]>
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 | ||
b27d777e SH |
66 | #define DEFAULT_EXPIRATION (24*60*60*HZ) |
67 | ||
1da177e4 LT |
68 | /* |
69 | * It is for full expiration check. | |
70 | * When there is no partial expiration check (garbage collection) | |
71 | * in a half hour, do a full expiration check to collect stale | |
72 | * entries that haven't been touched for a day. | |
73 | */ | |
74 | #define COUNT_FOR_FULL_EXPIRATION 30 | |
1da177e4 LT |
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 | 154 | /* |
25985edc | 155 | * We don't kfree dest because it is referred either |
1da177e4 LT |
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)) { | |
b552f7e3 | 183 | loh = ip_vs_dest_conn_overhead(least); |
1da177e4 LT |
184 | goto nextstage; |
185 | } | |
186 | } | |
1da177e4 LT |
187 | return NULL; |
188 | ||
189 | /* find the destination with the weighted least load */ | |
190 | nextstage: | |
51f0bc78 | 191 | list_for_each_entry(e, &set->list, list) { |
1da177e4 LT |
192 | dest = e->dest; |
193 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
194 | continue; | |
195 | ||
b552f7e3 | 196 | doh = ip_vs_dest_conn_overhead(dest); |
1da177e4 LT |
197 | if ((loh * atomic_read(&dest->weight) > |
198 | doh * atomic_read(&least->weight)) | |
199 | && (dest->flags & IP_VS_DEST_F_AVAILABLE)) { | |
200 | least = dest; | |
201 | loh = doh; | |
202 | } | |
203 | } | |
1da177e4 | 204 | |
1e3e238e | 205 | IP_VS_DBG_BUF(6, "%s(): server %s:%d " |
44548375 | 206 | "activeconns %d refcnt %d weight %d overhead %d\n", |
1e3e238e | 207 | __func__, |
44548375 JV |
208 | IP_VS_DBG_ADDR(least->af, &least->addr), |
209 | ntohs(least->port), | |
210 | atomic_read(&least->activeconns), | |
211 | atomic_read(&least->refcnt), | |
212 | atomic_read(&least->weight), loh); | |
1da177e4 LT |
213 | return least; |
214 | } | |
215 | ||
216 | ||
217 | /* get weighted most-connection node in the destination set */ | |
218 | static inline struct ip_vs_dest *ip_vs_dest_set_max(struct ip_vs_dest_set *set) | |
219 | { | |
51f0bc78 | 220 | register struct ip_vs_dest_set_elem *e; |
1da177e4 LT |
221 | struct ip_vs_dest *dest, *most; |
222 | int moh, doh; | |
223 | ||
224 | if (set == NULL) | |
225 | return NULL; | |
226 | ||
1da177e4 | 227 | /* select the first destination server, whose weight > 0 */ |
51f0bc78 | 228 | list_for_each_entry(e, &set->list, list) { |
1da177e4 LT |
229 | most = e->dest; |
230 | if (atomic_read(&most->weight) > 0) { | |
b552f7e3 | 231 | moh = ip_vs_dest_conn_overhead(most); |
1da177e4 LT |
232 | goto nextstage; |
233 | } | |
234 | } | |
1da177e4 LT |
235 | return NULL; |
236 | ||
237 | /* find the destination with the weighted most load */ | |
238 | nextstage: | |
51f0bc78 | 239 | list_for_each_entry(e, &set->list, list) { |
1da177e4 | 240 | dest = e->dest; |
b552f7e3 | 241 | doh = ip_vs_dest_conn_overhead(dest); |
1da177e4 LT |
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 | } | |
1da177e4 | 250 | |
1e3e238e | 251 | IP_VS_DBG_BUF(6, "%s(): server %s:%d " |
44548375 | 252 | "activeconns %d refcnt %d weight %d overhead %d\n", |
1e3e238e | 253 | __func__, |
44548375 JV |
254 | IP_VS_DBG_ADDR(most->af, &most->addr), ntohs(most->port), |
255 | atomic_read(&most->activeconns), | |
256 | atomic_read(&most->refcnt), | |
257 | atomic_read(&most->weight), moh); | |
1da177e4 LT |
258 | return most; |
259 | } | |
260 | ||
261 | ||
262 | /* | |
263 | * IPVS lblcr entry represents an association between destination | |
264 | * IP address and its destination server set | |
265 | */ | |
266 | struct ip_vs_lblcr_entry { | |
267 | struct list_head list; | |
44548375 JV |
268 | int af; /* address family */ |
269 | union nf_inet_addr addr; /* destination IP address */ | |
1da177e4 LT |
270 | struct ip_vs_dest_set set; /* destination server set */ |
271 | unsigned long lastuse; /* last used time */ | |
272 | }; | |
273 | ||
274 | ||
275 | /* | |
276 | * IPVS lblcr hash table | |
277 | */ | |
278 | struct ip_vs_lblcr_table { | |
1da177e4 LT |
279 | struct list_head bucket[IP_VS_LBLCR_TAB_SIZE]; /* hash bucket */ |
280 | atomic_t entries; /* number of entries */ | |
281 | int max_size; /* maximum size of entries */ | |
282 | struct timer_list periodic_timer; /* collect stale entries */ | |
283 | int rover; /* rover for expire check */ | |
284 | int counter; /* counter for no expire */ | |
285 | }; | |
286 | ||
287 | ||
fb1de432 | 288 | #ifdef CONFIG_SYSCTL |
1da177e4 LT |
289 | /* |
290 | * IPVS LBLCR sysctl table | |
291 | */ | |
292 | ||
293 | static ctl_table vs_vars_table[] = { | |
294 | { | |
1da177e4 | 295 | .procname = "lblcr_expiration", |
d0a1eef9 | 296 | .data = NULL, |
1da177e4 | 297 | .maxlen = sizeof(int), |
e905a9ed | 298 | .mode = 0644, |
6d9f239a | 299 | .proc_handler = proc_dointvec_jiffies, |
1da177e4 | 300 | }, |
f8572d8f | 301 | { } |
1da177e4 | 302 | }; |
fb1de432 | 303 | #endif |
1da177e4 | 304 | |
1da177e4 LT |
305 | static inline void ip_vs_lblcr_free(struct ip_vs_lblcr_entry *en) |
306 | { | |
307 | list_del(&en->list); | |
308 | ip_vs_dest_set_eraseall(&en->set); | |
309 | kfree(en); | |
310 | } | |
311 | ||
312 | ||
313 | /* | |
314 | * Returns hash value for IPVS LBLCR entry | |
315 | */ | |
44548375 JV |
316 | static inline unsigned |
317 | ip_vs_lblcr_hashkey(int af, const union nf_inet_addr *addr) | |
1da177e4 | 318 | { |
44548375 JV |
319 | __be32 addr_fold = addr->ip; |
320 | ||
321 | #ifdef CONFIG_IP_VS_IPV6 | |
322 | if (af == AF_INET6) | |
323 | addr_fold = addr->ip6[0]^addr->ip6[1]^ | |
324 | addr->ip6[2]^addr->ip6[3]; | |
325 | #endif | |
326 | return (ntohl(addr_fold)*2654435761UL) & IP_VS_LBLCR_TAB_MASK; | |
1da177e4 LT |
327 | } |
328 | ||
329 | ||
330 | /* | |
331 | * Hash an entry in the ip_vs_lblcr_table. | |
332 | * returns bool success. | |
333 | */ | |
f728bafb | 334 | static void |
1da177e4 LT |
335 | ip_vs_lblcr_hash(struct ip_vs_lblcr_table *tbl, struct ip_vs_lblcr_entry *en) |
336 | { | |
44548375 | 337 | unsigned hash = ip_vs_lblcr_hashkey(en->af, &en->addr); |
1da177e4 | 338 | |
1da177e4 LT |
339 | list_add(&en->list, &tbl->bucket[hash]); |
340 | atomic_inc(&tbl->entries); | |
1da177e4 LT |
341 | } |
342 | ||
343 | ||
1da177e4 | 344 | /* |
f728bafb SW |
345 | * Get ip_vs_lblcr_entry associated with supplied parameters. Called under |
346 | * read lock. | |
1da177e4 LT |
347 | */ |
348 | static inline struct ip_vs_lblcr_entry * | |
44548375 JV |
349 | ip_vs_lblcr_get(int af, struct ip_vs_lblcr_table *tbl, |
350 | const union nf_inet_addr *addr) | |
1da177e4 | 351 | { |
44548375 | 352 | unsigned hash = ip_vs_lblcr_hashkey(af, addr); |
1da177e4 LT |
353 | struct ip_vs_lblcr_entry *en; |
354 | ||
f728bafb | 355 | list_for_each_entry(en, &tbl->bucket[hash], list) |
44548375 | 356 | if (ip_vs_addr_equal(af, &en->addr, addr)) |
f728bafb SW |
357 | return en; |
358 | ||
359 | return NULL; | |
360 | } | |
1da177e4 | 361 | |
1da177e4 | 362 | |
f728bafb SW |
363 | /* |
364 | * Create or update an ip_vs_lblcr_entry, which is a mapping of a destination | |
365 | * IP address to a server. Called under write lock. | |
366 | */ | |
367 | static inline struct ip_vs_lblcr_entry * | |
44548375 | 368 | ip_vs_lblcr_new(struct ip_vs_lblcr_table *tbl, const union nf_inet_addr *daddr, |
f728bafb SW |
369 | struct ip_vs_dest *dest) |
370 | { | |
371 | struct ip_vs_lblcr_entry *en; | |
372 | ||
44548375 | 373 | en = ip_vs_lblcr_get(dest->af, tbl, daddr); |
f728bafb SW |
374 | if (!en) { |
375 | en = kmalloc(sizeof(*en), GFP_ATOMIC); | |
376 | if (!en) { | |
1e3e238e | 377 | pr_err("%s(): no memory\n", __func__); |
f728bafb | 378 | return NULL; |
1da177e4 | 379 | } |
f728bafb | 380 | |
44548375 JV |
381 | en->af = dest->af; |
382 | ip_vs_addr_copy(dest->af, &en->addr, daddr); | |
f728bafb SW |
383 | en->lastuse = jiffies; |
384 | ||
421f91d2 | 385 | /* initialize its dest set */ |
f728bafb | 386 | atomic_set(&(en->set.size), 0); |
51f0bc78 | 387 | INIT_LIST_HEAD(&en->set.list); |
f728bafb SW |
388 | rwlock_init(&en->set.lock); |
389 | ||
390 | ip_vs_lblcr_hash(tbl, en); | |
1da177e4 LT |
391 | } |
392 | ||
f728bafb SW |
393 | write_lock(&en->set.lock); |
394 | ip_vs_dest_set_insert(&en->set, dest); | |
395 | write_unlock(&en->set.lock); | |
1da177e4 | 396 | |
f728bafb | 397 | return en; |
1da177e4 LT |
398 | } |
399 | ||
400 | ||
401 | /* | |
402 | * Flush all the entries of the specified table. | |
403 | */ | |
404 | static void ip_vs_lblcr_flush(struct ip_vs_lblcr_table *tbl) | |
405 | { | |
406 | int i; | |
407 | struct ip_vs_lblcr_entry *en, *nxt; | |
408 | ||
f728bafb | 409 | /* No locking required, only called during cleanup. */ |
1da177e4 | 410 | for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) { |
1da177e4 LT |
411 | list_for_each_entry_safe(en, nxt, &tbl->bucket[i], list) { |
412 | ip_vs_lblcr_free(en); | |
1da177e4 | 413 | } |
1da177e4 LT |
414 | } |
415 | } | |
416 | ||
b27d777e SH |
417 | static int sysctl_lblcr_expiration(struct ip_vs_service *svc) |
418 | { | |
419 | #ifdef CONFIG_SYSCTL | |
420 | struct netns_ipvs *ipvs = net_ipvs(svc->net); | |
421 | return ipvs->sysctl_lblcr_expiration; | |
422 | #else | |
423 | return DEFAULT_EXPIRATION; | |
424 | #endif | |
425 | } | |
1da177e4 | 426 | |
f728bafb | 427 | static inline void ip_vs_lblcr_full_check(struct ip_vs_service *svc) |
1da177e4 | 428 | { |
f728bafb | 429 | struct ip_vs_lblcr_table *tbl = svc->sched_data; |
1da177e4 LT |
430 | unsigned long now = jiffies; |
431 | int i, j; | |
432 | struct ip_vs_lblcr_entry *en, *nxt; | |
433 | ||
434 | for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) { | |
435 | j = (j + 1) & IP_VS_LBLCR_TAB_MASK; | |
436 | ||
f728bafb | 437 | write_lock(&svc->sched_lock); |
1da177e4 | 438 | list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) { |
b27d777e SH |
439 | if (time_after(en->lastuse + |
440 | sysctl_lblcr_expiration(svc), now)) | |
1da177e4 LT |
441 | continue; |
442 | ||
443 | ip_vs_lblcr_free(en); | |
444 | atomic_dec(&tbl->entries); | |
445 | } | |
f728bafb | 446 | write_unlock(&svc->sched_lock); |
1da177e4 LT |
447 | } |
448 | tbl->rover = j; | |
449 | } | |
450 | ||
451 | ||
452 | /* | |
453 | * Periodical timer handler for IPVS lblcr table | |
454 | * It is used to collect stale entries when the number of entries | |
455 | * exceeds the maximum size of the table. | |
456 | * | |
457 | * Fixme: we probably need more complicated algorithm to collect | |
458 | * entries that have not been used for a long time even | |
459 | * if the number of entries doesn't exceed the maximum size | |
460 | * of the table. | |
461 | * The full expiration check is for this purpose now. | |
462 | */ | |
463 | static void ip_vs_lblcr_check_expire(unsigned long data) | |
464 | { | |
f728bafb SW |
465 | struct ip_vs_service *svc = (struct ip_vs_service *) data; |
466 | struct ip_vs_lblcr_table *tbl = svc->sched_data; | |
1da177e4 LT |
467 | unsigned long now = jiffies; |
468 | int goal; | |
469 | int i, j; | |
470 | struct ip_vs_lblcr_entry *en, *nxt; | |
471 | ||
1da177e4 LT |
472 | if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) { |
473 | /* do full expiration check */ | |
f728bafb | 474 | ip_vs_lblcr_full_check(svc); |
1da177e4 LT |
475 | tbl->counter = 1; |
476 | goto out; | |
477 | } | |
478 | ||
479 | if (atomic_read(&tbl->entries) <= tbl->max_size) { | |
480 | tbl->counter++; | |
481 | goto out; | |
482 | } | |
483 | ||
484 | goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3; | |
485 | if (goal > tbl->max_size/2) | |
486 | goal = tbl->max_size/2; | |
487 | ||
488 | for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) { | |
489 | j = (j + 1) & IP_VS_LBLCR_TAB_MASK; | |
490 | ||
f728bafb | 491 | write_lock(&svc->sched_lock); |
1da177e4 LT |
492 | list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) { |
493 | if (time_before(now, en->lastuse+ENTRY_TIMEOUT)) | |
494 | continue; | |
495 | ||
496 | ip_vs_lblcr_free(en); | |
497 | atomic_dec(&tbl->entries); | |
498 | goal--; | |
499 | } | |
f728bafb | 500 | write_unlock(&svc->sched_lock); |
1da177e4 LT |
501 | if (goal <= 0) |
502 | break; | |
503 | } | |
504 | tbl->rover = j; | |
505 | ||
506 | out: | |
507 | mod_timer(&tbl->periodic_timer, jiffies+CHECK_EXPIRE_INTERVAL); | |
508 | } | |
509 | ||
1da177e4 LT |
510 | static int ip_vs_lblcr_init_svc(struct ip_vs_service *svc) |
511 | { | |
512 | int i; | |
513 | struct ip_vs_lblcr_table *tbl; | |
514 | ||
515 | /* | |
516 | * Allocate the ip_vs_lblcr_table for this service | |
517 | */ | |
f728bafb | 518 | tbl = kmalloc(sizeof(*tbl), GFP_ATOMIC); |
1da177e4 | 519 | if (tbl == NULL) { |
1e3e238e | 520 | pr_err("%s(): no memory\n", __func__); |
1da177e4 LT |
521 | return -ENOMEM; |
522 | } | |
523 | svc->sched_data = tbl; | |
524 | IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) allocated for " | |
f728bafb | 525 | "current service\n", sizeof(*tbl)); |
1da177e4 LT |
526 | |
527 | /* | |
528 | * Initialize the hash buckets | |
529 | */ | |
530 | for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) { | |
531 | INIT_LIST_HEAD(&tbl->bucket[i]); | |
532 | } | |
1da177e4 LT |
533 | tbl->max_size = IP_VS_LBLCR_TAB_SIZE*16; |
534 | tbl->rover = 0; | |
535 | tbl->counter = 1; | |
536 | ||
537 | /* | |
538 | * Hook periodic timer for garbage collection | |
539 | */ | |
b24b8a24 | 540 | setup_timer(&tbl->periodic_timer, ip_vs_lblcr_check_expire, |
f728bafb SW |
541 | (unsigned long)svc); |
542 | mod_timer(&tbl->periodic_timer, jiffies + CHECK_EXPIRE_INTERVAL); | |
1da177e4 | 543 | |
1da177e4 LT |
544 | return 0; |
545 | } | |
546 | ||
547 | ||
548 | static int ip_vs_lblcr_done_svc(struct ip_vs_service *svc) | |
549 | { | |
550 | struct ip_vs_lblcr_table *tbl = svc->sched_data; | |
551 | ||
552 | /* remove periodic timer */ | |
553 | del_timer_sync(&tbl->periodic_timer); | |
554 | ||
555 | /* got to clean up table entries here */ | |
556 | ip_vs_lblcr_flush(tbl); | |
557 | ||
558 | /* release the table itself */ | |
f728bafb | 559 | kfree(tbl); |
1da177e4 | 560 | IP_VS_DBG(6, "LBLCR hash table (memory=%Zdbytes) released\n", |
f728bafb | 561 | sizeof(*tbl)); |
1da177e4 LT |
562 | |
563 | return 0; | |
564 | } | |
565 | ||
566 | ||
1da177e4 | 567 | static inline struct ip_vs_dest * |
44548375 | 568 | __ip_vs_lblcr_schedule(struct ip_vs_service *svc) |
1da177e4 LT |
569 | { |
570 | struct ip_vs_dest *dest, *least; | |
571 | int loh, doh; | |
572 | ||
573 | /* | |
b552f7e3 | 574 | * We use the following formula to estimate the load: |
1da177e4 LT |
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; | |
b552f7e3 | 591 | loh = ip_vs_dest_conn_overhead(least); |
1da177e4 LT |
592 | goto nextstage; |
593 | } | |
594 | } | |
595 | return NULL; | |
596 | ||
597 | /* | |
598 | * Find the destination with the least load. | |
599 | */ | |
600 | nextstage: | |
601 | list_for_each_entry_continue(dest, &svc->destinations, n_list) { | |
602 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
603 | continue; | |
604 | ||
b552f7e3 | 605 | doh = ip_vs_dest_conn_overhead(dest); |
1da177e4 LT |
606 | if (loh * atomic_read(&dest->weight) > |
607 | doh * atomic_read(&least->weight)) { | |
608 | least = dest; | |
609 | loh = doh; | |
610 | } | |
611 | } | |
612 | ||
44548375 JV |
613 | IP_VS_DBG_BUF(6, "LBLCR: server %s:%d " |
614 | "activeconns %d refcnt %d weight %d overhead %d\n", | |
615 | IP_VS_DBG_ADDR(least->af, &least->addr), | |
616 | ntohs(least->port), | |
617 | atomic_read(&least->activeconns), | |
618 | atomic_read(&least->refcnt), | |
619 | atomic_read(&least->weight), loh); | |
1da177e4 LT |
620 | |
621 | return least; | |
622 | } | |
623 | ||
624 | ||
625 | /* | |
626 | * If this destination server is overloaded and there is a less loaded | |
627 | * server, then return true. | |
628 | */ | |
629 | static inline int | |
630 | is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc) | |
631 | { | |
632 | if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) { | |
633 | struct ip_vs_dest *d; | |
634 | ||
635 | list_for_each_entry(d, &svc->destinations, n_list) { | |
636 | if (atomic_read(&d->activeconns)*2 | |
637 | < atomic_read(&d->weight)) { | |
638 | return 1; | |
639 | } | |
640 | } | |
641 | } | |
642 | return 0; | |
643 | } | |
644 | ||
645 | ||
646 | /* | |
647 | * Locality-Based (weighted) Least-Connection scheduling | |
648 | */ | |
649 | static struct ip_vs_dest * | |
650 | ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb) | |
651 | { | |
f728bafb | 652 | struct ip_vs_lblcr_table *tbl = svc->sched_data; |
44548375 | 653 | struct ip_vs_iphdr iph; |
f728bafb SW |
654 | struct ip_vs_dest *dest = NULL; |
655 | struct ip_vs_lblcr_entry *en; | |
1da177e4 | 656 | |
44548375 JV |
657 | ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph); |
658 | ||
1e3e238e | 659 | IP_VS_DBG(6, "%s(): Scheduling...\n", __func__); |
1da177e4 | 660 | |
f728bafb SW |
661 | /* First look in our cache */ |
662 | read_lock(&svc->sched_lock); | |
44548375 | 663 | en = ip_vs_lblcr_get(svc->af, tbl, &iph.daddr); |
f728bafb SW |
664 | if (en) { |
665 | /* We only hold a read lock, but this is atomic */ | |
666 | en->lastuse = jiffies; | |
667 | ||
668 | /* Get the least loaded destination */ | |
669 | read_lock(&en->set.lock); | |
1da177e4 | 670 | dest = ip_vs_dest_set_min(&en->set); |
f728bafb SW |
671 | read_unlock(&en->set.lock); |
672 | ||
673 | /* More than one destination + enough time passed by, cleanup */ | |
1da177e4 | 674 | if (atomic_read(&en->set.size) > 1 && |
f728bafb | 675 | time_after(jiffies, en->set.lastmod + |
b27d777e | 676 | sysctl_lblcr_expiration(svc))) { |
1da177e4 | 677 | struct ip_vs_dest *m; |
f728bafb SW |
678 | |
679 | write_lock(&en->set.lock); | |
1da177e4 LT |
680 | m = ip_vs_dest_set_max(&en->set); |
681 | if (m) | |
682 | ip_vs_dest_set_erase(&en->set, m); | |
f728bafb SW |
683 | write_unlock(&en->set.lock); |
684 | } | |
685 | ||
686 | /* If the destination is not overloaded, use it */ | |
687 | if (dest && !is_overloaded(dest, svc)) { | |
688 | read_unlock(&svc->sched_lock); | |
689 | goto out; | |
690 | } | |
691 | ||
692 | /* The cache entry is invalid, time to schedule */ | |
44548375 | 693 | dest = __ip_vs_lblcr_schedule(svc); |
f728bafb | 694 | if (!dest) { |
41ac51ee | 695 | ip_vs_scheduler_err(svc, "no destination available"); |
f728bafb SW |
696 | read_unlock(&svc->sched_lock); |
697 | return NULL; | |
1da177e4 | 698 | } |
f728bafb SW |
699 | |
700 | /* Update our cache entry */ | |
701 | write_lock(&en->set.lock); | |
702 | ip_vs_dest_set_insert(&en->set, dest); | |
703 | write_unlock(&en->set.lock); | |
704 | } | |
705 | read_unlock(&svc->sched_lock); | |
706 | ||
707 | if (dest) | |
708 | goto out; | |
709 | ||
710 | /* No cache entry, time to schedule */ | |
44548375 | 711 | dest = __ip_vs_lblcr_schedule(svc); |
f728bafb SW |
712 | if (!dest) { |
713 | IP_VS_DBG(1, "no destination available\n"); | |
714 | return NULL; | |
1da177e4 | 715 | } |
1da177e4 | 716 | |
f728bafb SW |
717 | /* If we fail to create a cache entry, we'll just use the valid dest */ |
718 | write_lock(&svc->sched_lock); | |
44548375 | 719 | ip_vs_lblcr_new(tbl, &iph.daddr, dest); |
f728bafb SW |
720 | write_unlock(&svc->sched_lock); |
721 | ||
722 | out: | |
44548375 JV |
723 | IP_VS_DBG_BUF(6, "LBLCR: destination IP address %s --> server %s:%d\n", |
724 | IP_VS_DBG_ADDR(svc->af, &iph.daddr), | |
725 | IP_VS_DBG_ADDR(svc->af, &dest->addr), ntohs(dest->port)); | |
1da177e4 LT |
726 | |
727 | return dest; | |
728 | } | |
729 | ||
730 | ||
731 | /* | |
732 | * IPVS LBLCR Scheduler structure | |
733 | */ | |
734 | static struct ip_vs_scheduler ip_vs_lblcr_scheduler = | |
735 | { | |
736 | .name = "lblcr", | |
737 | .refcnt = ATOMIC_INIT(0), | |
738 | .module = THIS_MODULE, | |
d149ccc9 | 739 | .n_list = LIST_HEAD_INIT(ip_vs_lblcr_scheduler.n_list), |
1da177e4 LT |
740 | .init_service = ip_vs_lblcr_init_svc, |
741 | .done_service = ip_vs_lblcr_done_svc, | |
1da177e4 LT |
742 | .schedule = ip_vs_lblcr_schedule, |
743 | }; | |
744 | ||
61b1ab45 HS |
745 | /* |
746 | * per netns init. | |
747 | */ | |
fb1de432 | 748 | #ifdef CONFIG_SYSCTL |
61b1ab45 HS |
749 | static int __net_init __ip_vs_lblcr_init(struct net *net) |
750 | { | |
d0a1eef9 HS |
751 | struct netns_ipvs *ipvs = net_ipvs(net); |
752 | ||
753 | if (!net_eq(net, &init_net)) { | |
754 | ipvs->lblcr_ctl_table = kmemdup(vs_vars_table, | |
755 | sizeof(vs_vars_table), | |
756 | GFP_KERNEL); | |
757 | if (ipvs->lblcr_ctl_table == NULL) | |
0443929f | 758 | return -ENOMEM; |
d0a1eef9 HS |
759 | } else |
760 | ipvs->lblcr_ctl_table = vs_vars_table; | |
b27d777e | 761 | ipvs->sysctl_lblcr_expiration = DEFAULT_EXPIRATION; |
d0a1eef9 HS |
762 | ipvs->lblcr_ctl_table[0].data = &ipvs->sysctl_lblcr_expiration; |
763 | ||
764 | ipvs->lblcr_ctl_header = | |
765 | register_net_sysctl_table(net, net_vs_ctl_path, | |
766 | ipvs->lblcr_ctl_table); | |
0443929f SH |
767 | if (!ipvs->lblcr_ctl_header) { |
768 | if (!net_eq(net, &init_net)) | |
769 | kfree(ipvs->lblcr_ctl_table); | |
770 | return -ENOMEM; | |
771 | } | |
61b1ab45 HS |
772 | |
773 | return 0; | |
774 | } | |
775 | ||
776 | static void __net_exit __ip_vs_lblcr_exit(struct net *net) | |
777 | { | |
d0a1eef9 HS |
778 | struct netns_ipvs *ipvs = net_ipvs(net); |
779 | ||
780 | unregister_net_sysctl_table(ipvs->lblcr_ctl_header); | |
61b1ab45 | 781 | |
d0a1eef9 HS |
782 | if (!net_eq(net, &init_net)) |
783 | kfree(ipvs->lblcr_ctl_table); | |
61b1ab45 HS |
784 | } |
785 | ||
fb1de432 SH |
786 | #else |
787 | ||
788 | static int __net_init __ip_vs_lblcr_init(struct net *net) { return 0; } | |
789 | static void __net_exit __ip_vs_lblcr_exit(struct net *net) { } | |
790 | ||
791 | #endif | |
792 | ||
61b1ab45 HS |
793 | static struct pernet_operations ip_vs_lblcr_ops = { |
794 | .init = __ip_vs_lblcr_init, | |
795 | .exit = __ip_vs_lblcr_exit, | |
796 | }; | |
1da177e4 LT |
797 | |
798 | static int __init ip_vs_lblcr_init(void) | |
799 | { | |
a014bc8f PE |
800 | int ret; |
801 | ||
61b1ab45 HS |
802 | ret = register_pernet_subsys(&ip_vs_lblcr_ops); |
803 | if (ret) | |
804 | return ret; | |
805 | ||
a014bc8f PE |
806 | ret = register_ip_vs_scheduler(&ip_vs_lblcr_scheduler); |
807 | if (ret) | |
61b1ab45 | 808 | unregister_pernet_subsys(&ip_vs_lblcr_ops); |
a014bc8f | 809 | return ret; |
1da177e4 LT |
810 | } |
811 | ||
1da177e4 LT |
812 | static void __exit ip_vs_lblcr_cleanup(void) |
813 | { | |
1da177e4 | 814 | unregister_ip_vs_scheduler(&ip_vs_lblcr_scheduler); |
61b1ab45 | 815 | unregister_pernet_subsys(&ip_vs_lblcr_ops); |
1da177e4 LT |
816 | } |
817 | ||
818 | ||
819 | module_init(ip_vs_lblcr_init); | |
820 | module_exit(ip_vs_lblcr_cleanup); | |
821 | MODULE_LICENSE("GPL"); |