | tbl = kmalloc(sizeof(*tbl), GFP_ATOMIC); |
1da177e4 | 514 | if (tbl == NULL) { |
1e3e238e | 515 | pr_err("%s(): no memory\n", __func__); |
1da177e4 LT |
516 | return -ENOMEM; |
517 | } | |
518 | svc->sched_data = tbl; | |
519 | IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) allocated for " | |
f728bafb | 520 | "current service\n", sizeof(*tbl)); |
1da177e4 LT |
521 | |
522 | /* | |
523 | * Initialize the hash buckets | |
524 | */ | |
525 | for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) { | |
526 | INIT_LIST_HEAD(&tbl->bucket[i]); | |
527 | } | |
1da177e4 LT |
528 | tbl->max_size = IP_VS_LBLCR_TAB_SIZE*16; |
529 | tbl->rover = 0; | |
530 | tbl->counter = 1; | |
531 | ||
532 | /* | |
533 | * Hook periodic timer for garbage collection | |
534 | */ | |
b24b8a24 | 535 | setup_timer(&tbl->periodic_timer, ip_vs_lblcr_check_expire, |
f728bafb SW |
536 | (unsigned long)svc); |
537 | mod_timer(&tbl->periodic_timer, jiffies + CHECK_EXPIRE_INTERVAL); | |
1da177e4 | 538 | |
1da177e4 LT |
539 | return 0; |
540 | } | |
541 | ||
542 | ||
543 | static int ip_vs_lblcr_done_svc(struct ip_vs_service *svc) | |
544 | { | |
545 | struct ip_vs_lblcr_table *tbl = svc->sched_data; | |
546 | ||
547 | /* remove periodic timer */ | |
548 | del_timer_sync(&tbl->periodic_timer); | |
549 | ||
550 | /* got to clean up table entries here */ | |
551 | ip_vs_lblcr_flush(tbl); | |
552 | ||
553 | /* release the table itself */ | |
f728bafb | 554 | kfree(tbl); |
1da177e4 | 555 | IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) released\n", |
f728bafb | 556 | sizeof(*tbl)); |
1da177e4 LT |
557 | |
558 | return 0; | |
559 | } | |
560 | ||
561 | ||
1da177e4 | 562 | static inline struct ip_vs_dest * |
44548375 | 563 | __ip_vs_lblcr_schedule(struct ip_vs_service *svc) |
1da177e4 LT |
564 | { |
565 | struct ip_vs_dest *dest, *least; | |
566 | int loh, doh; | |
567 | ||
568 | /* | |
569 | * We think the overhead of processing active connections is fifty | |
570 | * times higher than that of inactive connections in average. (This | |
571 | * fifty times might not be accurate, we will change it later.) We | |
572 | * use the following formula to estimate the overhead: | |
573 | * dest->activeconns*50 + dest->inactconns | |
574 | * and the load: | |
575 | * (dest overhead) / dest->weight | |
576 | * | |
577 | * Remember -- no floats in kernel mode!!! | |
578 | * The comparison of h1*w2 > h2*w1 is equivalent to that of | |
579 | * h1/w1 > h2/w2 | |
580 | * if every weight is larger than zero. | |
581 | * | |
582 | * The server with weight=0 is quiesced and will not receive any | |
583 | * new connection. | |
584 | */ | |
585 | list_for_each_entry(dest, &svc->destinations, n_list) { | |
586 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
587 | continue; | |
588 | ||
589 | if (atomic_read(&dest->weight) > 0) { | |
590 | least = dest; | |
591 | loh = atomic_read(&least->activeconns) * 50 | |
592 | + atomic_read(&least->inactconns); | |
593 | goto nextstage; | |
594 | } | |
595 | } | |
596 | return NULL; | |
597 | ||
598 | /* | |
599 | * Find the destination with the least load. | |
600 | */ | |
601 | nextstage: | |
602 | list_for_each_entry_continue(dest, &svc->destinations, n_list) { | |
603 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
604 | continue; | |
605 | ||
606 | doh = atomic_read(&dest->activeconns) * 50 | |
607 | + atomic_read(&dest->inactconns); | |
608 | if (loh * atomic_read(&dest->weight) > | |
609 | doh * atomic_read(&least->weight)) { | |
610 | least = dest; | |
611 | loh = doh; | |
612 | } | |
613 | } | |
614 | ||
44548375 JV |
615 | IP_VS_DBG_BUF(6, "LBLCR: server %s:%d " |
616 | "activeconns %d refcnt %d weight %d overhead %d\n", | |
617 | IP_VS_DBG_ADDR(least->af, &least->addr), | |
618 | ntohs(least->port), | |
619 | atomic_read(&least->activeconns), | |
620 | atomic_read(&least->refcnt), | |
621 | atomic_read(&least->weight), loh); | |
1da177e4 LT |
622 | |
623 | return least; | |
624 | } | |
625 | ||
626 | ||
627 | /* | |
628 | * If this destination server is overloaded and there is a less loaded | |
629 | * server, then return true. | |
630 | */ | |
631 | static inline int | |
632 | is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc) | |
633 | { | |
634 | if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) { | |
635 | struct ip_vs_dest *d; | |
636 | ||
637 | list_for_each_entry(d, &svc->destinations, n_list) { | |
638 | if (atomic_read(&d->activeconns)*2 | |
639 | < atomic_read(&d->weight)) { | |
640 | return 1; | |
641 | } | |
642 | } | |
643 | } | |
644 | return 0; | |
645 | } | |
646 | ||
647 | ||
648 | /* | |
649 | * Locality-Based (weighted) Least-Connection scheduling | |
650 | */ | |
651 | static struct ip_vs_dest * | |
652 | ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb) | |
653 | { | |
f728bafb | 654 | struct ip_vs_lblcr_table *tbl = svc->sched_data; |
44548375 | 655 | struct ip_vs_iphdr iph; |
f728bafb SW |
656 | struct ip_vs_dest *dest = NULL; |
657 | struct ip_vs_lblcr_entry *en; | |
1da177e4 | 658 | |
44548375 JV |
659 | ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph); |
660 | ||
1e3e238e | 661 | IP_VS_DBG(6, "%s(): Scheduling...\n", __func__); |
1da177e4 | 662 | |
f728bafb SW |
663 | /* First look in our cache */ |
664 | read_lock(&svc->sched_lock); | |
44548375 | 665 | en = ip_vs_lblcr_get(svc->af, tbl, &iph.daddr); |
f728bafb SW |
666 | if (en) { |
667 | /* We only hold a read lock, but this is atomic */ | |
668 | en->lastuse = jiffies; | |
669 | ||
670 | /* Get the least loaded destination */ | |
671 | read_lock(&en->set.lock); | |
1da177e4 | 672 | dest = ip_vs_dest_set_min(&en->set); |
f728bafb SW |
673 | read_unlock(&en->set.lock); |
674 | ||
675 | /* More than one destination + enough time passed by, cleanup */ | |
1da177e4 | 676 | if (atomic_read(&en->set.size) > 1 && |
f728bafb SW |
677 | time_after(jiffies, en->set.lastmod + |
678 | sysctl_ip_vs_lblcr_expiration)) { | |
1da177e4 | 679 | struct ip_vs_dest *m; |
f728bafb SW |
680 | |
681 | write_lock(&en->set.lock); | |
1da177e4 LT |
682 | m = ip_vs_dest_set_max(&en->set); |
683 | if (m) | |
684 | ip_vs_dest_set_erase(&en->set, m); | |
f728bafb SW |
685 | write_unlock(&en->set.lock); |
686 | } | |
687 | ||
688 | /* If the destination is not overloaded, use it */ | |
689 | if (dest && !is_overloaded(dest, svc)) { | |
690 | read_unlock(&svc->sched_lock); | |
691 | goto out; | |
692 | } | |
693 | ||
694 | /* The cache entry is invalid, time to schedule */ | |
44548375 | 695 | dest = __ip_vs_lblcr_schedule(svc); |
f728bafb | 696 | if (!dest) { |
68888d10 | 697 | IP_VS_ERR_RL("LBLCR: no destination available\n"); |
f728bafb SW |
698 | read_unlock(&svc->sched_lock); |
699 | return NULL; | |
1da177e4 | 700 | } |
f728bafb SW |
701 | |
702 | /* Update our cache entry */ | |
703 | write_lock(&en->set.lock); | |
704 | ip_vs_dest_set_insert(&en->set, dest); | |
705 | write_unlock(&en->set.lock); | |
706 | } | |
707 | read_unlock(&svc->sched_lock); | |
708 | ||
709 | if (dest) | |
710 | goto out; | |
711 | ||
712 | /* No cache entry, time to schedule */ | |
44548375 | 713 | dest = __ip_vs_lblcr_schedule(svc); |
f728bafb SW |
714 | if (!dest) { |
715 | IP_VS_DBG(1, "no destination available\n"); | |
716 | return NULL; | |
1da177e4 | 717 | } |
1da177e4 | 718 | |
f728bafb SW |
719 | /* If we fail to create a cache entry, we'll just use the valid dest */ |
720 | write_lock(&svc->sched_lock); | |
44548375 | 721 | ip_vs_lblcr_new(tbl, &iph.daddr, dest); |
f728bafb SW |
722 | write_unlock(&svc->sched_lock); |
723 | ||
724 | out: | |
44548375 JV |
725 | IP_VS_DBG_BUF(6, "LBLCR: destination IP address %s --> server %s:%d\n", |
726 | IP_VS_DBG_ADDR(svc->af, &iph.daddr), | |
727 | IP_VS_DBG_ADDR(svc->af, &dest->addr), ntohs(dest->port)); | |
1da177e4 LT |
728 | |
729 | return dest; | |
730 | } | |
731 | ||
732 | ||
733 | /* | |
734 | * IPVS LBLCR Scheduler structure | |
735 | */ | |
736 | static struct ip_vs_scheduler ip_vs_lblcr_scheduler = | |
737 | { | |
738 | .name = "lblcr", | |
739 | .refcnt = ATOMIC_INIT(0), | |
740 | .module = THIS_MODULE, | |
d149ccc9 | 741 | .n_list = LIST_HEAD_INIT(ip_vs_lblcr_scheduler.n_list), |
1da177e4 LT |
742 | .init_service = ip_vs_lblcr_init_svc, |
743 | .done_service = ip_vs_lblcr_done_svc, | |
1da177e4 LT |
744 | .schedule = ip_vs_lblcr_schedule, |
745 | }; | |
746 | ||
61b1ab45 HS |
747 | /* |
748 | * per netns init. | |
749 | */ | |
750 | static int __net_init __ip_vs_lblcr_init(struct net *net) | |
751 | { | |
752 | if (!net_eq(net, &init_net)) /* netns not enabled yet */ | |
753 | return -EPERM; | |
754 | ||
755 | sysctl_header = register_net_sysctl_table(net, net_vs_ctl_path, | |
756 | vs_vars_table); | |
757 | if (!sysctl_header) | |
758 | return -ENOMEM; | |
759 | ||
760 | return 0; | |
761 | } | |
762 | ||
763 | static void __net_exit __ip_vs_lblcr_exit(struct net *net) | |
764 | { | |
765 | if (!net_eq(net, &init_net)) /* netns not enabled yet */ | |
766 | return; | |
767 | ||
768 | unregister_net_sysctl_table(sysctl_header); | |
769 | } | |
770 | ||
771 | static struct pernet_operations ip_vs_lblcr_ops = { | |
772 | .init = __ip_vs_lblcr_init, | |
773 | .exit = __ip_vs_lblcr_exit, | |
774 | }; | |
1da177e4 LT |
775 | |
776 | static int __init ip_vs_lblcr_init(void) | |
777 | { | |
a014bc8f PE |
778 | int ret; |
779 | ||
61b1ab45 HS |
780 | ret = register_pernet_subsys(&ip_vs_lblcr_ops); |
781 | if (ret) | |
782 | return ret; | |
783 | ||
a014bc8f PE |
784 | ret = register_ip_vs_scheduler(&ip_vs_lblcr_scheduler); |
785 | if (ret) | |
61b1ab45 | 786 | unregister_pernet_subsys(&ip_vs_lblcr_ops); |
a014bc8f | 787 | return ret; |
1da177e4 LT |
788 | } |
789 | ||
1da177e4 LT |
790 | static void __exit ip_vs_lblcr_cleanup(void) |
791 | { | |
1da177e4 | 792 | unregister_ip_vs_scheduler(&ip_vs_lblcr_scheduler); |
61b1ab45 | 793 | unregister_pernet_subsys(&ip_vs_lblcr_ops); |
1da177e4 LT |
794 | } |
795 | ||
796 | ||
797 | module_init(ip_vs_lblcr_init); | |
798 | module_exit(ip_vs_lblcr_cleanup); | |
799 | MODULE_LICENSE("GPL"); |