Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * IPVS: Locality-Based Least-Connection scheduling module | |
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 | * Martin Hamilton : fixed the terrible locking bugs | |
13 | * *lock(tbl->lock) ==> *lock(&tbl->lock) | |
421f91d2 | 14 | * Wensong Zhang : fixed the uninitialized tbl->lock bug |
1da177e4 LT |
15 | * Wensong Zhang : added doing full expiration check to |
16 | * collect stale entries of 24+ hours when | |
17 | * no partial expire check in a half hour | |
18 | * Julian Anastasov : replaced del_timer call with del_timer_sync | |
19 | * to avoid the possible race between timer | |
20 | * handler and del_timer thread in SMP | |
21 | * | |
22 | */ | |
23 | ||
24 | /* | |
25 | * The lblc algorithm is as follows (pseudo code): | |
26 | * | |
27 | * if cachenode[dest_ip] is null then | |
28 | * n, cachenode[dest_ip] <- {weighted least-conn node}; | |
29 | * else | |
30 | * n <- cachenode[dest_ip]; | |
31 | * if (n is dead) OR | |
32 | * (n.conns>n.weight AND | |
33 | * there is a node m with m.conns<m.weight/2) then | |
34 | * n, cachenode[dest_ip] <- {weighted least-conn node}; | |
35 | * | |
36 | * return n; | |
37 | * | |
38 | * Thanks must go to Wenzhuo Zhang for talking WCCP to me and pushing | |
39 | * me to write this module. | |
40 | */ | |
41 | ||
9aada7ac HE |
42 | #define KMSG_COMPONENT "IPVS" |
43 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt | |
44 | ||
14c85021 | 45 | #include <linux/ip.h> |
5a0e3ad6 | 46 | #include <linux/slab.h> |
1da177e4 LT |
47 | #include <linux/module.h> |
48 | #include <linux/kernel.h> | |
14c85021 | 49 | #include <linux/skbuff.h> |
d7fe0f24 | 50 | #include <linux/jiffies.h> |
1da177e4 LT |
51 | |
52 | /* for sysctl */ | |
53 | #include <linux/fs.h> | |
54 | #include <linux/sysctl.h> | |
55 | ||
56 | #include <net/ip_vs.h> | |
57 | ||
58 | ||
59 | /* | |
60 | * It is for garbage collection of stale IPVS lblc 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 | /* | |
78 | * for IPVS lblc entry hash table | |
79 | */ | |
80 | #ifndef CONFIG_IP_VS_LBLC_TAB_BITS | |
81 | #define CONFIG_IP_VS_LBLC_TAB_BITS 10 | |
82 | #endif | |
83 | #define IP_VS_LBLC_TAB_BITS CONFIG_IP_VS_LBLC_TAB_BITS | |
84 | #define IP_VS_LBLC_TAB_SIZE (1 << IP_VS_LBLC_TAB_BITS) | |
85 | #define IP_VS_LBLC_TAB_MASK (IP_VS_LBLC_TAB_SIZE - 1) | |
86 | ||
87 | ||
88 | /* | |
89 | * IPVS lblc entry represents an association between destination | |
90 | * IP address and its destination server | |
91 | */ | |
92 | struct ip_vs_lblc_entry { | |
93 | struct list_head list; | |
44548375 JV |
94 | int af; /* address family */ |
95 | union nf_inet_addr addr; /* destination IP address */ | |
1da177e4 LT |
96 | struct ip_vs_dest *dest; /* real server (cache) */ |
97 | unsigned long lastuse; /* last used time */ | |
98 | }; | |
99 | ||
100 | ||
101 | /* | |
102 | * IPVS lblc hash table | |
103 | */ | |
104 | struct ip_vs_lblc_table { | |
1da177e4 LT |
105 | struct list_head bucket[IP_VS_LBLC_TAB_SIZE]; /* hash bucket */ |
106 | atomic_t entries; /* number of entries */ | |
107 | int max_size; /* maximum size of entries */ | |
108 | struct timer_list periodic_timer; /* collect stale entries */ | |
109 | int rover; /* rover for expire check */ | |
110 | int counter; /* counter for no expire */ | |
111 | }; | |
112 | ||
113 | ||
114 | /* | |
115 | * IPVS LBLC sysctl table | |
116 | */ | |
fb1de432 | 117 | #ifdef CONFIG_SYSCTL |
1da177e4 LT |
118 | static ctl_table vs_vars_table[] = { |
119 | { | |
1da177e4 | 120 | .procname = "lblc_expiration", |
b6e885dd | 121 | .data = NULL, |
1da177e4 | 122 | .maxlen = sizeof(int), |
e905a9ed | 123 | .mode = 0644, |
6d9f239a | 124 | .proc_handler = proc_dointvec_jiffies, |
1da177e4 | 125 | }, |
f8572d8f | 126 | { } |
1da177e4 | 127 | }; |
fb1de432 | 128 | #endif |
1da177e4 | 129 | |
1da177e4 LT |
130 | static inline void ip_vs_lblc_free(struct ip_vs_lblc_entry *en) |
131 | { | |
132 | list_del(&en->list); | |
133 | /* | |
25985edc | 134 | * We don't kfree dest because it is referred either by its service |
1da177e4 LT |
135 | * or the trash dest list. |
136 | */ | |
137 | atomic_dec(&en->dest->refcnt); | |
138 | kfree(en); | |
139 | } | |
140 | ||
141 | ||
142 | /* | |
143 | * Returns hash value for IPVS LBLC entry | |
144 | */ | |
95c96174 | 145 | static inline unsigned int |
44548375 | 146 | ip_vs_lblc_hashkey(int af, const union nf_inet_addr *addr) |
1da177e4 | 147 | { |
44548375 JV |
148 | __be32 addr_fold = addr->ip; |
149 | ||
150 | #ifdef CONFIG_IP_VS_IPV6 | |
151 | if (af == AF_INET6) | |
152 | addr_fold = addr->ip6[0]^addr->ip6[1]^ | |
153 | addr->ip6[2]^addr->ip6[3]; | |
154 | #endif | |
155 | return (ntohl(addr_fold)*2654435761UL) & IP_VS_LBLC_TAB_MASK; | |
1da177e4 LT |
156 | } |
157 | ||
158 | ||
159 | /* | |
160 | * Hash an entry in the ip_vs_lblc_table. | |
161 | * returns bool success. | |
162 | */ | |
39ac50d0 | 163 | static void |
1da177e4 LT |
164 | ip_vs_lblc_hash(struct ip_vs_lblc_table *tbl, struct ip_vs_lblc_entry *en) |
165 | { | |
95c96174 | 166 | unsigned int hash = ip_vs_lblc_hashkey(en->af, &en->addr); |
1da177e4 | 167 | |
1da177e4 LT |
168 | list_add(&en->list, &tbl->bucket[hash]); |
169 | atomic_inc(&tbl->entries); | |
1da177e4 LT |
170 | } |
171 | ||
172 | ||
1da177e4 | 173 | /* |
39ac50d0 SW |
174 | * Get ip_vs_lblc_entry associated with supplied parameters. Called under read |
175 | * lock | |
1da177e4 LT |
176 | */ |
177 | static inline struct ip_vs_lblc_entry * | |
44548375 JV |
178 | ip_vs_lblc_get(int af, struct ip_vs_lblc_table *tbl, |
179 | const union nf_inet_addr *addr) | |
1da177e4 | 180 | { |
95c96174 | 181 | unsigned int hash = ip_vs_lblc_hashkey(af, addr); |
1da177e4 LT |
182 | struct ip_vs_lblc_entry *en; |
183 | ||
39ac50d0 | 184 | list_for_each_entry(en, &tbl->bucket[hash], list) |
44548375 | 185 | if (ip_vs_addr_equal(af, &en->addr, addr)) |
39ac50d0 | 186 | return en; |
1da177e4 | 187 | |
39ac50d0 SW |
188 | return NULL; |
189 | } | |
1da177e4 | 190 | |
39ac50d0 SW |
191 | |
192 | /* | |
193 | * Create or update an ip_vs_lblc_entry, which is a mapping of a destination IP | |
194 | * address to a server. Called under write lock. | |
195 | */ | |
196 | static inline struct ip_vs_lblc_entry * | |
44548375 | 197 | ip_vs_lblc_new(struct ip_vs_lblc_table *tbl, const union nf_inet_addr *daddr, |
39ac50d0 SW |
198 | struct ip_vs_dest *dest) |
199 | { | |
200 | struct ip_vs_lblc_entry *en; | |
201 | ||
44548375 | 202 | en = ip_vs_lblc_get(dest->af, tbl, daddr); |
39ac50d0 SW |
203 | if (!en) { |
204 | en = kmalloc(sizeof(*en), GFP_ATOMIC); | |
0a9ee813 | 205 | if (!en) |
39ac50d0 | 206 | return NULL; |
1da177e4 | 207 | |
44548375 JV |
208 | en->af = dest->af; |
209 | ip_vs_addr_copy(dest->af, &en->addr, daddr); | |
39ac50d0 | 210 | en->lastuse = jiffies; |
1da177e4 | 211 | |
39ac50d0 SW |
212 | atomic_inc(&dest->refcnt); |
213 | en->dest = dest; | |
214 | ||
215 | ip_vs_lblc_hash(tbl, en); | |
216 | } else if (en->dest != dest) { | |
217 | atomic_dec(&en->dest->refcnt); | |
218 | atomic_inc(&dest->refcnt); | |
219 | en->dest = dest; | |
220 | } | |
221 | ||
222 | return en; | |
1da177e4 LT |
223 | } |
224 | ||
225 | ||
226 | /* | |
227 | * Flush all the entries of the specified table. | |
228 | */ | |
229 | static void ip_vs_lblc_flush(struct ip_vs_lblc_table *tbl) | |
230 | { | |
1da177e4 | 231 | struct ip_vs_lblc_entry *en, *nxt; |
39ac50d0 | 232 | int i; |
1da177e4 LT |
233 | |
234 | for (i=0; i<IP_VS_LBLC_TAB_SIZE; i++) { | |
1da177e4 LT |
235 | list_for_each_entry_safe(en, nxt, &tbl->bucket[i], list) { |
236 | ip_vs_lblc_free(en); | |
237 | atomic_dec(&tbl->entries); | |
238 | } | |
1da177e4 LT |
239 | } |
240 | } | |
241 | ||
b27d777e SH |
242 | static int sysctl_lblc_expiration(struct ip_vs_service *svc) |
243 | { | |
244 | #ifdef CONFIG_SYSCTL | |
245 | struct netns_ipvs *ipvs = net_ipvs(svc->net); | |
246 | return ipvs->sysctl_lblc_expiration; | |
247 | #else | |
248 | return DEFAULT_EXPIRATION; | |
249 | #endif | |
250 | } | |
1da177e4 | 251 | |
39ac50d0 | 252 | static inline void ip_vs_lblc_full_check(struct ip_vs_service *svc) |
1da177e4 | 253 | { |
39ac50d0 SW |
254 | struct ip_vs_lblc_table *tbl = svc->sched_data; |
255 | struct ip_vs_lblc_entry *en, *nxt; | |
1da177e4 LT |
256 | unsigned long now = jiffies; |
257 | int i, j; | |
1da177e4 LT |
258 | |
259 | for (i=0, j=tbl->rover; i<IP_VS_LBLC_TAB_SIZE; i++) { | |
260 | j = (j + 1) & IP_VS_LBLC_TAB_MASK; | |
261 | ||
39ac50d0 | 262 | write_lock(&svc->sched_lock); |
1da177e4 | 263 | list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) { |
e905a9ed | 264 | if (time_before(now, |
b6e885dd | 265 | en->lastuse + |
b27d777e | 266 | sysctl_lblc_expiration(svc))) |
1da177e4 LT |
267 | continue; |
268 | ||
269 | ip_vs_lblc_free(en); | |
270 | atomic_dec(&tbl->entries); | |
271 | } | |
39ac50d0 | 272 | write_unlock(&svc->sched_lock); |
1da177e4 LT |
273 | } |
274 | tbl->rover = j; | |
275 | } | |
276 | ||
277 | ||
278 | /* | |
279 | * Periodical timer handler for IPVS lblc table | |
280 | * It is used to collect stale entries when the number of entries | |
281 | * exceeds the maximum size of the table. | |
282 | * | |
283 | * Fixme: we probably need more complicated algorithm to collect | |
284 | * entries that have not been used for a long time even | |
285 | * if the number of entries doesn't exceed the maximum size | |
286 | * of the table. | |
287 | * The full expiration check is for this purpose now. | |
288 | */ | |
289 | static void ip_vs_lblc_check_expire(unsigned long data) | |
290 | { | |
39ac50d0 SW |
291 | struct ip_vs_service *svc = (struct ip_vs_service *) data; |
292 | struct ip_vs_lblc_table *tbl = svc->sched_data; | |
1da177e4 LT |
293 | unsigned long now = jiffies; |
294 | int goal; | |
295 | int i, j; | |
296 | struct ip_vs_lblc_entry *en, *nxt; | |
297 | ||
1da177e4 LT |
298 | if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) { |
299 | /* do full expiration check */ | |
39ac50d0 | 300 | ip_vs_lblc_full_check(svc); |
1da177e4 LT |
301 | tbl->counter = 1; |
302 | goto out; | |
303 | } | |
304 | ||
305 | if (atomic_read(&tbl->entries) <= tbl->max_size) { | |
306 | tbl->counter++; | |
307 | goto out; | |
308 | } | |
309 | ||
310 | goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3; | |
311 | if (goal > tbl->max_size/2) | |
312 | goal = tbl->max_size/2; | |
313 | ||
314 | for (i=0, j=tbl->rover; i<IP_VS_LBLC_TAB_SIZE; i++) { | |
315 | j = (j + 1) & IP_VS_LBLC_TAB_MASK; | |
316 | ||
39ac50d0 | 317 | write_lock(&svc->sched_lock); |
1da177e4 LT |
318 | list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) { |
319 | if (time_before(now, en->lastuse + ENTRY_TIMEOUT)) | |
320 | continue; | |
321 | ||
322 | ip_vs_lblc_free(en); | |
323 | atomic_dec(&tbl->entries); | |
324 | goal--; | |
325 | } | |
39ac50d0 | 326 | write_unlock(&svc->sched_lock); |
1da177e4 LT |
327 | if (goal <= 0) |
328 | break; | |
329 | } | |
330 | tbl->rover = j; | |
331 | ||
332 | out: | |
333 | mod_timer(&tbl->periodic_timer, jiffies+CHECK_EXPIRE_INTERVAL); | |
334 | } | |
335 | ||
336 | ||
337 | static int ip_vs_lblc_init_svc(struct ip_vs_service *svc) | |
338 | { | |
339 | int i; | |
340 | struct ip_vs_lblc_table *tbl; | |
341 | ||
342 | /* | |
343 | * Allocate the ip_vs_lblc_table for this service | |
344 | */ | |
748d845c | 345 | tbl = kmalloc(sizeof(*tbl), GFP_KERNEL); |
0a9ee813 | 346 | if (tbl == NULL) |
1da177e4 | 347 | return -ENOMEM; |
0a9ee813 | 348 | |
1da177e4 LT |
349 | svc->sched_data = tbl; |
350 | IP_VS_DBG(6, "LBLC hash table (memory=%Zdbytes) allocated for " | |
39ac50d0 | 351 | "current service\n", sizeof(*tbl)); |
1da177e4 LT |
352 | |
353 | /* | |
354 | * Initialize the hash buckets | |
355 | */ | |
356 | for (i=0; i<IP_VS_LBLC_TAB_SIZE; i++) { | |
357 | INIT_LIST_HEAD(&tbl->bucket[i]); | |
358 | } | |
1da177e4 LT |
359 | tbl->max_size = IP_VS_LBLC_TAB_SIZE*16; |
360 | tbl->rover = 0; | |
361 | tbl->counter = 1; | |
362 | ||
363 | /* | |
364 | * Hook periodic timer for garbage collection | |
365 | */ | |
b24b8a24 | 366 | setup_timer(&tbl->periodic_timer, ip_vs_lblc_check_expire, |
39ac50d0 SW |
367 | (unsigned long)svc); |
368 | mod_timer(&tbl->periodic_timer, jiffies + CHECK_EXPIRE_INTERVAL); | |
1da177e4 LT |
369 | |
370 | return 0; | |
371 | } | |
372 | ||
373 | ||
374 | static int ip_vs_lblc_done_svc(struct ip_vs_service *svc) | |
375 | { | |
376 | struct ip_vs_lblc_table *tbl = svc->sched_data; | |
377 | ||
378 | /* remove periodic timer */ | |
379 | del_timer_sync(&tbl->periodic_timer); | |
380 | ||
381 | /* got to clean up table entries here */ | |
382 | ip_vs_lblc_flush(tbl); | |
383 | ||
384 | /* release the table itself */ | |
39ac50d0 | 385 | kfree(tbl); |
1da177e4 | 386 | IP_VS_DBG(6, "LBLC hash table (memory=%Zdbytes) released\n", |
39ac50d0 | 387 | sizeof(*tbl)); |
1da177e4 LT |
388 | |
389 | return 0; | |
390 | } | |
391 | ||
392 | ||
1da177e4 | 393 | static inline struct ip_vs_dest * |
44548375 | 394 | __ip_vs_lblc_schedule(struct ip_vs_service *svc) |
1da177e4 LT |
395 | { |
396 | struct ip_vs_dest *dest, *least; | |
397 | int loh, doh; | |
398 | ||
399 | /* | |
b552f7e3 | 400 | * We use the following formula to estimate the load: |
1da177e4 LT |
401 | * (dest overhead) / dest->weight |
402 | * | |
403 | * Remember -- no floats in kernel mode!!! | |
404 | * The comparison of h1*w2 > h2*w1 is equivalent to that of | |
405 | * h1/w1 > h2/w2 | |
406 | * if every weight is larger than zero. | |
407 | * | |
408 | * The server with weight=0 is quiesced and will not receive any | |
409 | * new connection. | |
410 | */ | |
411 | list_for_each_entry(dest, &svc->destinations, n_list) { | |
412 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
413 | continue; | |
414 | if (atomic_read(&dest->weight) > 0) { | |
415 | least = dest; | |
b552f7e3 | 416 | loh = ip_vs_dest_conn_overhead(least); |
1da177e4 LT |
417 | goto nextstage; |
418 | } | |
419 | } | |
420 | return NULL; | |
421 | ||
422 | /* | |
423 | * Find the destination with the least load. | |
424 | */ | |
425 | nextstage: | |
426 | list_for_each_entry_continue(dest, &svc->destinations, n_list) { | |
427 | if (dest->flags & IP_VS_DEST_F_OVERLOAD) | |
428 | continue; | |
429 | ||
b552f7e3 | 430 | doh = ip_vs_dest_conn_overhead(dest); |
1da177e4 LT |
431 | if (loh * atomic_read(&dest->weight) > |
432 | doh * atomic_read(&least->weight)) { | |
433 | least = dest; | |
434 | loh = doh; | |
435 | } | |
436 | } | |
437 | ||
44548375 JV |
438 | IP_VS_DBG_BUF(6, "LBLC: server %s:%d " |
439 | "activeconns %d refcnt %d weight %d overhead %d\n", | |
440 | IP_VS_DBG_ADDR(least->af, &least->addr), | |
441 | ntohs(least->port), | |
442 | atomic_read(&least->activeconns), | |
443 | atomic_read(&least->refcnt), | |
444 | atomic_read(&least->weight), loh); | |
1da177e4 LT |
445 | |
446 | return least; | |
447 | } | |
448 | ||
449 | ||
450 | /* | |
451 | * If this destination server is overloaded and there is a less loaded | |
452 | * server, then return true. | |
453 | */ | |
454 | static inline int | |
455 | is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc) | |
456 | { | |
457 | if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) { | |
458 | struct ip_vs_dest *d; | |
459 | ||
460 | list_for_each_entry(d, &svc->destinations, n_list) { | |
461 | if (atomic_read(&d->activeconns)*2 | |
462 | < atomic_read(&d->weight)) { | |
463 | return 1; | |
464 | } | |
465 | } | |
466 | } | |
467 | return 0; | |
468 | } | |
469 | ||
470 | ||
471 | /* | |
472 | * Locality-Based (weighted) Least-Connection scheduling | |
473 | */ | |
474 | static struct ip_vs_dest * | |
475 | ip_vs_lblc_schedule(struct ip_vs_service *svc, const struct sk_buff *skb) | |
476 | { | |
39ac50d0 | 477 | struct ip_vs_lblc_table *tbl = svc->sched_data; |
44548375 | 478 | struct ip_vs_iphdr iph; |
39ac50d0 SW |
479 | struct ip_vs_dest *dest = NULL; |
480 | struct ip_vs_lblc_entry *en; | |
1da177e4 | 481 | |
63dca2c0 | 482 | ip_vs_fill_iph_addr_only(svc->af, skb, &iph); |
44548375 | 483 | |
1e3e238e | 484 | IP_VS_DBG(6, "%s(): Scheduling...\n", __func__); |
1da177e4 | 485 | |
39ac50d0 SW |
486 | /* First look in our cache */ |
487 | read_lock(&svc->sched_lock); | |
44548375 | 488 | en = ip_vs_lblc_get(svc->af, tbl, &iph.daddr); |
39ac50d0 SW |
489 | if (en) { |
490 | /* We only hold a read lock, but this is atomic */ | |
491 | en->lastuse = jiffies; | |
492 | ||
493 | /* | |
494 | * If the destination is not available, i.e. it's in the trash, | |
495 | * we must ignore it, as it may be removed from under our feet, | |
496 | * if someone drops our reference count. Our caller only makes | |
497 | * sure that destinations, that are not in the trash, are not | |
498 | * moved to the trash, while we are scheduling. But anyone can | |
499 | * free up entries from the trash at any time. | |
500 | */ | |
501 | ||
502 | if (en->dest->flags & IP_VS_DEST_F_AVAILABLE) | |
503 | dest = en->dest; | |
504 | } | |
505 | read_unlock(&svc->sched_lock); | |
506 | ||
507 | /* If the destination has a weight and is not overloaded, use it */ | |
508 | if (dest && atomic_read(&dest->weight) > 0 && !is_overloaded(dest, svc)) | |
509 | goto out; | |
510 | ||
511 | /* No cache entry or it is invalid, time to schedule */ | |
44548375 | 512 | dest = __ip_vs_lblc_schedule(svc); |
39ac50d0 | 513 | if (!dest) { |
41ac51ee | 514 | ip_vs_scheduler_err(svc, "no destination available"); |
39ac50d0 | 515 | return NULL; |
1da177e4 | 516 | } |
1da177e4 | 517 | |
39ac50d0 SW |
518 | /* If we fail to create a cache entry, we'll just use the valid dest */ |
519 | write_lock(&svc->sched_lock); | |
44548375 | 520 | ip_vs_lblc_new(tbl, &iph.daddr, dest); |
39ac50d0 SW |
521 | write_unlock(&svc->sched_lock); |
522 | ||
523 | out: | |
44548375 JV |
524 | IP_VS_DBG_BUF(6, "LBLC: destination IP address %s --> server %s:%d\n", |
525 | IP_VS_DBG_ADDR(svc->af, &iph.daddr), | |
526 | IP_VS_DBG_ADDR(svc->af, &dest->addr), ntohs(dest->port)); | |
1da177e4 LT |
527 | |
528 | return dest; | |
529 | } | |
530 | ||
531 | ||
532 | /* | |
533 | * IPVS LBLC Scheduler structure | |
534 | */ | |
535 | static struct ip_vs_scheduler ip_vs_lblc_scheduler = | |
536 | { | |
537 | .name = "lblc", | |
538 | .refcnt = ATOMIC_INIT(0), | |
539 | .module = THIS_MODULE, | |
d149ccc9 | 540 | .n_list = LIST_HEAD_INIT(ip_vs_lblc_scheduler.n_list), |
1da177e4 LT |
541 | .init_service = ip_vs_lblc_init_svc, |
542 | .done_service = ip_vs_lblc_done_svc, | |
1da177e4 LT |
543 | .schedule = ip_vs_lblc_schedule, |
544 | }; | |
545 | ||
61b1ab45 HS |
546 | /* |
547 | * per netns init. | |
548 | */ | |
fb1de432 | 549 | #ifdef CONFIG_SYSCTL |
61b1ab45 HS |
550 | static int __net_init __ip_vs_lblc_init(struct net *net) |
551 | { | |
b6e885dd HS |
552 | struct netns_ipvs *ipvs = net_ipvs(net); |
553 | ||
4b984cd5 HS |
554 | if (!ipvs) |
555 | return -ENOENT; | |
556 | ||
b6e885dd HS |
557 | if (!net_eq(net, &init_net)) { |
558 | ipvs->lblc_ctl_table = kmemdup(vs_vars_table, | |
559 | sizeof(vs_vars_table), | |
560 | GFP_KERNEL); | |
561 | if (ipvs->lblc_ctl_table == NULL) | |
0443929f | 562 | return -ENOMEM; |
464dc801 EB |
563 | |
564 | /* Don't export sysctls to unprivileged users */ | |
565 | if (net->user_ns != &init_user_ns) | |
566 | ipvs->lblc_ctl_table[0].procname = NULL; | |
567 | ||
b6e885dd HS |
568 | } else |
569 | ipvs->lblc_ctl_table = vs_vars_table; | |
b27d777e | 570 | ipvs->sysctl_lblc_expiration = DEFAULT_EXPIRATION; |
b6e885dd HS |
571 | ipvs->lblc_ctl_table[0].data = &ipvs->sysctl_lblc_expiration; |
572 | ||
573 | ipvs->lblc_ctl_header = | |
ec8f23ce | 574 | register_net_sysctl(net, "net/ipv4/vs", ipvs->lblc_ctl_table); |
0443929f SH |
575 | if (!ipvs->lblc_ctl_header) { |
576 | if (!net_eq(net, &init_net)) | |
464dc801 | 577 | kfree(ipvs->lblc_ctl_table);\ |
0443929f SH |
578 | return -ENOMEM; |
579 | } | |
61b1ab45 HS |
580 | |
581 | return 0; | |
582 | } | |
583 | ||
584 | static void __net_exit __ip_vs_lblc_exit(struct net *net) | |
585 | { | |
b6e885dd HS |
586 | struct netns_ipvs *ipvs = net_ipvs(net); |
587 | ||
588 | unregister_net_sysctl_table(ipvs->lblc_ctl_header); | |
61b1ab45 | 589 | |
b6e885dd HS |
590 | if (!net_eq(net, &init_net)) |
591 | kfree(ipvs->lblc_ctl_table); | |
61b1ab45 HS |
592 | } |
593 | ||
fb1de432 SH |
594 | #else |
595 | ||
596 | static int __net_init __ip_vs_lblc_init(struct net *net) { return 0; } | |
597 | static void __net_exit __ip_vs_lblc_exit(struct net *net) { } | |
598 | ||
599 | #endif | |
600 | ||
61b1ab45 HS |
601 | static struct pernet_operations ip_vs_lblc_ops = { |
602 | .init = __ip_vs_lblc_init, | |
603 | .exit = __ip_vs_lblc_exit, | |
604 | }; | |
1da177e4 LT |
605 | |
606 | static int __init ip_vs_lblc_init(void) | |
607 | { | |
a014bc8f PE |
608 | int ret; |
609 | ||
61b1ab45 HS |
610 | ret = register_pernet_subsys(&ip_vs_lblc_ops); |
611 | if (ret) | |
612 | return ret; | |
613 | ||
a014bc8f PE |
614 | ret = register_ip_vs_scheduler(&ip_vs_lblc_scheduler); |
615 | if (ret) | |
61b1ab45 | 616 | unregister_pernet_subsys(&ip_vs_lblc_ops); |
a014bc8f | 617 | return ret; |
1da177e4 LT |
618 | } |
619 | ||
1da177e4 LT |
620 | static void __exit ip_vs_lblc_cleanup(void) |
621 | { | |
1da177e4 | 622 | unregister_ip_vs_scheduler(&ip_vs_lblc_scheduler); |
61b1ab45 | 623 | unregister_pernet_subsys(&ip_vs_lblc_ops); |
1da177e4 LT |
624 | } |
625 | ||
626 | ||
627 | module_init(ip_vs_lblc_init); | |
628 | module_exit(ip_vs_lblc_cleanup); | |
629 | MODULE_LICENSE("GPL"); |