xref: /linux/net/netfilter/ipvs/ip_vs_dh.c (revision 564eb714f5f09ac733c26860d5f0831f213fbdf1)
1 /*
2  * IPVS:        Destination Hashing scheduling module
3  *
4  * Authors:     Wensong Zhang <wensong@gnuchina.org>
5  *
6  *              Inspired by the consistent hashing scheduler patch from
7  *              Thomas Proell <proellt@gmx.de>
8  *
9  *              This program is free software; you can redistribute it and/or
10  *              modify it under the terms of the GNU General Public License
11  *              as published by the Free Software Foundation; either version
12  *              2 of the License, or (at your option) any later version.
13  *
14  * Changes:
15  *
16  */
17 
18 /*
19  * The dh algorithm is to select server by the hash key of destination IP
20  * address. The pseudo code is as follows:
21  *
22  *       n <- servernode[dest_ip];
23  *       if (n is dead) OR
24  *          (n is overloaded) OR (n.weight <= 0) then
25  *                 return NULL;
26  *
27  *       return n;
28  *
29  * Notes that servernode is a 256-bucket hash table that maps the hash
30  * index derived from packet destination IP address to the current server
31  * array. If the dh scheduler is used in cache cluster, it is good to
32  * combine it with cache_bypass feature. When the statically assigned
33  * server is dead or overloaded, the load balancer can bypass the cache
34  * server and send requests to the original server directly.
35  *
36  */
37 
38 #define KMSG_COMPONENT "IPVS"
39 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
40 
41 #include <linux/ip.h>
42 #include <linux/slab.h>
43 #include <linux/module.h>
44 #include <linux/kernel.h>
45 #include <linux/skbuff.h>
46 
47 #include <net/ip_vs.h>
48 
49 
50 /*
51  *      IPVS DH bucket
52  */
53 struct ip_vs_dh_bucket {
54 	struct ip_vs_dest __rcu	*dest;	/* real server (cache) */
55 };
56 
57 /*
58  *     for IPVS DH entry hash table
59  */
60 #ifndef CONFIG_IP_VS_DH_TAB_BITS
61 #define CONFIG_IP_VS_DH_TAB_BITS        8
62 #endif
63 #define IP_VS_DH_TAB_BITS               CONFIG_IP_VS_DH_TAB_BITS
64 #define IP_VS_DH_TAB_SIZE               (1 << IP_VS_DH_TAB_BITS)
65 #define IP_VS_DH_TAB_MASK               (IP_VS_DH_TAB_SIZE - 1)
66 
67 struct ip_vs_dh_state {
68 	struct ip_vs_dh_bucket		buckets[IP_VS_DH_TAB_SIZE];
69 	struct rcu_head			rcu_head;
70 };
71 
72 /*
73  *	Returns hash value for IPVS DH entry
74  */
75 static inline unsigned int ip_vs_dh_hashkey(int af, const union nf_inet_addr *addr)
76 {
77 	__be32 addr_fold = addr->ip;
78 
79 #ifdef CONFIG_IP_VS_IPV6
80 	if (af == AF_INET6)
81 		addr_fold = addr->ip6[0]^addr->ip6[1]^
82 			    addr->ip6[2]^addr->ip6[3];
83 #endif
84 	return (ntohl(addr_fold)*2654435761UL) & IP_VS_DH_TAB_MASK;
85 }
86 
87 
88 /*
89  *      Get ip_vs_dest associated with supplied parameters.
90  */
91 static inline struct ip_vs_dest *
92 ip_vs_dh_get(int af, struct ip_vs_dh_state *s, const union nf_inet_addr *addr)
93 {
94 	return rcu_dereference(s->buckets[ip_vs_dh_hashkey(af, addr)].dest);
95 }
96 
97 
98 /*
99  *      Assign all the hash buckets of the specified table with the service.
100  */
101 static int
102 ip_vs_dh_reassign(struct ip_vs_dh_state *s, struct ip_vs_service *svc)
103 {
104 	int i;
105 	struct ip_vs_dh_bucket *b;
106 	struct list_head *p;
107 	struct ip_vs_dest *dest;
108 	bool empty;
109 
110 	b = &s->buckets[0];
111 	p = &svc->destinations;
112 	empty = list_empty(p);
113 	for (i=0; i<IP_VS_DH_TAB_SIZE; i++) {
114 		dest = rcu_dereference_protected(b->dest, 1);
115 		if (dest)
116 			ip_vs_dest_put(dest);
117 		if (empty)
118 			RCU_INIT_POINTER(b->dest, NULL);
119 		else {
120 			if (p == &svc->destinations)
121 				p = p->next;
122 
123 			dest = list_entry(p, struct ip_vs_dest, n_list);
124 			ip_vs_dest_hold(dest);
125 			RCU_INIT_POINTER(b->dest, dest);
126 
127 			p = p->next;
128 		}
129 		b++;
130 	}
131 	return 0;
132 }
133 
134 
135 /*
136  *      Flush all the hash buckets of the specified table.
137  */
138 static void ip_vs_dh_flush(struct ip_vs_dh_state *s)
139 {
140 	int i;
141 	struct ip_vs_dh_bucket *b;
142 	struct ip_vs_dest *dest;
143 
144 	b = &s->buckets[0];
145 	for (i=0; i<IP_VS_DH_TAB_SIZE; i++) {
146 		dest = rcu_dereference_protected(b->dest, 1);
147 		if (dest) {
148 			ip_vs_dest_put(dest);
149 			RCU_INIT_POINTER(b->dest, NULL);
150 		}
151 		b++;
152 	}
153 }
154 
155 
156 static int ip_vs_dh_init_svc(struct ip_vs_service *svc)
157 {
158 	struct ip_vs_dh_state *s;
159 
160 	/* allocate the DH table for this service */
161 	s = kzalloc(sizeof(struct ip_vs_dh_state), GFP_KERNEL);
162 	if (s == NULL)
163 		return -ENOMEM;
164 
165 	svc->sched_data = s;
166 	IP_VS_DBG(6, "DH hash table (memory=%Zdbytes) allocated for "
167 		  "current service\n",
168 		  sizeof(struct ip_vs_dh_bucket)*IP_VS_DH_TAB_SIZE);
169 
170 	/* assign the hash buckets with current dests */
171 	ip_vs_dh_reassign(s, svc);
172 
173 	return 0;
174 }
175 
176 
177 static void ip_vs_dh_done_svc(struct ip_vs_service *svc)
178 {
179 	struct ip_vs_dh_state *s = svc->sched_data;
180 
181 	/* got to clean up hash buckets here */
182 	ip_vs_dh_flush(s);
183 
184 	/* release the table itself */
185 	kfree_rcu(s, rcu_head);
186 	IP_VS_DBG(6, "DH hash table (memory=%Zdbytes) released\n",
187 		  sizeof(struct ip_vs_dh_bucket)*IP_VS_DH_TAB_SIZE);
188 }
189 
190 
191 static int ip_vs_dh_dest_changed(struct ip_vs_service *svc,
192 				 struct ip_vs_dest *dest)
193 {
194 	struct ip_vs_dh_state *s = svc->sched_data;
195 
196 	/* assign the hash buckets with the updated service */
197 	ip_vs_dh_reassign(s, svc);
198 
199 	return 0;
200 }
201 
202 
203 /*
204  *      If the dest flags is set with IP_VS_DEST_F_OVERLOAD,
205  *      consider that the server is overloaded here.
206  */
207 static inline int is_overloaded(struct ip_vs_dest *dest)
208 {
209 	return dest->flags & IP_VS_DEST_F_OVERLOAD;
210 }
211 
212 
213 /*
214  *      Destination hashing scheduling
215  */
216 static struct ip_vs_dest *
217 ip_vs_dh_schedule(struct ip_vs_service *svc, const struct sk_buff *skb,
218 		  struct ip_vs_iphdr *iph)
219 {
220 	struct ip_vs_dest *dest;
221 	struct ip_vs_dh_state *s;
222 
223 	IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
224 
225 	s = (struct ip_vs_dh_state *) svc->sched_data;
226 	dest = ip_vs_dh_get(svc->af, s, &iph->daddr);
227 	if (!dest
228 	    || !(dest->flags & IP_VS_DEST_F_AVAILABLE)
229 	    || atomic_read(&dest->weight) <= 0
230 	    || is_overloaded(dest)) {
231 		ip_vs_scheduler_err(svc, "no destination available");
232 		return NULL;
233 	}
234 
235 	IP_VS_DBG_BUF(6, "DH: destination IP address %s --> server %s:%d\n",
236 		      IP_VS_DBG_ADDR(svc->af, &iph->daddr),
237 		      IP_VS_DBG_ADDR(svc->af, &dest->addr),
238 		      ntohs(dest->port));
239 
240 	return dest;
241 }
242 
243 
244 /*
245  *      IPVS DH Scheduler structure
246  */
247 static struct ip_vs_scheduler ip_vs_dh_scheduler =
248 {
249 	.name =			"dh",
250 	.refcnt =		ATOMIC_INIT(0),
251 	.module =		THIS_MODULE,
252 	.n_list =		LIST_HEAD_INIT(ip_vs_dh_scheduler.n_list),
253 	.init_service =		ip_vs_dh_init_svc,
254 	.done_service =		ip_vs_dh_done_svc,
255 	.add_dest =		ip_vs_dh_dest_changed,
256 	.del_dest =		ip_vs_dh_dest_changed,
257 	.schedule =		ip_vs_dh_schedule,
258 };
259 
260 
261 static int __init ip_vs_dh_init(void)
262 {
263 	return register_ip_vs_scheduler(&ip_vs_dh_scheduler);
264 }
265 
266 
267 static void __exit ip_vs_dh_cleanup(void)
268 {
269 	unregister_ip_vs_scheduler(&ip_vs_dh_scheduler);
270 	synchronize_rcu();
271 }
272 
273 
274 module_init(ip_vs_dh_init);
275 module_exit(ip_vs_dh_cleanup);
276 MODULE_LICENSE("GPL");
277