Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * IPVS: Weighted Round-Robin Scheduling module |
| 3 | * |
| 4 | * Version: $Id: ip_vs_wrr.c,v 1.12 2002/09/15 08:14:08 wensong Exp $ |
| 5 | * |
| 6 | * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> |
| 7 | * |
| 8 | * This program is free software; you can redistribute it and/or |
| 9 | * modify it under the terms of the GNU General Public License |
| 10 | * as published by the Free Software Foundation; either version |
| 11 | * 2 of the License, or (at your option) any later version. |
| 12 | * |
| 13 | * Changes: |
| 14 | * Wensong Zhang : changed the ip_vs_wrr_schedule to return dest |
| 15 | * Wensong Zhang : changed some comestics things for debugging |
| 16 | * Wensong Zhang : changed for the d-linked destination list |
| 17 | * Wensong Zhang : added the ip_vs_wrr_update_svc |
| 18 | * Julian Anastasov : fixed the bug of returning destination |
| 19 | * with weight 0 when all weights are zero |
| 20 | * |
| 21 | */ |
| 22 | |
| 23 | #include <linux/module.h> |
| 24 | #include <linux/kernel.h> |
| 25 | |
| 26 | #include <net/ip_vs.h> |
| 27 | |
| 28 | /* |
| 29 | * current destination pointer for weighted round-robin scheduling |
| 30 | */ |
| 31 | struct ip_vs_wrr_mark { |
| 32 | struct list_head *cl; /* current list head */ |
| 33 | int cw; /* current weight */ |
| 34 | int mw; /* maximum weight */ |
| 35 | int di; /* decreasing interval */ |
| 36 | }; |
| 37 | |
| 38 | |
| 39 | /* |
| 40 | * Get the gcd of server weights |
| 41 | */ |
| 42 | static int gcd(int a, int b) |
| 43 | { |
| 44 | int c; |
| 45 | |
| 46 | while ((c = a % b)) { |
| 47 | a = b; |
| 48 | b = c; |
| 49 | } |
| 50 | return b; |
| 51 | } |
| 52 | |
| 53 | static int ip_vs_wrr_gcd_weight(struct ip_vs_service *svc) |
| 54 | { |
| 55 | struct ip_vs_dest *dest; |
| 56 | int weight; |
| 57 | int g = 0; |
| 58 | |
| 59 | list_for_each_entry(dest, &svc->destinations, n_list) { |
| 60 | weight = atomic_read(&dest->weight); |
| 61 | if (weight > 0) { |
| 62 | if (g > 0) |
| 63 | g = gcd(weight, g); |
| 64 | else |
| 65 | g = weight; |
| 66 | } |
| 67 | } |
| 68 | return g ? g : 1; |
| 69 | } |
| 70 | |
| 71 | |
| 72 | /* |
| 73 | * Get the maximum weight of the service destinations. |
| 74 | */ |
| 75 | static int ip_vs_wrr_max_weight(struct ip_vs_service *svc) |
| 76 | { |
| 77 | struct ip_vs_dest *dest; |
| 78 | int weight = 0; |
| 79 | |
| 80 | list_for_each_entry(dest, &svc->destinations, n_list) { |
| 81 | if (atomic_read(&dest->weight) > weight) |
| 82 | weight = atomic_read(&dest->weight); |
| 83 | } |
| 84 | |
| 85 | return weight; |
| 86 | } |
| 87 | |
| 88 | |
| 89 | static int ip_vs_wrr_init_svc(struct ip_vs_service *svc) |
| 90 | { |
| 91 | struct ip_vs_wrr_mark *mark; |
| 92 | |
| 93 | /* |
| 94 | * Allocate the mark variable for WRR scheduling |
| 95 | */ |
| 96 | mark = kmalloc(sizeof(struct ip_vs_wrr_mark), GFP_ATOMIC); |
| 97 | if (mark == NULL) { |
| 98 | IP_VS_ERR("ip_vs_wrr_init_svc(): no memory\n"); |
| 99 | return -ENOMEM; |
| 100 | } |
| 101 | mark->cl = &svc->destinations; |
| 102 | mark->cw = 0; |
| 103 | mark->mw = ip_vs_wrr_max_weight(svc); |
| 104 | mark->di = ip_vs_wrr_gcd_weight(svc); |
| 105 | svc->sched_data = mark; |
| 106 | |
| 107 | return 0; |
| 108 | } |
| 109 | |
| 110 | |
| 111 | static int ip_vs_wrr_done_svc(struct ip_vs_service *svc) |
| 112 | { |
| 113 | /* |
| 114 | * Release the mark variable |
| 115 | */ |
| 116 | kfree(svc->sched_data); |
| 117 | |
| 118 | return 0; |
| 119 | } |
| 120 | |
| 121 | |
| 122 | static int ip_vs_wrr_update_svc(struct ip_vs_service *svc) |
| 123 | { |
| 124 | struct ip_vs_wrr_mark *mark = svc->sched_data; |
| 125 | |
| 126 | mark->cl = &svc->destinations; |
| 127 | mark->mw = ip_vs_wrr_max_weight(svc); |
| 128 | mark->di = ip_vs_wrr_gcd_weight(svc); |
| 129 | if (mark->cw > mark->mw) |
| 130 | mark->cw = 0; |
| 131 | return 0; |
| 132 | } |
| 133 | |
| 134 | |
| 135 | /* |
| 136 | * Weighted Round-Robin Scheduling |
| 137 | */ |
| 138 | static struct ip_vs_dest * |
| 139 | ip_vs_wrr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb) |
| 140 | { |
| 141 | struct ip_vs_dest *dest; |
| 142 | struct ip_vs_wrr_mark *mark = svc->sched_data; |
| 143 | struct list_head *p; |
| 144 | |
| 145 | IP_VS_DBG(6, "ip_vs_wrr_schedule(): Scheduling...\n"); |
| 146 | |
| 147 | /* |
| 148 | * This loop will always terminate, because mark->cw in (0, max_weight] |
| 149 | * and at least one server has its weight equal to max_weight. |
| 150 | */ |
| 151 | write_lock(&svc->sched_lock); |
| 152 | p = mark->cl; |
| 153 | while (1) { |
| 154 | if (mark->cl == &svc->destinations) { |
| 155 | /* it is at the head of the destination list */ |
| 156 | |
| 157 | if (mark->cl == mark->cl->next) { |
| 158 | /* no dest entry */ |
| 159 | dest = NULL; |
| 160 | goto out; |
| 161 | } |
| 162 | |
| 163 | mark->cl = svc->destinations.next; |
| 164 | mark->cw -= mark->di; |
| 165 | if (mark->cw <= 0) { |
| 166 | mark->cw = mark->mw; |
| 167 | /* |
| 168 | * Still zero, which means no available servers. |
| 169 | */ |
| 170 | if (mark->cw == 0) { |
| 171 | mark->cl = &svc->destinations; |
| 172 | IP_VS_INFO("ip_vs_wrr_schedule(): " |
| 173 | "no available servers\n"); |
| 174 | dest = NULL; |
| 175 | goto out; |
| 176 | } |
| 177 | } |
| 178 | } else |
| 179 | mark->cl = mark->cl->next; |
| 180 | |
| 181 | if (mark->cl != &svc->destinations) { |
| 182 | /* not at the head of the list */ |
| 183 | dest = list_entry(mark->cl, struct ip_vs_dest, n_list); |
| 184 | if (!(dest->flags & IP_VS_DEST_F_OVERLOAD) && |
| 185 | atomic_read(&dest->weight) >= mark->cw) { |
| 186 | /* got it */ |
| 187 | break; |
| 188 | } |
| 189 | } |
| 190 | |
| 191 | if (mark->cl == p && mark->cw == mark->di) { |
| 192 | /* back to the start, and no dest is found. |
| 193 | It is only possible when all dests are OVERLOADED */ |
| 194 | dest = NULL; |
| 195 | goto out; |
| 196 | } |
| 197 | } |
| 198 | |
| 199 | IP_VS_DBG(6, "WRR: server %u.%u.%u.%u:%u " |
| 200 | "activeconns %d refcnt %d weight %d\n", |
| 201 | NIPQUAD(dest->addr), ntohs(dest->port), |
| 202 | atomic_read(&dest->activeconns), |
| 203 | atomic_read(&dest->refcnt), |
| 204 | atomic_read(&dest->weight)); |
| 205 | |
| 206 | out: |
| 207 | write_unlock(&svc->sched_lock); |
| 208 | return dest; |
| 209 | } |
| 210 | |
| 211 | |
| 212 | static struct ip_vs_scheduler ip_vs_wrr_scheduler = { |
| 213 | .name = "wrr", |
| 214 | .refcnt = ATOMIC_INIT(0), |
| 215 | .module = THIS_MODULE, |
| 216 | .init_service = ip_vs_wrr_init_svc, |
| 217 | .done_service = ip_vs_wrr_done_svc, |
| 218 | .update_service = ip_vs_wrr_update_svc, |
| 219 | .schedule = ip_vs_wrr_schedule, |
| 220 | }; |
| 221 | |
| 222 | static int __init ip_vs_wrr_init(void) |
| 223 | { |
| 224 | INIT_LIST_HEAD(&ip_vs_wrr_scheduler.n_list); |
| 225 | return register_ip_vs_scheduler(&ip_vs_wrr_scheduler) ; |
| 226 | } |
| 227 | |
| 228 | static void __exit ip_vs_wrr_cleanup(void) |
| 229 | { |
| 230 | unregister_ip_vs_scheduler(&ip_vs_wrr_scheduler); |
| 231 | } |
| 232 | |
| 233 | module_init(ip_vs_wrr_init); |
| 234 | module_exit(ip_vs_wrr_cleanup); |
| 235 | MODULE_LICENSE("GPL"); |