Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * IPVS: Weighted Round-Robin Scheduling module |
| 3 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4 | * Authors: Wensong Zhang <wensong@linuxvirtualserver.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 | * Wensong Zhang : changed the ip_vs_wrr_schedule to return dest |
| 13 | * Wensong Zhang : changed some comestics things for debugging |
| 14 | * Wensong Zhang : changed for the d-linked destination list |
| 15 | * Wensong Zhang : added the ip_vs_wrr_update_svc |
| 16 | * Julian Anastasov : fixed the bug of returning destination |
| 17 | * with weight 0 when all weights are zero |
| 18 | * |
| 19 | */ |
| 20 | |
Hannes Eder | 9aada7a | 2009-07-30 14:29:44 -0700 | [diff] [blame] | 21 | #define KMSG_COMPONENT "IPVS" |
| 22 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt |
| 23 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include <linux/module.h> |
| 25 | #include <linux/kernel.h> |
Sven Wegener | 9c1ca6e | 2008-02-05 20:00:10 -0800 | [diff] [blame] | 26 | #include <linux/net.h> |
Florian Fainelli | ae24e57 | 2009-12-22 09:42:06 +0100 | [diff] [blame] | 27 | #include <linux/gcd.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | |
| 29 | #include <net/ip_vs.h> |
| 30 | |
| 31 | /* |
| 32 | * current destination pointer for weighted round-robin scheduling |
| 33 | */ |
| 34 | struct ip_vs_wrr_mark { |
| 35 | struct list_head *cl; /* current list head */ |
| 36 | int cw; /* current weight */ |
| 37 | int mw; /* maximum weight */ |
| 38 | int di; /* decreasing interval */ |
| 39 | }; |
| 40 | |
| 41 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 42 | static int ip_vs_wrr_gcd_weight(struct ip_vs_service *svc) |
| 43 | { |
| 44 | struct ip_vs_dest *dest; |
| 45 | int weight; |
| 46 | int g = 0; |
| 47 | |
| 48 | list_for_each_entry(dest, &svc->destinations, n_list) { |
| 49 | weight = atomic_read(&dest->weight); |
| 50 | if (weight > 0) { |
| 51 | if (g > 0) |
| 52 | g = gcd(weight, g); |
| 53 | else |
| 54 | g = weight; |
| 55 | } |
| 56 | } |
| 57 | return g ? g : 1; |
| 58 | } |
| 59 | |
| 60 | |
| 61 | /* |
| 62 | * Get the maximum weight of the service destinations. |
| 63 | */ |
| 64 | static int ip_vs_wrr_max_weight(struct ip_vs_service *svc) |
| 65 | { |
| 66 | struct ip_vs_dest *dest; |
Simon Horman | 1e66daf | 2009-08-31 14:18:48 +0200 | [diff] [blame] | 67 | int new_weight, weight = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 68 | |
| 69 | list_for_each_entry(dest, &svc->destinations, n_list) { |
Simon Horman | 1e66daf | 2009-08-31 14:18:48 +0200 | [diff] [blame] | 70 | new_weight = atomic_read(&dest->weight); |
| 71 | if (new_weight > weight) |
| 72 | weight = new_weight; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 73 | } |
| 74 | |
| 75 | return weight; |
| 76 | } |
| 77 | |
| 78 | |
| 79 | static int ip_vs_wrr_init_svc(struct ip_vs_service *svc) |
| 80 | { |
| 81 | struct ip_vs_wrr_mark *mark; |
| 82 | |
| 83 | /* |
| 84 | * Allocate the mark variable for WRR scheduling |
| 85 | */ |
| 86 | mark = kmalloc(sizeof(struct ip_vs_wrr_mark), GFP_ATOMIC); |
| 87 | if (mark == NULL) { |
Hannes Eder | 1e3e238 | 2009-08-02 11:05:41 +0000 | [diff] [blame] | 88 | pr_err("%s(): no memory\n", __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | return -ENOMEM; |
| 90 | } |
| 91 | mark->cl = &svc->destinations; |
| 92 | mark->cw = 0; |
| 93 | mark->mw = ip_vs_wrr_max_weight(svc); |
| 94 | mark->di = ip_vs_wrr_gcd_weight(svc); |
| 95 | svc->sched_data = mark; |
| 96 | |
| 97 | return 0; |
| 98 | } |
| 99 | |
| 100 | |
| 101 | static int ip_vs_wrr_done_svc(struct ip_vs_service *svc) |
| 102 | { |
| 103 | /* |
| 104 | * Release the mark variable |
| 105 | */ |
| 106 | kfree(svc->sched_data); |
| 107 | |
| 108 | return 0; |
| 109 | } |
| 110 | |
| 111 | |
| 112 | static int ip_vs_wrr_update_svc(struct ip_vs_service *svc) |
| 113 | { |
| 114 | struct ip_vs_wrr_mark *mark = svc->sched_data; |
| 115 | |
| 116 | mark->cl = &svc->destinations; |
| 117 | mark->mw = ip_vs_wrr_max_weight(svc); |
| 118 | mark->di = ip_vs_wrr_gcd_weight(svc); |
| 119 | if (mark->cw > mark->mw) |
| 120 | mark->cw = 0; |
| 121 | return 0; |
| 122 | } |
| 123 | |
| 124 | |
| 125 | /* |
| 126 | * Weighted Round-Robin Scheduling |
| 127 | */ |
| 128 | static struct ip_vs_dest * |
| 129 | ip_vs_wrr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb) |
| 130 | { |
| 131 | struct ip_vs_dest *dest; |
| 132 | struct ip_vs_wrr_mark *mark = svc->sched_data; |
| 133 | struct list_head *p; |
| 134 | |
Hannes Eder | 1e3e238 | 2009-08-02 11:05:41 +0000 | [diff] [blame] | 135 | IP_VS_DBG(6, "%s(): Scheduling...\n", __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 136 | |
| 137 | /* |
| 138 | * This loop will always terminate, because mark->cw in (0, max_weight] |
| 139 | * and at least one server has its weight equal to max_weight. |
| 140 | */ |
| 141 | write_lock(&svc->sched_lock); |
| 142 | p = mark->cl; |
| 143 | while (1) { |
| 144 | if (mark->cl == &svc->destinations) { |
| 145 | /* it is at the head of the destination list */ |
| 146 | |
| 147 | if (mark->cl == mark->cl->next) { |
| 148 | /* no dest entry */ |
Simon Horman | 68888d1 | 2008-12-29 18:37:36 -0800 | [diff] [blame] | 149 | IP_VS_ERR_RL("WRR: no destination available: " |
| 150 | "no destinations present\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 151 | dest = NULL; |
| 152 | goto out; |
| 153 | } |
| 154 | |
| 155 | mark->cl = svc->destinations.next; |
| 156 | mark->cw -= mark->di; |
| 157 | if (mark->cw <= 0) { |
| 158 | mark->cw = mark->mw; |
| 159 | /* |
| 160 | * Still zero, which means no available servers. |
| 161 | */ |
| 162 | if (mark->cw == 0) { |
| 163 | mark->cl = &svc->destinations; |
Simon Horman | 68888d1 | 2008-12-29 18:37:36 -0800 | [diff] [blame] | 164 | IP_VS_ERR_RL("WRR: no destination " |
| 165 | "available\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 166 | dest = NULL; |
| 167 | goto out; |
| 168 | } |
| 169 | } |
| 170 | } else |
| 171 | mark->cl = mark->cl->next; |
| 172 | |
| 173 | if (mark->cl != &svc->destinations) { |
| 174 | /* not at the head of the list */ |
| 175 | dest = list_entry(mark->cl, struct ip_vs_dest, n_list); |
| 176 | if (!(dest->flags & IP_VS_DEST_F_OVERLOAD) && |
| 177 | atomic_read(&dest->weight) >= mark->cw) { |
| 178 | /* got it */ |
| 179 | break; |
| 180 | } |
| 181 | } |
| 182 | |
| 183 | if (mark->cl == p && mark->cw == mark->di) { |
| 184 | /* back to the start, and no dest is found. |
| 185 | It is only possible when all dests are OVERLOADED */ |
| 186 | dest = NULL; |
Simon Horman | 68888d1 | 2008-12-29 18:37:36 -0800 | [diff] [blame] | 187 | IP_VS_ERR_RL("WRR: no destination available: " |
| 188 | "all destinations are overloaded\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 189 | goto out; |
| 190 | } |
| 191 | } |
| 192 | |
Julius Volz | b14198f | 2008-09-02 15:55:39 +0200 | [diff] [blame] | 193 | IP_VS_DBG_BUF(6, "WRR: server %s:%u " |
| 194 | "activeconns %d refcnt %d weight %d\n", |
| 195 | IP_VS_DBG_ADDR(svc->af, &dest->addr), ntohs(dest->port), |
| 196 | atomic_read(&dest->activeconns), |
| 197 | atomic_read(&dest->refcnt), |
| 198 | atomic_read(&dest->weight)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | |
| 200 | out: |
| 201 | write_unlock(&svc->sched_lock); |
| 202 | return dest; |
| 203 | } |
| 204 | |
| 205 | |
| 206 | static struct ip_vs_scheduler ip_vs_wrr_scheduler = { |
| 207 | .name = "wrr", |
| 208 | .refcnt = ATOMIC_INIT(0), |
| 209 | .module = THIS_MODULE, |
Sven Wegener | d149ccc | 2008-08-10 09:18:02 +0000 | [diff] [blame] | 210 | .n_list = LIST_HEAD_INIT(ip_vs_wrr_scheduler.n_list), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 211 | .init_service = ip_vs_wrr_init_svc, |
| 212 | .done_service = ip_vs_wrr_done_svc, |
| 213 | .update_service = ip_vs_wrr_update_svc, |
| 214 | .schedule = ip_vs_wrr_schedule, |
| 215 | }; |
| 216 | |
| 217 | static int __init ip_vs_wrr_init(void) |
| 218 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 219 | return register_ip_vs_scheduler(&ip_vs_wrr_scheduler) ; |
| 220 | } |
| 221 | |
| 222 | static void __exit ip_vs_wrr_cleanup(void) |
| 223 | { |
| 224 | unregister_ip_vs_scheduler(&ip_vs_wrr_scheduler); |
| 225 | } |
| 226 | |
| 227 | module_init(ip_vs_wrr_init); |
| 228 | module_exit(ip_vs_wrr_cleanup); |
| 229 | MODULE_LICENSE("GPL"); |