Merge tag 'trace-fixes-v3.10-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git...
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / netfilter / ipvs / ip_vs_sh.c
1 /*
2 * IPVS: Source Hashing scheduling module
3 *
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 * The weight destination attribute can be used to control the
34 * distribution of connections to the destinations in servernode. The
35 * greater the weight, the more connections the destination
36 * will receive.
37 *
38 */
39
40 #define KMSG_COMPONENT "IPVS"
41 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
42
43 #include <linux/ip.h>
44 #include <linux/slab.h>
45 #include <linux/module.h>
46 #include <linux/kernel.h>
47 #include <linux/skbuff.h>
48
49 #include <net/ip_vs.h>
50
51
52 /*
53 * IPVS SH bucket
54 */
55 struct ip_vs_sh_bucket {
56 struct ip_vs_dest __rcu *dest; /* real server (cache) */
57 };
58
59 /*
60 * for IPVS SH entry hash table
61 */
62 #ifndef CONFIG_IP_VS_SH_TAB_BITS
63 #define CONFIG_IP_VS_SH_TAB_BITS 8
64 #endif
65 #define IP_VS_SH_TAB_BITS CONFIG_IP_VS_SH_TAB_BITS
66 #define IP_VS_SH_TAB_SIZE (1 << IP_VS_SH_TAB_BITS)
67 #define IP_VS_SH_TAB_MASK (IP_VS_SH_TAB_SIZE - 1)
68
69 struct ip_vs_sh_state {
70 struct rcu_head rcu_head;
71 struct ip_vs_sh_bucket buckets[IP_VS_SH_TAB_SIZE];
72 };
73
74 /*
75 * Returns hash value for IPVS SH entry
76 */
77 static inline unsigned int ip_vs_sh_hashkey(int af, const union nf_inet_addr *addr)
78 {
79 __be32 addr_fold = addr->ip;
80
81 #ifdef CONFIG_IP_VS_IPV6
82 if (af == AF_INET6)
83 addr_fold = addr->ip6[0]^addr->ip6[1]^
84 addr->ip6[2]^addr->ip6[3];
85 #endif
86 return (ntohl(addr_fold)*2654435761UL) & IP_VS_SH_TAB_MASK;
87 }
88
89
90 /*
91 * Get ip_vs_dest associated with supplied parameters.
92 */
93 static inline struct ip_vs_dest *
94 ip_vs_sh_get(int af, struct ip_vs_sh_state *s, const union nf_inet_addr *addr)
95 {
96 return rcu_dereference(s->buckets[ip_vs_sh_hashkey(af, addr)].dest);
97 }
98
99
100 /*
101 * Assign all the hash buckets of the specified table with the service.
102 */
103 static int
104 ip_vs_sh_reassign(struct ip_vs_sh_state *s, struct ip_vs_service *svc)
105 {
106 int i;
107 struct ip_vs_sh_bucket *b;
108 struct list_head *p;
109 struct ip_vs_dest *dest;
110 int d_count;
111 bool empty;
112
113 b = &s->buckets[0];
114 p = &svc->destinations;
115 empty = list_empty(p);
116 d_count = 0;
117 for (i=0; i<IP_VS_SH_TAB_SIZE; i++) {
118 dest = rcu_dereference_protected(b->dest, 1);
119 if (dest)
120 ip_vs_dest_put(dest);
121 if (empty)
122 RCU_INIT_POINTER(b->dest, NULL);
123 else {
124 if (p == &svc->destinations)
125 p = p->next;
126
127 dest = list_entry(p, struct ip_vs_dest, n_list);
128 ip_vs_dest_hold(dest);
129 RCU_INIT_POINTER(b->dest, dest);
130
131 IP_VS_DBG_BUF(6, "assigned i: %d dest: %s weight: %d\n",
132 i, IP_VS_DBG_ADDR(svc->af, &dest->addr),
133 atomic_read(&dest->weight));
134
135 /* Don't move to next dest until filling weight */
136 if (++d_count >= atomic_read(&dest->weight)) {
137 p = p->next;
138 d_count = 0;
139 }
140
141 }
142 b++;
143 }
144 return 0;
145 }
146
147
148 /*
149 * Flush all the hash buckets of the specified table.
150 */
151 static void ip_vs_sh_flush(struct ip_vs_sh_state *s)
152 {
153 int i;
154 struct ip_vs_sh_bucket *b;
155 struct ip_vs_dest *dest;
156
157 b = &s->buckets[0];
158 for (i=0; i<IP_VS_SH_TAB_SIZE; i++) {
159 dest = rcu_dereference_protected(b->dest, 1);
160 if (dest) {
161 ip_vs_dest_put(dest);
162 RCU_INIT_POINTER(b->dest, NULL);
163 }
164 b++;
165 }
166 }
167
168
169 static int ip_vs_sh_init_svc(struct ip_vs_service *svc)
170 {
171 struct ip_vs_sh_state *s;
172
173 /* allocate the SH table for this service */
174 s = kzalloc(sizeof(struct ip_vs_sh_state), GFP_KERNEL);
175 if (s == NULL)
176 return -ENOMEM;
177
178 svc->sched_data = s;
179 IP_VS_DBG(6, "SH hash table (memory=%Zdbytes) allocated for "
180 "current service\n",
181 sizeof(struct ip_vs_sh_bucket)*IP_VS_SH_TAB_SIZE);
182
183 /* assign the hash buckets with current dests */
184 ip_vs_sh_reassign(s, svc);
185
186 return 0;
187 }
188
189
190 static void ip_vs_sh_done_svc(struct ip_vs_service *svc)
191 {
192 struct ip_vs_sh_state *s = svc->sched_data;
193
194 /* got to clean up hash buckets here */
195 ip_vs_sh_flush(s);
196
197 /* release the table itself */
198 kfree_rcu(s, rcu_head);
199 IP_VS_DBG(6, "SH hash table (memory=%Zdbytes) released\n",
200 sizeof(struct ip_vs_sh_bucket)*IP_VS_SH_TAB_SIZE);
201 }
202
203
204 static int ip_vs_sh_dest_changed(struct ip_vs_service *svc,
205 struct ip_vs_dest *dest)
206 {
207 struct ip_vs_sh_state *s = svc->sched_data;
208
209 /* assign the hash buckets with the updated service */
210 ip_vs_sh_reassign(s, svc);
211
212 return 0;
213 }
214
215
216 /*
217 * If the dest flags is set with IP_VS_DEST_F_OVERLOAD,
218 * consider that the server is overloaded here.
219 */
220 static inline int is_overloaded(struct ip_vs_dest *dest)
221 {
222 return dest->flags & IP_VS_DEST_F_OVERLOAD;
223 }
224
225
226 /*
227 * Source Hashing scheduling
228 */
229 static struct ip_vs_dest *
230 ip_vs_sh_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
231 {
232 struct ip_vs_dest *dest;
233 struct ip_vs_sh_state *s;
234 struct ip_vs_iphdr iph;
235
236 ip_vs_fill_iph_addr_only(svc->af, skb, &iph);
237
238 IP_VS_DBG(6, "ip_vs_sh_schedule(): Scheduling...\n");
239
240 s = (struct ip_vs_sh_state *) svc->sched_data;
241 dest = ip_vs_sh_get(svc->af, s, &iph.saddr);
242 if (!dest
243 || !(dest->flags & IP_VS_DEST_F_AVAILABLE)
244 || atomic_read(&dest->weight) <= 0
245 || is_overloaded(dest)) {
246 ip_vs_scheduler_err(svc, "no destination available");
247 return NULL;
248 }
249
250 IP_VS_DBG_BUF(6, "SH: source IP address %s --> server %s:%d\n",
251 IP_VS_DBG_ADDR(svc->af, &iph.saddr),
252 IP_VS_DBG_ADDR(svc->af, &dest->addr),
253 ntohs(dest->port));
254
255 return dest;
256 }
257
258
259 /*
260 * IPVS SH Scheduler structure
261 */
262 static struct ip_vs_scheduler ip_vs_sh_scheduler =
263 {
264 .name = "sh",
265 .refcnt = ATOMIC_INIT(0),
266 .module = THIS_MODULE,
267 .n_list = LIST_HEAD_INIT(ip_vs_sh_scheduler.n_list),
268 .init_service = ip_vs_sh_init_svc,
269 .done_service = ip_vs_sh_done_svc,
270 .add_dest = ip_vs_sh_dest_changed,
271 .del_dest = ip_vs_sh_dest_changed,
272 .upd_dest = ip_vs_sh_dest_changed,
273 .schedule = ip_vs_sh_schedule,
274 };
275
276
277 static int __init ip_vs_sh_init(void)
278 {
279 return register_ip_vs_scheduler(&ip_vs_sh_scheduler);
280 }
281
282
283 static void __exit ip_vs_sh_cleanup(void)
284 {
285 unregister_ip_vs_scheduler(&ip_vs_sh_scheduler);
286 synchronize_rcu();
287 }
288
289
290 module_init(ip_vs_sh_init);
291 module_exit(ip_vs_sh_cleanup);
292 MODULE_LICENSE("GPL");