2 * IPVS: Source Hashing scheduling module
4 * Authors: Wensong Zhang <wensong@gnuchina.org>
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.
16 * The sh algorithm is to select server by the hash key of source IP
17 * address. The pseudo code is as follows:
19 * n <- servernode[src_ip];
21 * (n is overloaded) or (n.weight <= 0) then
26 * Notes that servernode is a 256-bucket hash table that maps the hash
27 * index derived from packet source IP address to the current server
28 * array. If the sh scheduler is used in cache cluster, it is good to
29 * combine it with cache_bypass feature. When the statically assigned
30 * server is dead or overloaded, the load balancer can bypass the cache
31 * server and send requests to the original server directly.
36 #include <linux/module.h>
37 #include <linux/kernel.h>
38 #include <linux/skbuff.h>
40 #include <net/ip_vs.h>
46 struct ip_vs_sh_bucket
{
47 struct ip_vs_dest
*dest
; /* real server (cache) */
51 * for IPVS SH entry hash table
53 #ifndef CONFIG_IP_VS_SH_TAB_BITS
54 #define CONFIG_IP_VS_SH_TAB_BITS 8
56 #define IP_VS_SH_TAB_BITS CONFIG_IP_VS_SH_TAB_BITS
57 #define IP_VS_SH_TAB_SIZE (1 << IP_VS_SH_TAB_BITS)
58 #define IP_VS_SH_TAB_MASK (IP_VS_SH_TAB_SIZE - 1)
62 * Returns hash value for IPVS SH entry
64 static inline unsigned ip_vs_sh_hashkey(int af
, const union nf_inet_addr
*addr
)
66 __be32 addr_fold
= addr
->ip
;
68 #ifdef CONFIG_IP_VS_IPV6
70 addr_fold
= addr
->ip6
[0]^addr
->ip6
[1]^
71 addr
->ip6
[2]^addr
->ip6
[3];
73 return (ntohl(addr_fold
)*2654435761UL) & IP_VS_SH_TAB_MASK
;
78 * Get ip_vs_dest associated with supplied parameters.
80 static inline struct ip_vs_dest
*
81 ip_vs_sh_get(int af
, struct ip_vs_sh_bucket
*tbl
,
82 const union nf_inet_addr
*addr
)
84 return (tbl
[ip_vs_sh_hashkey(af
, addr
)]).dest
;
89 * Assign all the hash buckets of the specified table with the service.
92 ip_vs_sh_assign(struct ip_vs_sh_bucket
*tbl
, struct ip_vs_service
*svc
)
95 struct ip_vs_sh_bucket
*b
;
97 struct ip_vs_dest
*dest
;
100 p
= &svc
->destinations
;
101 for (i
=0; i
<IP_VS_SH_TAB_SIZE
; i
++) {
105 if (p
== &svc
->destinations
)
108 dest
= list_entry(p
, struct ip_vs_dest
, n_list
);
109 atomic_inc(&dest
->refcnt
);
121 * Flush all the hash buckets of the specified table.
123 static void ip_vs_sh_flush(struct ip_vs_sh_bucket
*tbl
)
126 struct ip_vs_sh_bucket
*b
;
129 for (i
=0; i
<IP_VS_SH_TAB_SIZE
; i
++) {
131 atomic_dec(&b
->dest
->refcnt
);
139 static int ip_vs_sh_init_svc(struct ip_vs_service
*svc
)
141 struct ip_vs_sh_bucket
*tbl
;
143 /* allocate the SH table for this service */
144 tbl
= kmalloc(sizeof(struct ip_vs_sh_bucket
)*IP_VS_SH_TAB_SIZE
,
147 IP_VS_ERR("ip_vs_sh_init_svc(): no memory\n");
150 svc
->sched_data
= tbl
;
151 IP_VS_DBG(6, "SH hash table (memory=%Zdbytes) allocated for "
153 sizeof(struct ip_vs_sh_bucket
)*IP_VS_SH_TAB_SIZE
);
155 /* assign the hash buckets with the updated service */
156 ip_vs_sh_assign(tbl
, svc
);
162 static int ip_vs_sh_done_svc(struct ip_vs_service
*svc
)
164 struct ip_vs_sh_bucket
*tbl
= svc
->sched_data
;
166 /* got to clean up hash buckets here */
169 /* release the table itself */
170 kfree(svc
->sched_data
);
171 IP_VS_DBG(6, "SH hash table (memory=%Zdbytes) released\n",
172 sizeof(struct ip_vs_sh_bucket
)*IP_VS_SH_TAB_SIZE
);
178 static int ip_vs_sh_update_svc(struct ip_vs_service
*svc
)
180 struct ip_vs_sh_bucket
*tbl
= svc
->sched_data
;
182 /* got to clean up hash buckets here */
185 /* assign the hash buckets with the updated service */
186 ip_vs_sh_assign(tbl
, svc
);
193 * If the dest flags is set with IP_VS_DEST_F_OVERLOAD,
194 * consider that the server is overloaded here.
196 static inline int is_overloaded(struct ip_vs_dest
*dest
)
198 return dest
->flags
& IP_VS_DEST_F_OVERLOAD
;
203 * Source Hashing scheduling
205 static struct ip_vs_dest
*
206 ip_vs_sh_schedule(struct ip_vs_service
*svc
, const struct sk_buff
*skb
)
208 struct ip_vs_dest
*dest
;
209 struct ip_vs_sh_bucket
*tbl
;
210 struct ip_vs_iphdr iph
;
212 ip_vs_fill_iphdr(svc
->af
, skb_network_header(skb
), &iph
);
214 IP_VS_DBG(6, "ip_vs_sh_schedule(): Scheduling...\n");
216 tbl
= (struct ip_vs_sh_bucket
*)svc
->sched_data
;
217 dest
= ip_vs_sh_get(svc
->af
, tbl
, &iph
.saddr
);
219 || !(dest
->flags
& IP_VS_DEST_F_AVAILABLE
)
220 || atomic_read(&dest
->weight
) <= 0
221 || is_overloaded(dest
)) {
222 IP_VS_ERR_RL("SH: no destination available\n");
226 IP_VS_DBG_BUF(6, "SH: source IP address %s --> server %s:%d\n",
227 IP_VS_DBG_ADDR(svc
->af
, &iph
.saddr
),
228 IP_VS_DBG_ADDR(svc
->af
, &dest
->addr
),
236 * IPVS SH Scheduler structure
238 static struct ip_vs_scheduler ip_vs_sh_scheduler
=
241 .refcnt
= ATOMIC_INIT(0),
242 .module
= THIS_MODULE
,
243 .n_list
= LIST_HEAD_INIT(ip_vs_sh_scheduler
.n_list
),
244 .init_service
= ip_vs_sh_init_svc
,
245 .done_service
= ip_vs_sh_done_svc
,
246 .update_service
= ip_vs_sh_update_svc
,
247 .schedule
= ip_vs_sh_schedule
,
251 static int __init
ip_vs_sh_init(void)
253 return register_ip_vs_scheduler(&ip_vs_sh_scheduler
);
257 static void __exit
ip_vs_sh_cleanup(void)
259 unregister_ip_vs_scheduler(&ip_vs_sh_scheduler
);
263 module_init(ip_vs_sh_init
);
264 module_exit(ip_vs_sh_cleanup
);
265 MODULE_LICENSE("GPL");