Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * IPVS: Locality-Based Least-Connection scheduling module |
| 3 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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) |
Uwe Kleine-König | 421f91d | 2010-06-11 12:17:00 +0200 | [diff] [blame] | 14 | * Wensong Zhang : fixed the uninitialized tbl->lock bug |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | |
Hannes Eder | 9aada7a | 2009-07-30 14:29:44 -0700 | [diff] [blame] | 42 | #define KMSG_COMPONENT "IPVS" |
| 43 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt |
| 44 | |
Arnaldo Carvalho de Melo | 14c8502 | 2005-12-27 02:43:12 -0200 | [diff] [blame] | 45 | #include <linux/ip.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 46 | #include <linux/slab.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 47 | #include <linux/module.h> |
| 48 | #include <linux/kernel.h> |
Arnaldo Carvalho de Melo | 14c8502 | 2005-12-27 02:43:12 -0200 | [diff] [blame] | 49 | #include <linux/skbuff.h> |
Al Viro | d7fe0f2 | 2006-12-03 23:15:30 -0500 | [diff] [blame] | 50 | #include <linux/jiffies.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | |
Simon Horman | b27d777 | 2011-02-04 18:33:01 +0900 | [diff] [blame] | 66 | #define DEFAULT_EXPIRATION (24*60*60*HZ) |
| 67 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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; |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 94 | int af; /* address family */ |
| 95 | union nf_inet_addr addr; /* destination IP address */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | */ |
Simon Horman | fb1de43 | 2011-02-04 18:33:02 +0900 | [diff] [blame] | 117 | #ifdef CONFIG_SYSCTL |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 118 | static ctl_table vs_vars_table[] = { |
| 119 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 120 | .procname = "lblc_expiration", |
Hans Schillstrom | b6e885d | 2011-01-03 14:44:45 +0100 | [diff] [blame] | 121 | .data = NULL, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 122 | .maxlen = sizeof(int), |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 123 | .mode = 0644, |
Alexey Dobriyan | 6d9f239 | 2008-11-03 18:21:05 -0800 | [diff] [blame] | 124 | .proc_handler = proc_dointvec_jiffies, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 125 | }, |
Eric W. Biederman | f8572d8 | 2009-11-05 13:32:03 -0800 | [diff] [blame] | 126 | { } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | }; |
Simon Horman | fb1de43 | 2011-02-04 18:33:02 +0900 | [diff] [blame] | 128 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 129 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 130 | static inline void ip_vs_lblc_free(struct ip_vs_lblc_entry *en) |
| 131 | { |
| 132 | list_del(&en->list); |
| 133 | /* |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 134 | * We don't kfree dest because it is referred either by its service |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | */ |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 145 | static inline unsigned |
| 146 | ip_vs_lblc_hashkey(int af, const union nf_inet_addr *addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 147 | { |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 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; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 156 | } |
| 157 | |
| 158 | |
| 159 | /* |
| 160 | * Hash an entry in the ip_vs_lblc_table. |
| 161 | * returns bool success. |
| 162 | */ |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 163 | static void |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 164 | ip_vs_lblc_hash(struct ip_vs_lblc_table *tbl, struct ip_vs_lblc_entry *en) |
| 165 | { |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 166 | unsigned hash = ip_vs_lblc_hashkey(en->af, &en->addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 168 | list_add(&en->list, &tbl->bucket[hash]); |
| 169 | atomic_inc(&tbl->entries); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 170 | } |
| 171 | |
| 172 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | /* |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 174 | * Get ip_vs_lblc_entry associated with supplied parameters. Called under read |
| 175 | * lock |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 176 | */ |
| 177 | static inline struct ip_vs_lblc_entry * |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 178 | ip_vs_lblc_get(int af, struct ip_vs_lblc_table *tbl, |
| 179 | const union nf_inet_addr *addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | { |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 181 | unsigned hash = ip_vs_lblc_hashkey(af, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | struct ip_vs_lblc_entry *en; |
| 183 | |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 184 | list_for_each_entry(en, &tbl->bucket[hash], list) |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 185 | if (ip_vs_addr_equal(af, &en->addr, addr)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | return en; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 187 | |
| 188 | return NULL; |
| 189 | } |
| 190 | |
| 191 | |
| 192 | /* |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 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 * |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 197 | ip_vs_lblc_new(struct ip_vs_lblc_table *tbl, const union nf_inet_addr *daddr, |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 198 | struct ip_vs_dest *dest) |
| 199 | { |
| 200 | struct ip_vs_lblc_entry *en; |
| 201 | |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 202 | en = ip_vs_lblc_get(dest->af, tbl, daddr); |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 203 | if (!en) { |
| 204 | en = kmalloc(sizeof(*en), GFP_ATOMIC); |
Joe Perches | 0a9ee81 | 2011-08-29 14:17:25 -0700 | [diff] [blame] | 205 | if (!en) |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 206 | return NULL; |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 207 | |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 208 | en->af = dest->af; |
| 209 | ip_vs_addr_copy(dest->af, &en->addr, daddr); |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 210 | en->lastuse = jiffies; |
| 211 | |
| 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; |
| 223 | } |
| 224 | |
| 225 | |
| 226 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | * Flush all the entries of the specified table. |
| 228 | */ |
| 229 | static void ip_vs_lblc_flush(struct ip_vs_lblc_table *tbl) |
| 230 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 231 | struct ip_vs_lblc_entry *en, *nxt; |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 232 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 233 | |
| 234 | for (i=0; i<IP_VS_LBLC_TAB_SIZE; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 235 | list_for_each_entry_safe(en, nxt, &tbl->bucket[i], list) { |
| 236 | ip_vs_lblc_free(en); |
| 237 | atomic_dec(&tbl->entries); |
| 238 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | } |
| 240 | } |
| 241 | |
Simon Horman | b27d777 | 2011-02-04 18:33:01 +0900 | [diff] [blame] | 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 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 251 | |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 252 | static inline void ip_vs_lblc_full_check(struct ip_vs_service *svc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 253 | { |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 254 | struct ip_vs_lblc_table *tbl = svc->sched_data; |
| 255 | struct ip_vs_lblc_entry *en, *nxt; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 256 | unsigned long now = jiffies; |
| 257 | int i, j; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 262 | write_lock(&svc->sched_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 263 | list_for_each_entry_safe(en, nxt, &tbl->bucket[j], list) { |
YOSHIFUJI Hideaki | e905a9e | 2007-02-09 23:24:47 +0900 | [diff] [blame] | 264 | if (time_before(now, |
Hans Schillstrom | b6e885d | 2011-01-03 14:44:45 +0100 | [diff] [blame] | 265 | en->lastuse + |
Simon Horman | b27d777 | 2011-02-04 18:33:01 +0900 | [diff] [blame] | 266 | sysctl_lblc_expiration(svc))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 267 | continue; |
| 268 | |
| 269 | ip_vs_lblc_free(en); |
| 270 | atomic_dec(&tbl->entries); |
| 271 | } |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 272 | write_unlock(&svc->sched_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | { |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 291 | struct ip_vs_service *svc = (struct ip_vs_service *) data; |
| 292 | struct ip_vs_lblc_table *tbl = svc->sched_data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 293 | unsigned long now = jiffies; |
| 294 | int goal; |
| 295 | int i, j; |
| 296 | struct ip_vs_lblc_entry *en, *nxt; |
| 297 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 298 | if ((tbl->counter % COUNT_FOR_FULL_EXPIRATION) == 0) { |
| 299 | /* do full expiration check */ |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 300 | ip_vs_lblc_full_check(svc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 317 | write_lock(&svc->sched_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | } |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 326 | write_unlock(&svc->sched_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | */ |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 345 | tbl = kmalloc(sizeof(*tbl), GFP_ATOMIC); |
Joe Perches | 0a9ee81 | 2011-08-29 14:17:25 -0700 | [diff] [blame] | 346 | if (tbl == NULL) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 347 | return -ENOMEM; |
Joe Perches | 0a9ee81 | 2011-08-29 14:17:25 -0700 | [diff] [blame] | 348 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 349 | svc->sched_data = tbl; |
| 350 | IP_VS_DBG(6, "LBLC hash table (memory=%Zdbytes) allocated for " |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 351 | "current service\n", sizeof(*tbl)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | */ |
Pavel Emelyanov | b24b8a2 | 2008-01-23 21:20:07 -0800 | [diff] [blame] | 366 | setup_timer(&tbl->periodic_timer, ip_vs_lblc_check_expire, |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 367 | (unsigned long)svc); |
| 368 | mod_timer(&tbl->periodic_timer, jiffies + CHECK_EXPIRE_INTERVAL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 */ |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 385 | kfree(tbl); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 386 | IP_VS_DBG(6, "LBLC hash table (memory=%Zdbytes) released\n", |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 387 | sizeof(*tbl)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 388 | |
| 389 | return 0; |
| 390 | } |
| 391 | |
| 392 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 393 | static inline struct ip_vs_dest * |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 394 | __ip_vs_lblc_schedule(struct ip_vs_service *svc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 395 | { |
| 396 | struct ip_vs_dest *dest, *least; |
| 397 | int loh, doh; |
| 398 | |
| 399 | /* |
Changli Gao | b552f7e | 2011-02-19 17:32:28 +0800 | [diff] [blame] | 400 | * We use the following formula to estimate the load: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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; |
Changli Gao | b552f7e | 2011-02-19 17:32:28 +0800 | [diff] [blame] | 416 | loh = ip_vs_dest_conn_overhead(least); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | |
Changli Gao | b552f7e | 2011-02-19 17:32:28 +0800 | [diff] [blame] | 430 | doh = ip_vs_dest_conn_overhead(dest); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 431 | if (loh * atomic_read(&dest->weight) > |
| 432 | doh * atomic_read(&least->weight)) { |
| 433 | least = dest; |
| 434 | loh = doh; |
| 435 | } |
| 436 | } |
| 437 | |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 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); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | { |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 477 | struct ip_vs_lblc_table *tbl = svc->sched_data; |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 478 | struct ip_vs_iphdr iph; |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 479 | struct ip_vs_dest *dest = NULL; |
| 480 | struct ip_vs_lblc_entry *en; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 481 | |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 482 | ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph); |
| 483 | |
Hannes Eder | 1e3e238 | 2009-08-02 11:05:41 +0000 | [diff] [blame] | 484 | IP_VS_DBG(6, "%s(): Scheduling...\n", __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 485 | |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 486 | /* First look in our cache */ |
| 487 | read_lock(&svc->sched_lock); |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 488 | en = ip_vs_lblc_get(svc->af, tbl, &iph.daddr); |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 489 | if (en) { |
| 490 | /* We only hold a read lock, but this is atomic */ |
| 491 | en->lastuse = jiffies; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 492 | |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 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 */ |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 512 | dest = __ip_vs_lblc_schedule(svc); |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 513 | if (!dest) { |
Patrick Schaaf | 41ac51e | 2011-02-11 14:01:12 +0100 | [diff] [blame] | 514 | ip_vs_scheduler_err(svc, "no destination available"); |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 515 | return NULL; |
| 516 | } |
| 517 | |
| 518 | /* If we fail to create a cache entry, we'll just use the valid dest */ |
| 519 | write_lock(&svc->sched_lock); |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 520 | ip_vs_lblc_new(tbl, &iph.daddr, dest); |
Sven Wegener | 39ac50d | 2008-08-18 00:52:08 +0200 | [diff] [blame] | 521 | write_unlock(&svc->sched_lock); |
| 522 | |
| 523 | out: |
Julius Volz | 4454837 | 2008-11-03 17:08:28 -0800 | [diff] [blame] | 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)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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, |
Sven Wegener | d149ccc | 2008-08-10 09:18:02 +0000 | [diff] [blame] | 540 | .n_list = LIST_HEAD_INIT(ip_vs_lblc_scheduler.n_list), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 541 | .init_service = ip_vs_lblc_init_svc, |
| 542 | .done_service = ip_vs_lblc_done_svc, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 543 | .schedule = ip_vs_lblc_schedule, |
| 544 | }; |
| 545 | |
Hans Schillstrom | 61b1ab4 | 2011-01-03 14:44:42 +0100 | [diff] [blame] | 546 | /* |
| 547 | * per netns init. |
| 548 | */ |
Simon Horman | fb1de43 | 2011-02-04 18:33:02 +0900 | [diff] [blame] | 549 | #ifdef CONFIG_SYSCTL |
Hans Schillstrom | 61b1ab4 | 2011-01-03 14:44:42 +0100 | [diff] [blame] | 550 | static int __net_init __ip_vs_lblc_init(struct net *net) |
| 551 | { |
Hans Schillstrom | b6e885d | 2011-01-03 14:44:45 +0100 | [diff] [blame] | 552 | struct netns_ipvs *ipvs = net_ipvs(net); |
Hans Schillstrom | 61b1ab4 | 2011-01-03 14:44:42 +0100 | [diff] [blame] | 553 | |
Hans Schillstrom | 4b984cd | 2012-04-26 09:45:34 +0200 | [diff] [blame] | 554 | if (!ipvs) |
| 555 | return -ENOENT; |
| 556 | |
Hans Schillstrom | b6e885d | 2011-01-03 14:44:45 +0100 | [diff] [blame] | 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) |
Simon Horman | 0443929 | 2011-02-01 18:29:04 +0100 | [diff] [blame] | 562 | return -ENOMEM; |
Hans Schillstrom | b6e885d | 2011-01-03 14:44:45 +0100 | [diff] [blame] | 563 | } else |
| 564 | ipvs->lblc_ctl_table = vs_vars_table; |
Simon Horman | b27d777 | 2011-02-04 18:33:01 +0900 | [diff] [blame] | 565 | ipvs->sysctl_lblc_expiration = DEFAULT_EXPIRATION; |
Hans Schillstrom | b6e885d | 2011-01-03 14:44:45 +0100 | [diff] [blame] | 566 | ipvs->lblc_ctl_table[0].data = &ipvs->sysctl_lblc_expiration; |
| 567 | |
| 568 | ipvs->lblc_ctl_header = |
| 569 | register_net_sysctl_table(net, net_vs_ctl_path, |
| 570 | ipvs->lblc_ctl_table); |
Simon Horman | 0443929 | 2011-02-01 18:29:04 +0100 | [diff] [blame] | 571 | if (!ipvs->lblc_ctl_header) { |
| 572 | if (!net_eq(net, &init_net)) |
| 573 | kfree(ipvs->lblc_ctl_table); |
| 574 | return -ENOMEM; |
| 575 | } |
Hans Schillstrom | 61b1ab4 | 2011-01-03 14:44:42 +0100 | [diff] [blame] | 576 | |
| 577 | return 0; |
| 578 | } |
| 579 | |
| 580 | static void __net_exit __ip_vs_lblc_exit(struct net *net) |
| 581 | { |
Hans Schillstrom | b6e885d | 2011-01-03 14:44:45 +0100 | [diff] [blame] | 582 | struct netns_ipvs *ipvs = net_ipvs(net); |
Hans Schillstrom | 61b1ab4 | 2011-01-03 14:44:42 +0100 | [diff] [blame] | 583 | |
Hans Schillstrom | b6e885d | 2011-01-03 14:44:45 +0100 | [diff] [blame] | 584 | unregister_net_sysctl_table(ipvs->lblc_ctl_header); |
| 585 | |
| 586 | if (!net_eq(net, &init_net)) |
| 587 | kfree(ipvs->lblc_ctl_table); |
Hans Schillstrom | 61b1ab4 | 2011-01-03 14:44:42 +0100 | [diff] [blame] | 588 | } |
| 589 | |
Simon Horman | fb1de43 | 2011-02-04 18:33:02 +0900 | [diff] [blame] | 590 | #else |
| 591 | |
| 592 | static int __net_init __ip_vs_lblc_init(struct net *net) { return 0; } |
| 593 | static void __net_exit __ip_vs_lblc_exit(struct net *net) { } |
| 594 | |
| 595 | #endif |
| 596 | |
Hans Schillstrom | 61b1ab4 | 2011-01-03 14:44:42 +0100 | [diff] [blame] | 597 | static struct pernet_operations ip_vs_lblc_ops = { |
| 598 | .init = __ip_vs_lblc_init, |
| 599 | .exit = __ip_vs_lblc_exit, |
| 600 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 601 | |
| 602 | static int __init ip_vs_lblc_init(void) |
| 603 | { |
Pavel Emelyanov | a014bc8 | 2007-12-04 00:43:24 -0800 | [diff] [blame] | 604 | int ret; |
| 605 | |
Hans Schillstrom | 61b1ab4 | 2011-01-03 14:44:42 +0100 | [diff] [blame] | 606 | ret = register_pernet_subsys(&ip_vs_lblc_ops); |
| 607 | if (ret) |
| 608 | return ret; |
| 609 | |
Pavel Emelyanov | a014bc8 | 2007-12-04 00:43:24 -0800 | [diff] [blame] | 610 | ret = register_ip_vs_scheduler(&ip_vs_lblc_scheduler); |
| 611 | if (ret) |
Hans Schillstrom | 61b1ab4 | 2011-01-03 14:44:42 +0100 | [diff] [blame] | 612 | unregister_pernet_subsys(&ip_vs_lblc_ops); |
Pavel Emelyanov | a014bc8 | 2007-12-04 00:43:24 -0800 | [diff] [blame] | 613 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 614 | } |
| 615 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 616 | static void __exit ip_vs_lblc_cleanup(void) |
| 617 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 618 | unregister_ip_vs_scheduler(&ip_vs_lblc_scheduler); |
Hans Schillstrom | 61b1ab4 | 2011-01-03 14:44:42 +0100 | [diff] [blame] | 619 | unregister_pernet_subsys(&ip_vs_lblc_ops); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 620 | } |
| 621 | |
| 622 | |
| 623 | module_init(ip_vs_lblc_init); |
| 624 | module_exit(ip_vs_lblc_cleanup); |
| 625 | MODULE_LICENSE("GPL"); |