ipv4: fix checkpatch errors
[deliverable/linux.git] / net / netfilter / ipvs / ip_vs_lblcr.c
CommitLineData
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
90struct ip_vs_dest_set_elem {
91 struct list_head list; /* list link */
1da177e4
LT
92 struct ip_vs_dest *dest; /* destination server */
93};
94
95struct 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 103static struct ip_vs_dest_set_elem *
1da177e4
LT
104ip_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);
0a9ee813 115 if (e == NULL)
1da177e4 116 return NULL;
1da177e4
LT
117
118 atomic_inc(&dest->refcnt);
119 e->dest = dest;
120
51f0bc78 121 list_add(&e->list, &set->list);
1da177e4 122 atomic_inc(&set->size);
1da177e4
LT
123
124 set->lastmod = jiffies;
125 return e;
126}
127
128static void
129ip_vs_dest_set_erase(struct ip_vs_dest_set *set, struct ip_vs_dest *dest)
130{
51f0bc78 131 struct ip_vs_dest_set_elem *e;
1da177e4 132
51f0bc78 133 list_for_each_entry(e, &set->list, list) {
1da177e4
LT
134 if (e->dest == dest) {
135 /* HIT */
1da177e4
LT
136 atomic_dec(&set->size);
137 set->lastmod = jiffies;
138 atomic_dec(&e->dest->refcnt);
51f0bc78 139 list_del(&e->list);
1da177e4
LT
140 kfree(e);
141 break;
142 }
1da177e4 143 }
1da177e4
LT
144}
145
146static void ip_vs_dest_set_eraseall(struct ip_vs_dest_set *set)
147{
51f0bc78 148 struct ip_vs_dest_set_elem *e, *ep;
1da177e4
LT
149
150 write_lock(&set->lock);
51f0bc78 151 list_for_each_entry_safe(e, ep, &set->list, list) {
1da177e4 152 /*
25985edc 153 * We don't kfree dest because it is referred either
1da177e4
LT
154 * by its service or by the trash dest list.
155 */
156 atomic_dec(&e->dest->refcnt);
51f0bc78 157 list_del(&e->list);
1da177e4
LT
158 kfree(e);
159 }
160 write_unlock(&set->lock);
161}
162
163/* get weighted least-connection node in the destination set */
164static inline struct ip_vs_dest *ip_vs_dest_set_min(struct ip_vs_dest_set *set)
165{
51f0bc78 166 register struct ip_vs_dest_set_elem *e;
1da177e4
LT
167 struct ip_vs_dest *dest, *least;
168 int loh, doh;
169
170 if (set == NULL)
171 return NULL;
172
1da177e4 173 /* select the first destination server, whose weight > 0 */
51f0bc78 174 list_for_each_entry(e, &set->list, list) {
1da177e4
LT
175 least = e->dest;
176 if (least->flags & IP_VS_DEST_F_OVERLOAD)
177 continue;
178
179 if ((atomic_read(&least->weight) > 0)
180 && (least->flags & IP_VS_DEST_F_AVAILABLE)) {
b552f7e3 181 loh = ip_vs_dest_conn_overhead(least);
1da177e4
LT
182 goto nextstage;
183 }
184 }
1da177e4
LT
185 return NULL;
186
187 /* find the destination with the weighted least load */
188 nextstage:
51f0bc78 189 list_for_each_entry(e, &set->list, list) {
1da177e4
LT
190 dest = e->dest;
191 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
192 continue;
193
b552f7e3 194 doh = ip_vs_dest_conn_overhead(dest);
1da177e4
LT
195 if ((loh * atomic_read(&dest->weight) >
196 doh * atomic_read(&least->weight))
197 && (dest->flags & IP_VS_DEST_F_AVAILABLE)) {
198 least = dest;
199 loh = doh;
200 }
201 }
1da177e4 202
1e3e238e 203 IP_VS_DBG_BUF(6, "%s(): server %s:%d "
44548375 204 "activeconns %d refcnt %d weight %d overhead %d\n",
1e3e238e 205 __func__,
44548375
JV
206 IP_VS_DBG_ADDR(least->af, &least->addr),
207 ntohs(least->port),
208 atomic_read(&least->activeconns),
209 atomic_read(&least->refcnt),
210 atomic_read(&least->weight), loh);
1da177e4
LT
211 return least;
212}
213
214
215/* get weighted most-connection node in the destination set */
216static inline struct ip_vs_dest *ip_vs_dest_set_max(struct ip_vs_dest_set *set)
217{
51f0bc78 218 register struct ip_vs_dest_set_elem *e;
1da177e4
LT
219 struct ip_vs_dest *dest, *most;
220 int moh, doh;
221
222 if (set == NULL)
223 return NULL;
224
1da177e4 225 /* select the first destination server, whose weight > 0 */
51f0bc78 226 list_for_each_entry(e, &set->list, list) {
1da177e4
LT
227 most = e->dest;
228 if (atomic_read(&most->weight) > 0) {
b552f7e3 229 moh = ip_vs_dest_conn_overhead(most);
1da177e4
LT
230 goto nextstage;
231 }
232 }
1da177e4
LT
233 return NULL;
234
235 /* find the destination with the weighted most load */
236 nextstage:
51f0bc78 237 list_for_each_entry(e, &set->list, list) {
1da177e4 238 dest = e->dest;
b552f7e3 239 doh = ip_vs_dest_conn_overhead(dest);
1da177e4
LT
240 /* moh/mw < doh/dw ==> moh*dw < doh*mw, where mw,dw>0 */
241 if ((moh * atomic_read(&dest->weight) <
242 doh * atomic_read(&most->weight))
243 && (atomic_read(&dest->weight) > 0)) {
244 most = dest;
245 moh = doh;
246 }
247 }
1da177e4 248
1e3e238e 249 IP_VS_DBG_BUF(6, "%s(): server %s:%d "
44548375 250 "activeconns %d refcnt %d weight %d overhead %d\n",
1e3e238e 251 __func__,
44548375
JV
252 IP_VS_DBG_ADDR(most->af, &most->addr), ntohs(most->port),
253 atomic_read(&most->activeconns),
254 atomic_read(&most->refcnt),
255 atomic_read(&most->weight), moh);
1da177e4
LT
256 return most;
257}
258
259
260/*
261 * IPVS lblcr entry represents an association between destination
262 * IP address and its destination server set
263 */
264struct ip_vs_lblcr_entry {
265 struct list_head list;
44548375
JV
266 int af; /* address family */
267 union nf_inet_addr addr; /* destination IP address */
1da177e4
LT
268 struct ip_vs_dest_set set; /* destination server set */
269 unsigned long lastuse; /* last used time */
270};
271
272
273/*
274 * IPVS lblcr hash table
275 */
276struct ip_vs_lblcr_table {
1da177e4
LT
277 struct list_head bucket[IP_VS_LBLCR_TAB_SIZE]; /* hash bucket */
278 atomic_t entries; /* number of entries */
279 int max_size; /* maximum size of entries */
280 struct timer_list periodic_timer; /* collect stale entries */
281 int rover; /* rover for expire check */
282 int counter; /* counter for no expire */
283};
284
285
fb1de432 286#ifdef CONFIG_SYSCTL
1da177e4
LT
287/*
288 * IPVS LBLCR sysctl table
289 */
290
291static ctl_table vs_vars_table[] = {
292 {
1da177e4 293 .procname = "lblcr_expiration",
d0a1eef9 294 .data = NULL,
1da177e4 295 .maxlen = sizeof(int),
e905a9ed 296 .mode = 0644,
6d9f239a 297 .proc_handler = proc_dointvec_jiffies,
1da177e4 298 },
f8572d8f 299 { }
1da177e4 300};
fb1de432 301#endif
1da177e4 302
1da177e4
LT
303static inline void ip_vs_lblcr_free(struct ip_vs_lblcr_entry *en)
304{
305 list_del(&en->list);
306 ip_vs_dest_set_eraseall(&en->set);
307 kfree(en);
308}
309
310
311/*
312 * Returns hash value for IPVS LBLCR entry
313 */
44548375
JV
314static inline unsigned
315ip_vs_lblcr_hashkey(int af, const union nf_inet_addr *addr)
1da177e4 316{
44548375
JV
317 __be32 addr_fold = addr->ip;
318
319#ifdef CONFIG_IP_VS_IPV6
320 if (af == AF_INET6)
321 addr_fold = addr->ip6[0]^addr->ip6[1]^
322 addr->ip6[2]^addr->ip6[3];
323#endif
324 return (ntohl(addr_fold)*2654435761UL) & IP_VS_LBLCR_TAB_MASK;
1da177e4
LT
325}
326
327
328/*
329 * Hash an entry in the ip_vs_lblcr_table.
330 * returns bool success.
331 */
f728bafb 332static void
1da177e4
LT
333ip_vs_lblcr_hash(struct ip_vs_lblcr_table *tbl, struct ip_vs_lblcr_entry *en)
334{
44548375 335 unsigned hash = ip_vs_lblcr_hashkey(en->af, &en->addr);
1da177e4 336
1da177e4
LT
337 list_add(&en->list, &tbl->bucket[hash]);
338 atomic_inc(&tbl->entries);
1da177e4
LT
339}
340
341
1da177e4 342/*
f728bafb
SW
343 * Get ip_vs_lblcr_entry associated with supplied parameters. Called under
344 * read lock.
1da177e4
LT
345 */
346static inline struct ip_vs_lblcr_entry *
44548375
JV
347ip_vs_lblcr_get(int af, struct ip_vs_lblcr_table *tbl,
348 const union nf_inet_addr *addr)
1da177e4 349{
44548375 350 unsigned hash = ip_vs_lblcr_hashkey(af, addr);
1da177e4
LT
351 struct ip_vs_lblcr_entry *en;
352
f728bafb 353 list_for_each_entry(en, &tbl->bucket[hash], list)
44548375 354 if (ip_vs_addr_equal(af, &en->addr, addr))
f728bafb
SW
355 return en;
356
357 return NULL;
358}
1da177e4 359
1da177e4 360
f728bafb
SW
361/*
362 * Create or update an ip_vs_lblcr_entry, which is a mapping of a destination
363 * IP address to a server. Called under write lock.
364 */
365static inline struct ip_vs_lblcr_entry *
44548375 366ip_vs_lblcr_new(struct ip_vs_lblcr_table *tbl, const union nf_inet_addr *daddr,
f728bafb
SW
367 struct ip_vs_dest *dest)
368{
369 struct ip_vs_lblcr_entry *en;
370
44548375 371 en = ip_vs_lblcr_get(dest->af, tbl, daddr);
f728bafb
SW
372 if (!en) {
373 en = kmalloc(sizeof(*en), GFP_ATOMIC);
0a9ee813 374 if (!en)
f728bafb 375 return NULL;
f728bafb 376
44548375
JV
377 en->af = dest->af;
378 ip_vs_addr_copy(dest->af, &en->addr, daddr);
f728bafb
SW
379 en->lastuse = jiffies;
380
421f91d2 381 /* initialize its dest set */
f728bafb 382 atomic_set(&(en->set.size), 0);
51f0bc78 383 INIT_LIST_HEAD(&en->set.list);
f728bafb
SW
384 rwlock_init(&en->set.lock);
385
386 ip_vs_lblcr_hash(tbl, en);
1da177e4
LT
387 }
388
f728bafb
SW
389 write_lock(&en->set.lock);
390 ip_vs_dest_set_insert(&en->set, dest);
391 write_unlock(&en->set.lock);
1da177e4 392
f728bafb 393 return en;
1da177e4
LT
394}
395
396
397/*
398 * Flush all the entries of the specified table.
399 */
400static void ip_vs_lblcr_flush(struct ip_vs_lblcr_table *tbl)
401{
402 int i;
403 struct ip_vs_lblcr_entry *en, *nxt;
404
f728bafb 405 /* No locking required, only called during cleanup. */
1da177e4 406 for (i=0; i<IP_VS_LBLCR_TAB_SIZE; i++) {
1da177e4
LT
407 list_for_each_entry_safe(en, nxt, &tbl->bucket[i], list) {
408 ip_vs_lblcr_free(en);
1da177e4 409 }
1da177e4
LT
410 }
411}
412
b27d777e
SH
413static int sysctl_lblcr_expiration(struct ip_vs_service *svc)
414{
415#ifdef CONFIG_SYSCTL
416 struct netns_ipvs *ipvs = net_ipvs(svc->net);
417 return ipvs->sysctl_lblcr_expiration;
418#else
419 return DEFAULT_EXPIRATION;
420#endif
421}
1da177e4 422
f728bafb 423static inline void ip_vs_lblcr_full_check(struct ip_vs_service *svc)
1da177e4 424{
f728bafb 425 struct ip_vs_lblcr_table *tbl = svc->sched_data;
1da177e4
LT
426 unsigned long now = jiffies;
427 int i, j;
428 struct ip_vs_lblcr_entry *en, *nxt;
429
430 for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) {
431 j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
432
f728bafb 433 write_lock(&svc->sched_lock);
1da177e4 434 list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) {
b27d777e
SH
435 if (time_after(en->lastuse +
436 sysctl_lblcr_expiration(svc), now))
1da177e4
LT
437 continue;
438
439 ip_vs_lblcr_free(en);
440 atomic_dec(&tbl->entries);
441 }
f728bafb 442 write_unlock(&svc->sched_lock);
1da177e4
LT
443 }
444 tbl->rover = j;
445}
446
447
448/*
449 * Periodical timer handler for IPVS lblcr table
450 * It is used to collect stale entries when the number of entries
451 * exceeds the maximum size of the table.
452 *
453 * Fixme: we probably need more complicated algorithm to collect
454 * entries that have not been used for a long time even
455 * if the number of entries doesn't exceed the maximum size
456 * of the table.
457 * The full expiration check is for this purpose now.
458 */
459static void ip_vs_lblcr_check_expire(unsigned long data)
460{
f728bafb
SW
461 struct ip_vs_service *svc = (struct ip_vs_service *) data;
462 struct ip_vs_lblcr_table *tbl = svc->sched_data;
1da177e4
LT
463 unsigned long now = jiffies;
464 int goal;
465 int i, j;
466 struct ip_vs_lblcr_entry *en, *nxt;
467
1da177e4
LT
468 if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) {
469 /* do full expiration check */
f728bafb 470 ip_vs_lblcr_full_check(svc);
1da177e4
LT
471 tbl->counter = 1;
472 goto out;
473 }
474
475 if (atomic_read(&tbl->entries) <= tbl->max_size) {
476 tbl->counter++;
477 goto out;
478 }
479
480 goal = (atomic_read(&tbl->entries) - tbl->max_size)*4/3;
481 if (goal > tbl->max_size/2)
482 goal = tbl->max_size/2;
483
484 for (i=0, j=tbl->rover; i<IP_VS_LBLCR_TAB_SIZE; i++) {
485 j = (j + 1) & IP_VS_LBLCR_TAB_MASK;
486
f728bafb 487 write_lock(&svc->sched_lock);
1da177e4
LT
488 list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) {
489 if (time_before(now, en->lastuse+ENTRY_TIMEOUT))
490 continue;
491
492 ip_vs_lblcr_free(en);
493 atomic_dec(&tbl->entries);
494 goal--;
495 }
f728bafb 496 write_unlock(&svc->sched_lock);
1da177e4
LT
497 if (goal <= 0)
498 break;
499 }
500 tbl->rover = j;
501
502 out:
503 mod_timer(&tbl->periodic_timer, jiffies+CHECK_EXPIRE_INTERVAL);
504}
505
1da177e4
LT
506static int ip_vs_lblcr_init_svc(struct ip_vs_service *svc)
507{
508 int i;
509 struct ip_vs_lblcr_table *tbl;
510
511 /*
512 * Allocate the ip_vs_lblcr_table for this service
513 */
f728bafb 514 tbl = kmalloc(sizeof(*tbl), GFP_ATOMIC);
0a9ee813 515 if (tbl == NULL)
1da177e4 516 return -ENOMEM;
0a9ee813 517
1da177e4
LT
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
543static 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 562static 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 /*
b552f7e3 569 * We use the following formula to estimate the load:
1da177e4
LT
570 * (dest overhead) / dest->weight
571 *
572 * Remember -- no floats in kernel mode!!!
573 * The comparison of h1*w2 > h2*w1 is equivalent to that of
574 * h1/w1 > h2/w2
575 * if every weight is larger than zero.
576 *
577 * The server with weight=0 is quiesced and will not receive any
578 * new connection.
579 */
580 list_for_each_entry(dest, &svc->destinations, n_list) {
581 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
582 continue;
583
584 if (atomic_read(&dest->weight) > 0) {
585 least = dest;
b552f7e3 586 loh = ip_vs_dest_conn_overhead(least);
1da177e4
LT
587 goto nextstage;
588 }
589 }
590 return NULL;
591
592 /*
593 * Find the destination with the least load.
594 */
595 nextstage:
596 list_for_each_entry_continue(dest, &svc->destinations, n_list) {
597 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
598 continue;
599
b552f7e3 600 doh = ip_vs_dest_conn_overhead(dest);
1da177e4
LT
601 if (loh * atomic_read(&dest->weight) >
602 doh * atomic_read(&least->weight)) {
603 least = dest;
604 loh = doh;
605 }
606 }
607
44548375
JV
608 IP_VS_DBG_BUF(6, "LBLCR: server %s:%d "
609 "activeconns %d refcnt %d weight %d overhead %d\n",
610 IP_VS_DBG_ADDR(least->af, &least->addr),
611 ntohs(least->port),
612 atomic_read(&least->activeconns),
613 atomic_read(&least->refcnt),
614 atomic_read(&least->weight), loh);
1da177e4
LT
615
616 return least;
617}
618
619
620/*
621 * If this destination server is overloaded and there is a less loaded
622 * server, then return true.
623 */
624static inline int
625is_overloaded(struct ip_vs_dest *dest, struct ip_vs_service *svc)
626{
627 if (atomic_read(&dest->activeconns) > atomic_read(&dest->weight)) {
628 struct ip_vs_dest *d;
629
630 list_for_each_entry(d, &svc->destinations, n_list) {
631 if (atomic_read(&d->activeconns)*2
632 < atomic_read(&d->weight)) {
633 return 1;
634 }
635 }
636 }
637 return 0;
638}
639
640
641/*
642 * Locality-Based (weighted) Least-Connection scheduling
643 */
644static struct ip_vs_dest *
645ip_vs_lblcr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
646{
f728bafb 647 struct ip_vs_lblcr_table *tbl = svc->sched_data;
44548375 648 struct ip_vs_iphdr iph;
f728bafb
SW
649 struct ip_vs_dest *dest = NULL;
650 struct ip_vs_lblcr_entry *en;
1da177e4 651
44548375
JV
652 ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
653
1e3e238e 654 IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
1da177e4 655
f728bafb
SW
656 /* First look in our cache */
657 read_lock(&svc->sched_lock);
44548375 658 en = ip_vs_lblcr_get(svc->af, tbl, &iph.daddr);
f728bafb
SW
659 if (en) {
660 /* We only hold a read lock, but this is atomic */
661 en->lastuse = jiffies;
662
663 /* Get the least loaded destination */
664 read_lock(&en->set.lock);
1da177e4 665 dest = ip_vs_dest_set_min(&en->set);
f728bafb
SW
666 read_unlock(&en->set.lock);
667
668 /* More than one destination + enough time passed by, cleanup */
1da177e4 669 if (atomic_read(&en->set.size) > 1 &&
f728bafb 670 time_after(jiffies, en->set.lastmod +
b27d777e 671 sysctl_lblcr_expiration(svc))) {
1da177e4 672 struct ip_vs_dest *m;
f728bafb
SW
673
674 write_lock(&en->set.lock);
1da177e4
LT
675 m = ip_vs_dest_set_max(&en->set);
676 if (m)
677 ip_vs_dest_set_erase(&en->set, m);
f728bafb
SW
678 write_unlock(&en->set.lock);
679 }
680
681 /* If the destination is not overloaded, use it */
682 if (dest && !is_overloaded(dest, svc)) {
683 read_unlock(&svc->sched_lock);
684 goto out;
685 }
686
687 /* The cache entry is invalid, time to schedule */
44548375 688 dest = __ip_vs_lblcr_schedule(svc);
f728bafb 689 if (!dest) {
41ac51ee 690 ip_vs_scheduler_err(svc, "no destination available");
f728bafb
SW
691 read_unlock(&svc->sched_lock);
692 return NULL;
1da177e4 693 }
f728bafb
SW
694
695 /* Update our cache entry */
696 write_lock(&en->set.lock);
697 ip_vs_dest_set_insert(&en->set, dest);
698 write_unlock(&en->set.lock);
699 }
700 read_unlock(&svc->sched_lock);
701
702 if (dest)
703 goto out;
704
705 /* No cache entry, time to schedule */
44548375 706 dest = __ip_vs_lblcr_schedule(svc);
f728bafb
SW
707 if (!dest) {
708 IP_VS_DBG(1, "no destination available\n");
709 return NULL;
1da177e4 710 }
1da177e4 711
f728bafb
SW
712 /* If we fail to create a cache entry, we'll just use the valid dest */
713 write_lock(&svc->sched_lock);
44548375 714 ip_vs_lblcr_new(tbl, &iph.daddr, dest);
f728bafb
SW
715 write_unlock(&svc->sched_lock);
716
717out:
44548375
JV
718 IP_VS_DBG_BUF(6, "LBLCR: destination IP address %s --> server %s:%d\n",
719 IP_VS_DBG_ADDR(svc->af, &iph.daddr),
720 IP_VS_DBG_ADDR(svc->af, &dest->addr), ntohs(dest->port));
1da177e4
LT
721
722 return dest;
723}
724
725
726/*
727 * IPVS LBLCR Scheduler structure
728 */
729static struct ip_vs_scheduler ip_vs_lblcr_scheduler =
730{
731 .name = "lblcr",
732 .refcnt = ATOMIC_INIT(0),
733 .module = THIS_MODULE,
d149ccc9 734 .n_list = LIST_HEAD_INIT(ip_vs_lblcr_scheduler.n_list),
1da177e4
LT
735 .init_service = ip_vs_lblcr_init_svc,
736 .done_service = ip_vs_lblcr_done_svc,
1da177e4
LT
737 .schedule = ip_vs_lblcr_schedule,
738};
739
61b1ab45
HS
740/*
741 * per netns init.
742 */
fb1de432 743#ifdef CONFIG_SYSCTL
61b1ab45
HS
744static int __net_init __ip_vs_lblcr_init(struct net *net)
745{
d0a1eef9
HS
746 struct netns_ipvs *ipvs = net_ipvs(net);
747
748 if (!net_eq(net, &init_net)) {
749 ipvs->lblcr_ctl_table = kmemdup(vs_vars_table,
750 sizeof(vs_vars_table),
751 GFP_KERNEL);
752 if (ipvs->lblcr_ctl_table == NULL)
0443929f 753 return -ENOMEM;
d0a1eef9
HS
754 } else
755 ipvs->lblcr_ctl_table = vs_vars_table;
b27d777e 756 ipvs->sysctl_lblcr_expiration = DEFAULT_EXPIRATION;
d0a1eef9
HS
757 ipvs->lblcr_ctl_table[0].data = &ipvs->sysctl_lblcr_expiration;
758
759 ipvs->lblcr_ctl_header =
760 register_net_sysctl_table(net, net_vs_ctl_path,
761 ipvs->lblcr_ctl_table);
0443929f
SH
762 if (!ipvs->lblcr_ctl_header) {
763 if (!net_eq(net, &init_net))
764 kfree(ipvs->lblcr_ctl_table);
765 return -ENOMEM;
766 }
61b1ab45
HS
767
768 return 0;
769}
770
771static void __net_exit __ip_vs_lblcr_exit(struct net *net)
772{
d0a1eef9
HS
773 struct netns_ipvs *ipvs = net_ipvs(net);
774
775 unregister_net_sysctl_table(ipvs->lblcr_ctl_header);
61b1ab45 776
d0a1eef9
HS
777 if (!net_eq(net, &init_net))
778 kfree(ipvs->lblcr_ctl_table);
61b1ab45
HS
779}
780
fb1de432
SH
781#else
782
783static int __net_init __ip_vs_lblcr_init(struct net *net) { return 0; }
784static void __net_exit __ip_vs_lblcr_exit(struct net *net) { }
785
786#endif
787
61b1ab45
HS
788static struct pernet_operations ip_vs_lblcr_ops = {
789 .init = __ip_vs_lblcr_init,
790 .exit = __ip_vs_lblcr_exit,
791};
1da177e4
LT
792
793static int __init ip_vs_lblcr_init(void)
794{
a014bc8f
PE
795 int ret;
796
61b1ab45
HS
797 ret = register_pernet_subsys(&ip_vs_lblcr_ops);
798 if (ret)
799 return ret;
800
a014bc8f
PE
801 ret = register_ip_vs_scheduler(&ip_vs_lblcr_scheduler);
802 if (ret)
61b1ab45 803 unregister_pernet_subsys(&ip_vs_lblcr_ops);
a014bc8f 804 return ret;
1da177e4
LT
805}
806
1da177e4
LT
807static void __exit ip_vs_lblcr_cleanup(void)
808{
1da177e4 809 unregister_ip_vs_scheduler(&ip_vs_lblcr_scheduler);
61b1ab45 810 unregister_pernet_subsys(&ip_vs_lblcr_ops);
1da177e4
LT
811}
812
813
814module_init(ip_vs_lblcr_init);
815module_exit(ip_vs_lblcr_cleanup);
816MODULE_LICENSE("GPL");
This page took 0.620566 seconds and 5 git commands to generate.