Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * IPVS: Source Hashing 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 | * | |
13 | */ | |
14 | ||
15 | /* | |
16 | * The sh algorithm is to select server by the hash key of source IP | |
17 | * address. The pseudo code is as follows: | |
18 | * | |
19 | * n <- servernode[src_ip]; | |
20 | * if (n is dead) OR | |
21 | * (n is overloaded) or (n.weight <= 0) then | |
22 | * return NULL; | |
23 | * | |
24 | * return n; | |
25 | * | |
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. | |
32 | * | |
33 | */ | |
34 | ||
14c85021 | 35 | #include <linux/ip.h> |
1da177e4 LT |
36 | #include <linux/module.h> |
37 | #include <linux/kernel.h> | |
14c85021 | 38 | #include <linux/skbuff.h> |
1da177e4 LT |
39 | |
40 | #include <net/ip_vs.h> | |
41 | ||
42 | ||
43 | /* | |
44 | * IPVS SH bucket | |
45 | */ | |
46 | struct ip_vs_sh_bucket { | |
47 | struct ip_vs_dest *dest; /* real server (cache) */ | |
48 | }; | |
49 | ||
50 | /* | |
51 | * for IPVS SH entry hash table | |
52 | */ | |
53 | #ifndef CONFIG_IP_VS_SH_TAB_BITS | |
54 | #define CONFIG_IP_VS_SH_TAB_BITS 8 | |
55 | #endif | |
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) | |
59 | ||
60 | ||
61 | /* | |
62 | * Returns hash value for IPVS SH entry | |
63 | */ | |
014d730d | 64 | static inline unsigned ip_vs_sh_hashkey(__be32 addr) |
1da177e4 LT |
65 | { |
66 | return (ntohl(addr)*2654435761UL) & IP_VS_SH_TAB_MASK; | |
67 | } | |
68 | ||
69 | ||
70 | /* | |
71 | * Get ip_vs_dest associated with supplied parameters. | |
72 | */ | |
73 | static inline struct ip_vs_dest * | |
014d730d | 74 | ip_vs_sh_get(struct ip_vs_sh_bucket *tbl, __be32 addr) |
1da177e4 LT |
75 | { |
76 | return (tbl[ip_vs_sh_hashkey(addr)]).dest; | |
77 | } | |
78 | ||
79 | ||
80 | /* | |
81 | * Assign all the hash buckets of the specified table with the service. | |
82 | */ | |
83 | static int | |
84 | ip_vs_sh_assign(struct ip_vs_sh_bucket *tbl, struct ip_vs_service *svc) | |
85 | { | |
86 | int i; | |
87 | struct ip_vs_sh_bucket *b; | |
88 | struct list_head *p; | |
89 | struct ip_vs_dest *dest; | |
90 | ||
91 | b = tbl; | |
92 | p = &svc->destinations; | |
93 | for (i=0; i<IP_VS_SH_TAB_SIZE; i++) { | |
94 | if (list_empty(p)) { | |
95 | b->dest = NULL; | |
96 | } else { | |
97 | if (p == &svc->destinations) | |
98 | p = p->next; | |
99 | ||
100 | dest = list_entry(p, struct ip_vs_dest, n_list); | |
101 | atomic_inc(&dest->refcnt); | |
102 | b->dest = dest; | |
103 | ||
104 | p = p->next; | |
105 | } | |
106 | b++; | |
107 | } | |
108 | return 0; | |
109 | } | |
110 | ||
111 | ||
112 | /* | |
113 | * Flush all the hash buckets of the specified table. | |
114 | */ | |
115 | static void ip_vs_sh_flush(struct ip_vs_sh_bucket *tbl) | |
116 | { | |
117 | int i; | |
118 | struct ip_vs_sh_bucket *b; | |
119 | ||
120 | b = tbl; | |
121 | for (i=0; i<IP_VS_SH_TAB_SIZE; i++) { | |
122 | if (b->dest) { | |
123 | atomic_dec(&b->dest->refcnt); | |
124 | b->dest = NULL; | |
125 | } | |
126 | b++; | |
127 | } | |
128 | } | |
129 | ||
130 | ||
131 | static int ip_vs_sh_init_svc(struct ip_vs_service *svc) | |
132 | { | |
133 | struct ip_vs_sh_bucket *tbl; | |
134 | ||
135 | /* allocate the SH table for this service */ | |
136 | tbl = kmalloc(sizeof(struct ip_vs_sh_bucket)*IP_VS_SH_TAB_SIZE, | |
137 | GFP_ATOMIC); | |
138 | if (tbl == NULL) { | |
139 | IP_VS_ERR("ip_vs_sh_init_svc(): no memory\n"); | |
140 | return -ENOMEM; | |
141 | } | |
142 | svc->sched_data = tbl; | |
143 | IP_VS_DBG(6, "SH hash table (memory=%Zdbytes) allocated for " | |
144 | "current service\n", | |
145 | sizeof(struct ip_vs_sh_bucket)*IP_VS_SH_TAB_SIZE); | |
146 | ||
147 | /* assign the hash buckets with the updated service */ | |
148 | ip_vs_sh_assign(tbl, svc); | |
149 | ||
150 | return 0; | |
151 | } | |
152 | ||
153 | ||
154 | static int ip_vs_sh_done_svc(struct ip_vs_service *svc) | |
155 | { | |
156 | struct ip_vs_sh_bucket *tbl = svc->sched_data; | |
157 | ||
158 | /* got to clean up hash buckets here */ | |
159 | ip_vs_sh_flush(tbl); | |
160 | ||
161 | /* release the table itself */ | |
162 | kfree(svc->sched_data); | |
163 | IP_VS_DBG(6, "SH hash table (memory=%Zdbytes) released\n", | |
164 | sizeof(struct ip_vs_sh_bucket)*IP_VS_SH_TAB_SIZE); | |
165 | ||
166 | return 0; | |
167 | } | |
168 | ||
169 | ||
170 | static int ip_vs_sh_update_svc(struct ip_vs_service *svc) | |
171 | { | |
172 | struct ip_vs_sh_bucket *tbl = svc->sched_data; | |
173 | ||
174 | /* got to clean up hash buckets here */ | |
175 | ip_vs_sh_flush(tbl); | |
176 | ||
177 | /* assign the hash buckets with the updated service */ | |
178 | ip_vs_sh_assign(tbl, svc); | |
179 | ||
180 | return 0; | |
181 | } | |
182 | ||
183 | ||
184 | /* | |
185 | * If the dest flags is set with IP_VS_DEST_F_OVERLOAD, | |
186 | * consider that the server is overloaded here. | |
187 | */ | |
188 | static inline int is_overloaded(struct ip_vs_dest *dest) | |
189 | { | |
190 | return dest->flags & IP_VS_DEST_F_OVERLOAD; | |
191 | } | |
192 | ||
193 | ||
194 | /* | |
195 | * Source Hashing scheduling | |
196 | */ | |
197 | static struct ip_vs_dest * | |
198 | ip_vs_sh_schedule(struct ip_vs_service *svc, const struct sk_buff *skb) | |
199 | { | |
200 | struct ip_vs_dest *dest; | |
201 | struct ip_vs_sh_bucket *tbl; | |
eddc9ec5 | 202 | struct iphdr *iph = ip_hdr(skb); |
1da177e4 LT |
203 | |
204 | IP_VS_DBG(6, "ip_vs_sh_schedule(): Scheduling...\n"); | |
205 | ||
206 | tbl = (struct ip_vs_sh_bucket *)svc->sched_data; | |
207 | dest = ip_vs_sh_get(tbl, iph->saddr); | |
208 | if (!dest | |
209 | || !(dest->flags & IP_VS_DEST_F_AVAILABLE) | |
210 | || atomic_read(&dest->weight) <= 0 | |
211 | || is_overloaded(dest)) { | |
212 | return NULL; | |
213 | } | |
214 | ||
215 | IP_VS_DBG(6, "SH: source IP address %u.%u.%u.%u " | |
216 | "--> server %u.%u.%u.%u:%d\n", | |
217 | NIPQUAD(iph->saddr), | |
218 | NIPQUAD(dest->addr), | |
219 | ntohs(dest->port)); | |
220 | ||
221 | return dest; | |
222 | } | |
223 | ||
224 | ||
225 | /* | |
226 | * IPVS SH Scheduler structure | |
227 | */ | |
228 | static struct ip_vs_scheduler ip_vs_sh_scheduler = | |
229 | { | |
230 | .name = "sh", | |
231 | .refcnt = ATOMIC_INIT(0), | |
232 | .module = THIS_MODULE, | |
233 | .init_service = ip_vs_sh_init_svc, | |
234 | .done_service = ip_vs_sh_done_svc, | |
235 | .update_service = ip_vs_sh_update_svc, | |
236 | .schedule = ip_vs_sh_schedule, | |
237 | }; | |
238 | ||
239 | ||
240 | static int __init ip_vs_sh_init(void) | |
241 | { | |
242 | INIT_LIST_HEAD(&ip_vs_sh_scheduler.n_list); | |
243 | return register_ip_vs_scheduler(&ip_vs_sh_scheduler); | |
244 | } | |
245 | ||
246 | ||
247 | static void __exit ip_vs_sh_cleanup(void) | |
248 | { | |
249 | unregister_ip_vs_scheduler(&ip_vs_sh_scheduler); | |
250 | } | |
251 | ||
252 | ||
253 | module_init(ip_vs_sh_init); | |
254 | module_exit(ip_vs_sh_cleanup); | |
255 | MODULE_LICENSE("GPL"); |