kernel/2.6.29.6-aldebaran-rt/net/netfilter/ipvs/ip_vs_dh.c

   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 #include <linux/ip.h>
  39 #include <linux/module.h>
  40 #include <linux/kernel.h>
  41 #include <linux/skbuff.h>
  42
  43 #include <net/ip_vs.h>
  44
  45
  46 /*
  47  *      IPVS DH bucket
  48  */
  49 struct ip_vs_dh_bucket {
  50         struct ip_vs_dest       *dest;          /* real server (cache) */
  51 };
  52
  53 /*
  54  *     for IPVS DH entry hash table
  55  */
  56 #ifndef CONFIG_IP_VS_DH_TAB_BITS
  57 #define CONFIG_IP_VS_DH_TAB_BITS        8
  58 #endif
  59 #define IP_VS_DH_TAB_BITS               CONFIG_IP_VS_DH_TAB_BITS
  60 #define IP_VS_DH_TAB_SIZE               (1 << IP_VS_DH_TAB_BITS)
  61 #define IP_VS_DH_TAB_MASK               (IP_VS_DH_TAB_SIZE - 1)
  62
  63
  64 /*
  65  *      Returns hash value for IPVS DH entry
  66  */
  67 static inline unsigned ip_vs_dh_hashkey(int af, const union nf_inet_addr *addr)
  68 {
  69         __be32 addr_fold = addr->ip;
  70
  71 #ifdef CONFIG_IP_VS_IPV6
  72         if (af == AF_INET6)
  73                 addr_fold = addr->ip6[0]^addr->ip6[1]^
  74                             addr->ip6[2]^addr->ip6[3];
  75 #endif
  76         return (ntohl(addr_fold)*2654435761UL) & IP_VS_DH_TAB_MASK;
  77 }
  78
  79
  80 /*
  81  *      Get ip_vs_dest associated with supplied parameters.
  82  */
  83 static inline struct ip_vs_dest *
  84 ip_vs_dh_get(int af, struct ip_vs_dh_bucket *tbl,
  85              const union nf_inet_addr *addr)
  86 {
  87         return (tbl[ip_vs_dh_hashkey(af, addr)]).dest;
  88 }
  89
  90
  91 /*
  92  *      Assign all the hash buckets of the specified table with the service.
  93  */
  94 static int
  95 ip_vs_dh_assign(struct ip_vs_dh_bucket *tbl, struct ip_vs_service *svc)
  96 {
  97         int i;
  98         struct ip_vs_dh_bucket *b;
  99         struct list_head *p;
 100         struct ip_vs_dest *dest;
 101
 102         b = tbl;
 103         p = &svc->destinations;
 104         for (i=0; i<IP_VS_DH_TAB_SIZE; i++) {
 105                 if (list_empty(p)) {
 106                         b->dest = NULL;
 107                 } else {
 108                         if (p == &svc->destinations)
 109                                 p = p->next;
 110
 111                         dest = list_entry(p, struct ip_vs_dest, n_list);
 112                         atomic_inc(&dest->refcnt);
 113                         b->dest = dest;
 114
 115                         p = p->next;
 116                 }
 117                 b++;
 118         }
 119         return 0;
 120 }
 121
 122
 123 /*
 124  *      Flush all the hash buckets of the specified table.
 125  */
 126 static void ip_vs_dh_flush(struct ip_vs_dh_bucket *tbl)
 127 {
 128         int i;
 129         struct ip_vs_dh_bucket *b;
 130
 131         b = tbl;
 132         for (i=0; i<IP_VS_DH_TAB_SIZE; i++) {
 133                 if (b->dest) {
 134                         atomic_dec(&b->dest->refcnt);
 135                         b->dest = NULL;
 136                 }
 137                 b++;
 138         }
 139 }
 140
 141
 142 static int ip_vs_dh_init_svc(struct ip_vs_service *svc)
 143 {
 144         struct ip_vs_dh_bucket *tbl;
 145
 146         /* allocate the DH table for this service */
 147         tbl = kmalloc(sizeof(struct ip_vs_dh_bucket)*IP_VS_DH_TAB_SIZE,
 148                       GFP_ATOMIC);
 149         if (tbl == NULL) {
 150                 IP_VS_ERR("ip_vs_dh_init_svc(): no memory\n");
 151                 return -ENOMEM;
 152         }
 153         svc->sched_data = tbl;
 154         IP_VS_DBG(6, "DH hash table (memory=%Zdbytes) allocated for "
 155                   "current service\n",
 156                   sizeof(struct ip_vs_dh_bucket)*IP_VS_DH_TAB_SIZE);
 157
 158         /* assign the hash buckets with the updated service */
 159         ip_vs_dh_assign(tbl, svc);
 160
 161         return 0;
 162 }
 163
 164
 165 static int ip_vs_dh_done_svc(struct ip_vs_service *svc)
 166 {
 167         struct ip_vs_dh_bucket *tbl = svc->sched_data;
 168
 169         /* got to clean up hash buckets here */
 170         ip_vs_dh_flush(tbl);
 171
 172         /* release the table itself */
 173         kfree(svc->sched_data);
 174         IP_VS_DBG(6, "DH hash table (memory=%Zdbytes) released\n",
 175                   sizeof(struct ip_vs_dh_bucket)*IP_VS_DH_TAB_SIZE);
 176
 177         return 0;
 178 }
 179
 180
 181 static int ip_vs_dh_update_svc(struct ip_vs_service *svc)
 182 {
 183         struct ip_vs_dh_bucket *tbl = svc->sched_data;
 184
 185         /* got to clean up hash buckets here */
 186         ip_vs_dh_flush(tbl);
 187
 188         /* assign the hash buckets with the updated service */
 189         ip_vs_dh_assign(tbl, svc);
 190
 191         return 0;
 192 }
 193
 194
 195 /*
 196  *      If the dest flags is set with IP_VS_DEST_F_OVERLOAD,
 197  *      consider that the server is overloaded here.
 198  */
 199 static inline int is_overloaded(struct ip_vs_dest *dest)
 200 {
 201         return dest->flags & IP_VS_DEST_F_OVERLOAD;
 202 }
 203
 204
 205 /*
 206  *      Destination hashing scheduling
 207  */
 208 static struct ip_vs_dest *
 209 ip_vs_dh_schedule(struct ip_vs_service *svc, const struct sk_buff *skb)
 210 {
 211         struct ip_vs_dest *dest;
 212         struct ip_vs_dh_bucket *tbl;
 213         struct ip_vs_iphdr iph;
 214
 215         ip_vs_fill_iphdr(svc->af, skb_network_header(skb), &iph);
 216
 217         IP_VS_DBG(6, "ip_vs_dh_schedule(): Scheduling...\n");
 218
 219         tbl = (struct ip_vs_dh_bucket *)svc->sched_data;
 220         dest = ip_vs_dh_get(svc->af, tbl, &iph.daddr);
 221         if (!dest
 222             || !(dest->flags & IP_VS_DEST_F_AVAILABLE)
 223             || atomic_read(&dest->weight) <= 0
 224             || is_overloaded(dest)) {
 225                 return NULL;
 226         }
 227
 228         IP_VS_DBG_BUF(6, "DH: destination IP address %s --> server %s:%d\n",
 229                       IP_VS_DBG_ADDR(svc->af, &iph.daddr),
 230                       IP_VS_DBG_ADDR(svc->af, &dest->addr),
 231                       ntohs(dest->port));
 232
 233         return dest;
 234 }
 235
 236
 237 /*
 238  *      IPVS DH Scheduler structure
 239  */
 240 static struct ip_vs_scheduler ip_vs_dh_scheduler =
 241 {
 242         .name =                 "dh",
 243         .refcnt =               ATOMIC_INIT(0),
 244         .module =               THIS_MODULE,
 245         .n_list =               LIST_HEAD_INIT(ip_vs_dh_scheduler.n_list),
 246         .init_service =         ip_vs_dh_init_svc,
 247         .done_service =         ip_vs_dh_done_svc,
 248         .update_service =       ip_vs_dh_update_svc,
 249         .schedule =             ip_vs_dh_schedule,
 250 };
 251
 252
 253 static int __init ip_vs_dh_init(void)
 254 {
 255         return register_ip_vs_scheduler(&ip_vs_dh_scheduler);
 256 }
 257
 258
 259 static void __exit ip_vs_dh_cleanup(void)
 260 {
 261         unregister_ip_vs_scheduler(&ip_vs_dh_scheduler);
 262 }
 263
 264
 265 module_init(ip_vs_dh_init);
 266 module_exit(ip_vs_dh_cleanup);
 267 MODULE_LICENSE("GPL");