xref: /openbmc/linux/net/netfilter/ipvs/ip_vs_sh.c (revision d0b73b48)
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       *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 
70 /*
71  *	Returns hash value for IPVS SH entry
72  */
73 static inline unsigned int ip_vs_sh_hashkey(int af, const union nf_inet_addr *addr)
74 {
75 	__be32 addr_fold = addr->ip;
76 
77 #ifdef CONFIG_IP_VS_IPV6
78 	if (af == AF_INET6)
79 		addr_fold = addr->ip6[0]^addr->ip6[1]^
80 			    addr->ip6[2]^addr->ip6[3];
81 #endif
82 	return (ntohl(addr_fold)*2654435761UL) & IP_VS_SH_TAB_MASK;
83 }
84 
85 
86 /*
87  *      Get ip_vs_dest associated with supplied parameters.
88  */
89 static inline struct ip_vs_dest *
90 ip_vs_sh_get(int af, struct ip_vs_sh_bucket *tbl,
91 	     const union nf_inet_addr *addr)
92 {
93 	return (tbl[ip_vs_sh_hashkey(af, addr)]).dest;
94 }
95 
96 
97 /*
98  *      Assign all the hash buckets of the specified table with the service.
99  */
100 static int
101 ip_vs_sh_assign(struct ip_vs_sh_bucket *tbl, struct ip_vs_service *svc)
102 {
103 	int i;
104 	struct ip_vs_sh_bucket *b;
105 	struct list_head *p;
106 	struct ip_vs_dest *dest;
107 	int d_count;
108 
109 	b = tbl;
110 	p = &svc->destinations;
111 	d_count = 0;
112 	for (i=0; i<IP_VS_SH_TAB_SIZE; i++) {
113 		if (list_empty(p)) {
114 			b->dest = NULL;
115 		} else {
116 			if (p == &svc->destinations)
117 				p = p->next;
118 
119 			dest = list_entry(p, struct ip_vs_dest, n_list);
120 			atomic_inc(&dest->refcnt);
121 			b->dest = dest;
122 
123 			IP_VS_DBG_BUF(6, "assigned i: %d dest: %s weight: %d\n",
124 				      i, IP_VS_DBG_ADDR(svc->af, &dest->addr),
125 				      atomic_read(&dest->weight));
126 
127 			/* Don't move to next dest until filling weight */
128 			if (++d_count >= atomic_read(&dest->weight)) {
129 				p = p->next;
130 				d_count = 0;
131 			}
132 
133 		}
134 		b++;
135 	}
136 	return 0;
137 }
138 
139 
140 /*
141  *      Flush all the hash buckets of the specified table.
142  */
143 static void ip_vs_sh_flush(struct ip_vs_sh_bucket *tbl)
144 {
145 	int i;
146 	struct ip_vs_sh_bucket *b;
147 
148 	b = tbl;
149 	for (i=0; i<IP_VS_SH_TAB_SIZE; i++) {
150 		if (b->dest) {
151 			atomic_dec(&b->dest->refcnt);
152 			b->dest = NULL;
153 		}
154 		b++;
155 	}
156 }
157 
158 
159 static int ip_vs_sh_init_svc(struct ip_vs_service *svc)
160 {
161 	struct ip_vs_sh_bucket *tbl;
162 
163 	/* allocate the SH table for this service */
164 	tbl = kmalloc(sizeof(struct ip_vs_sh_bucket)*IP_VS_SH_TAB_SIZE,
165 		      GFP_KERNEL);
166 	if (tbl == NULL)
167 		return -ENOMEM;
168 
169 	svc->sched_data = tbl;
170 	IP_VS_DBG(6, "SH hash table (memory=%Zdbytes) allocated for "
171 		  "current service\n",
172 		  sizeof(struct ip_vs_sh_bucket)*IP_VS_SH_TAB_SIZE);
173 
174 	/* assign the hash buckets with the updated service */
175 	ip_vs_sh_assign(tbl, svc);
176 
177 	return 0;
178 }
179 
180 
181 static int ip_vs_sh_done_svc(struct ip_vs_service *svc)
182 {
183 	struct ip_vs_sh_bucket *tbl = svc->sched_data;
184 
185 	/* got to clean up hash buckets here */
186 	ip_vs_sh_flush(tbl);
187 
188 	/* release the table itself */
189 	kfree(svc->sched_data);
190 	IP_VS_DBG(6, "SH hash table (memory=%Zdbytes) released\n",
191 		  sizeof(struct ip_vs_sh_bucket)*IP_VS_SH_TAB_SIZE);
192 
193 	return 0;
194 }
195 
196 
197 static int ip_vs_sh_update_svc(struct ip_vs_service *svc)
198 {
199 	struct ip_vs_sh_bucket *tbl = svc->sched_data;
200 
201 	/* got to clean up hash buckets here */
202 	ip_vs_sh_flush(tbl);
203 
204 	/* assign the hash buckets with the updated service */
205 	ip_vs_sh_assign(tbl, svc);
206 
207 	return 0;
208 }
209 
210 
211 /*
212  *      If the dest flags is set with IP_VS_DEST_F_OVERLOAD,
213  *      consider that the server is overloaded here.
214  */
215 static inline int is_overloaded(struct ip_vs_dest *dest)
216 {
217 	return dest->flags & IP_VS_DEST_F_OVERLOAD;
218 }
219 
220 
221 /*
222  *      Source Hashing scheduling
223  */
224 static struct ip_vs_dest *
225 ip_vs_sh_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
226 {
227 	struct ip_vs_dest *dest;
228 	struct ip_vs_sh_bucket *tbl;
229 	struct ip_vs_iphdr iph;
230 
231 	ip_vs_fill_iph_addr_only(svc->af, skb, &iph);
232 
233 	IP_VS_DBG(6, "ip_vs_sh_schedule(): Scheduling...\n");
234 
235 	tbl = (struct ip_vs_sh_bucket *)svc->sched_data;
236 	dest = ip_vs_sh_get(svc->af, tbl, &iph.saddr);
237 	if (!dest
238 	    || !(dest->flags & IP_VS_DEST_F_AVAILABLE)
239 	    || atomic_read(&dest->weight) <= 0
240 	    || is_overloaded(dest)) {
241 		ip_vs_scheduler_err(svc, "no destination available");
242 		return NULL;
243 	}
244 
245 	IP_VS_DBG_BUF(6, "SH: source IP address %s --> server %s:%d\n",
246 		      IP_VS_DBG_ADDR(svc->af, &iph.saddr),
247 		      IP_VS_DBG_ADDR(svc->af, &dest->addr),
248 		      ntohs(dest->port));
249 
250 	return dest;
251 }
252 
253 
254 /*
255  *      IPVS SH Scheduler structure
256  */
257 static struct ip_vs_scheduler ip_vs_sh_scheduler =
258 {
259 	.name =			"sh",
260 	.refcnt =		ATOMIC_INIT(0),
261 	.module =		THIS_MODULE,
262 	.n_list	 =		LIST_HEAD_INIT(ip_vs_sh_scheduler.n_list),
263 	.init_service =		ip_vs_sh_init_svc,
264 	.done_service =		ip_vs_sh_done_svc,
265 	.update_service =	ip_vs_sh_update_svc,
266 	.schedule =		ip_vs_sh_schedule,
267 };
268 
269 
270 static int __init ip_vs_sh_init(void)
271 {
272 	return register_ip_vs_scheduler(&ip_vs_sh_scheduler);
273 }
274 
275 
276 static void __exit ip_vs_sh_cleanup(void)
277 {
278 	unregister_ip_vs_scheduler(&ip_vs_sh_scheduler);
279 }
280 
281 
282 module_init(ip_vs_sh_init);
283 module_exit(ip_vs_sh_cleanup);
284 MODULE_LICENSE("GPL");
285