]>
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 | */ | |
20971a0a | 64 | static inline unsigned ip_vs_sh_hashkey(int af, const union nf_inet_addr *addr) |
1da177e4 | 65 | { |
20971a0a JV |
66 | __be32 addr_fold = addr->ip; |
67 | ||
68 | #ifdef CONFIG_IP_VS_IPV6 | |
69 | if (af == AF_INET6) | |
70 | addr_fold = addr->ip6[0]^addr->ip6[1]^ | |
71 | addr->ip6[2]^addr->ip6[3]; | |
72 | #endif | |
73 | return (ntohl(addr_fold)*2654435761UL) & IP_VS_SH_TAB_MASK; | |
1da177e4 LT |
74 | } |
75 | ||
76 | ||
77 | /* | |
78 | * Get ip_vs_dest associated with supplied parameters. | |
79 | */ | |
80 | static inline struct ip_vs_dest * | |
20971a0a JV |
81 | ip_vs_sh_get(int af, struct ip_vs_sh_bucket *tbl, |
82 | const union nf_inet_addr *addr) | |
1da177e4 | 83 | { |
20971a0a | 84 | return (tbl[ip_vs_sh_hashkey(af, addr)]).dest; |
1da177e4 LT |
85 | } |
86 | ||
87 | ||
88 | /* | |
89 | * Assign all the hash buckets of the specified table with the service. | |
90 | */ | |
91 | static int | |
92 | ip_vs_sh_assign(struct ip_vs_sh_bucket *tbl, struct ip_vs_service *svc) | |
93 | { | |
94 | int i; | |
95 | struct ip_vs_sh_bucket *b; | |
96 | struct list_head *p; | |
97 | struct ip_vs_dest *dest; | |
98 | ||
99 | b = tbl; | |
100 | p = &svc->destinations; | |
101 | for (i=0; i<IP_VS_SH_TAB_SIZE; i++) { | |
102 | if (list_empty(p)) { | |
103 | b->dest = NULL; | |
104 | } else { | |
105 | if (p == &svc->destinations) | |
106 | p = p->next; | |
107 | ||
108 | dest = list_entry(p, struct ip_vs_dest, n_list); | |
109 | atomic_inc(&dest->refcnt); | |
110 | b->dest = dest; | |
111 | ||
112 | p = p->next; | |
113 | } | |
114 | b++; | |
115 | } | |
116 | return 0; | |
117 | } | |
118 | ||
119 | ||
120 | /* | |
121 | * Flush all the hash buckets of the specified table. | |
122 | */ | |
123 | static void ip_vs_sh_flush(struct ip_vs_sh_bucket *tbl) | |
124 | { | |
125 | int i; | |
126 | struct ip_vs_sh_bucket *b; | |
127 | ||
128 | b = tbl; | |
129 | for (i=0; i<IP_VS_SH_TAB_SIZE; i++) { | |
130 | if (b->dest) { | |
131 | atomic_dec(&b->dest->refcnt); | |
132 | b->dest = NULL; | |
133 | } | |
134 | b++; | |
135 | } | |
136 | } | |
137 | ||
138 | ||
139 | static int ip_vs_sh_init_svc(struct ip_vs_service *svc) | |
140 | { | |
141 | struct ip_vs_sh_bucket *tbl; | |
142 | ||
143 | /* allocate the SH table for this service */ | |
144 | tbl = kmalloc(sizeof(struct ip_vs_sh_bucket)*IP_VS_SH_TAB_SIZE, | |
145 | GFP_ATOMIC); | |
146 | if (tbl == NULL) { | |
147 | IP_VS_ERR("ip_vs_sh_init_svc(): no memory\n"); | |
148 | return -ENOMEM; | |
149 | } | |
150 | svc->sched_data = tbl; | |
151 | IP_VS_DBG(6, "SH hash table (memory=%Zdbytes) allocated for " | |
152 | "current service\n", | |
153 | sizeof(struct ip_vs_sh_bucket)*IP_VS_SH_TAB_SIZE); | |
154 | ||
155 | /* assign the hash buckets with the updated service */ | |
156 | ip_vs_sh_assign(tbl, svc); | |
157 | ||
158 | return 0; | |
159 | } | |
160 | ||
161 | ||
162 | static int ip_vs_sh_done_svc(struct ip_vs_service *svc) | |
163 | { | |
164 | struct ip_vs_sh_bucket *tbl = svc->sched_data; | |
165 | ||
166 | /* got to clean up hash buckets here */ | |
167 | ip_vs_sh_flush(tbl); | |
168 | ||
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); | |
173 | ||
174 | return 0; | |
175 | } | |
176 | ||
177 | ||
178 | static int ip_vs_sh_update_svc(struct ip_vs_service *svc) | |
179 | { | |
180 | struct ip_vs_sh_bucket *tbl = svc->sched_data; | |
181 | ||
182 | /* got to clean up hash buckets here */ | |
183 | ip_vs_sh_flush(tbl); | |
184 | ||
185 | /* assign the hash buckets with the updated service */ | |
186 | ip_vs_sh_assign(tbl, svc); | |
187 | ||
188 | return 0; | |
189 | } | |
190 | ||
191 | ||
192 | /* | |
193 | * If the dest flags is set with IP_VS_DEST_F_OVERLOAD, | |
194 | * consider that the server is overloaded here. | |
195 | */ | |
196 | static inline int is_overloaded(struct ip_vs_dest *dest) | |
197 | { | |
198 | return dest->flags & IP_VS_DEST_F_OVERLOAD; | |
199 | } | |
200 | ||
201 | ||
202 | /* | |
203 | * Source Hashing scheduling | |
204 | */ | |
205 | static struct ip_vs_dest * | |
206 | ip_vs_sh_schedule(struct ip_vs_service *svc, const struct sk_buff *skb) | |
207 | { | |
208 | struct ip_vs_dest *dest; | |
209 | struct ip_vs_sh_bucket *tbl; | |
20971a0a JV |
210 | struct ip_vs_iphdr iph; |
211 | ||
212 | ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph); | |
1da177e4 LT |
213 | |
214 | IP_VS_DBG(6, "ip_vs_sh_schedule(): Scheduling...\n"); | |
215 | ||
216 | tbl = (struct ip_vs_sh_bucket *)svc->sched_data; | |
20971a0a | 217 | dest = ip_vs_sh_get(svc->af, tbl, &iph.saddr); |
1da177e4 LT |
218 | if (!dest |
219 | || !(dest->flags & IP_VS_DEST_F_AVAILABLE) | |
220 | || atomic_read(&dest->weight) <= 0 | |
221 | || is_overloaded(dest)) { | |
68888d10 | 222 | IP_VS_ERR_RL("SH: no destination available\n"); |
1da177e4 LT |
223 | return NULL; |
224 | } | |
225 | ||
20971a0a JV |
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), | |
229 | ntohs(dest->port)); | |
1da177e4 LT |
230 | |
231 | return dest; | |
232 | } | |
233 | ||
234 | ||
235 | /* | |
236 | * IPVS SH Scheduler structure | |
237 | */ | |
238 | static struct ip_vs_scheduler ip_vs_sh_scheduler = | |
239 | { | |
240 | .name = "sh", | |
241 | .refcnt = ATOMIC_INIT(0), | |
242 | .module = THIS_MODULE, | |
d149ccc9 | 243 | .n_list = LIST_HEAD_INIT(ip_vs_sh_scheduler.n_list), |
1da177e4 LT |
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, | |
248 | }; | |
249 | ||
250 | ||
251 | static int __init ip_vs_sh_init(void) | |
252 | { | |
1da177e4 LT |
253 | return register_ip_vs_scheduler(&ip_vs_sh_scheduler); |
254 | } | |
255 | ||
256 | ||
257 | static void __exit ip_vs_sh_cleanup(void) | |
258 | { | |
259 | unregister_ip_vs_scheduler(&ip_vs_sh_scheduler); | |
260 | } | |
261 | ||
262 | ||
263 | module_init(ip_vs_sh_init); | |
264 | module_exit(ip_vs_sh_cleanup); | |
265 | MODULE_LICENSE("GPL"); |