/net/netfilter/ipvs/ip_vs_wrr.c

  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/*
  3 * IPVS:        Weighted Round-Robin Scheduling module
  4 *
  5 * Authors:     Wensong Zhang <wensong@linuxvirtualserver.org>
  6 *
  7 * Changes:
  8 *     Wensong Zhang            :     changed the ip_vs_wrr_schedule to return dest
  9 *     Wensong Zhang            :     changed some comestics things for debugging
 10 *     Wensong Zhang            :     changed for the d-linked destination list
 11 *     Wensong Zhang            :     added the ip_vs_wrr_update_svc
 12 *     Julian Anastasov         :     fixed the bug of returning destination
 13 *                                    with weight 0 when all weights are zero
 14 */
 15
 16#define KMSG_COMPONENT "IPVS"
 17#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
 18
 19#include <linux/module.h>
 20#include <linux/kernel.h>
 21#include <linux/slab.h>
 22#include <linux/net.h>
 23#include <linux/gcd.h>
 24
 25#include <net/ip_vs.h>
 26
 27/* The WRR algorithm depends on some caclulations:
 28 * - mw: maximum weight
 29 * - di: weight step, greatest common divisor from all weights
 30 * - cw: current required weight
 31 * As result, all weights are in the [di..mw] range with a step=di.
 32 *
 33 * First, we start with cw = mw and select dests with weight >= cw.
 34 * Then cw is reduced with di and all dests are checked again.
 35 * Last pass should be with cw = di. We have mw/di passes in total:
 36 *
 37 * pass 1: cw = max weight
 38 * pass 2: cw = max weight - di
 39 * pass 3: cw = max weight - 2 * di
 40 * ...
 41 * last pass: cw = di
 42 *
 43 * Weights are supposed to be >= di but we run in parallel with
 44 * weight changes, it is possible some dest weight to be reduced
 45 * below di, bad if it is the only available dest.
 46 *
 47 * So, we modify how mw is calculated, now it is reduced with (di - 1),
 48 * so that last cw is 1 to catch such dests with weight below di:
 49 * pass 1: cw = max weight - (di - 1)
 50 * pass 2: cw = max weight - di - (di - 1)
 51 * pass 3: cw = max weight - 2 * di - (di - 1)
 52 * ...
 53 * last pass: cw = 1
 54 *
 55 */
 56
 57/*
 58 * current destination pointer for weighted round-robin scheduling
 59 */
 60struct ip_vs_wrr_mark {
 61	struct ip_vs_dest *cl;	/* current dest or head */
 62	int cw;			/* current weight */
 63	int mw;			/* maximum weight */
 64	int di;			/* decreasing interval */
 65	struct rcu_head		rcu_head;
 66};
 67
 68
 69static int ip_vs_wrr_gcd_weight(struct ip_vs_service *svc)
 70{
 71	struct ip_vs_dest *dest;
 72	int weight;
 73	int g = 0;
 74
 75	list_for_each_entry(dest, &svc->destinations, n_list) {
 76		weight = atomic_read(&dest->weight);
 77		if (weight > 0) {
 78			if (g > 0)
 79				g = gcd(weight, g);
 80			else
 81				g = weight;
 82		}
 83	}
 84	return g ? g : 1;
 85}
 86
 87
 88/*
 89 *    Get the maximum weight of the service destinations.
 90 */
 91static int ip_vs_wrr_max_weight(struct ip_vs_service *svc)
 92{
 93	struct ip_vs_dest *dest;
 94	int new_weight, weight = 0;
 95
 96	list_for_each_entry(dest, &svc->destinations, n_list) {
 97		new_weight = atomic_read(&dest->weight);
 98		if (new_weight > weight)
 99			weight = new_weight;
100	}
101
102	return weight;
103}
104
105
106static int ip_vs_wrr_init_svc(struct ip_vs_service *svc)
107{
108	struct ip_vs_wrr_mark *mark;
109
110	/*
111	 *    Allocate the mark variable for WRR scheduling
112	 */
113	mark = kmalloc(sizeof(struct ip_vs_wrr_mark), GFP_KERNEL);
114	if (mark == NULL)
115		return -ENOMEM;
116
117	mark->cl = list_entry(&svc->destinations, struct ip_vs_dest, n_list);
118	mark->di = ip_vs_wrr_gcd_weight(svc);
119	mark->mw = ip_vs_wrr_max_weight(svc) - (mark->di - 1);
120	mark->cw = mark->mw;
121	svc->sched_data = mark;
122
123	return 0;
124}
125
126
127static void ip_vs_wrr_done_svc(struct ip_vs_service *svc)
128{
129	struct ip_vs_wrr_mark *mark = svc->sched_data;
130
131	/*
132	 *    Release the mark variable
133	 */
134	kfree_rcu(mark, rcu_head);
135}
136
137
138static int ip_vs_wrr_dest_changed(struct ip_vs_service *svc,
139				  struct ip_vs_dest *dest)
140{
141	struct ip_vs_wrr_mark *mark = svc->sched_data;
142
143	spin_lock_bh(&svc->sched_lock);
144	mark->cl = list_entry(&svc->destinations, struct ip_vs_dest, n_list);
145	mark->di = ip_vs_wrr_gcd_weight(svc);
146	mark->mw = ip_vs_wrr_max_weight(svc) - (mark->di - 1);
147	if (mark->cw > mark->mw || !mark->cw)
148		mark->cw = mark->mw;
149	else if (mark->di > 1)
150		mark->cw = (mark->cw / mark->di) * mark->di + 1;
151	spin_unlock_bh(&svc->sched_lock);
152	return 0;
153}
154
155
156/*
157 *    Weighted Round-Robin Scheduling
158 */
159static struct ip_vs_dest *
160ip_vs_wrr_schedule(struct ip_vs_service *svc, const struct sk_buff *skb,
161		   struct ip_vs_iphdr *iph)
162{
163	struct ip_vs_dest *dest, *last, *stop = NULL;
164	struct ip_vs_wrr_mark *mark = svc->sched_data;
165	bool last_pass = false, restarted = false;
166
167	IP_VS_DBG(6, "%s(): Scheduling...\n", __func__);
168
169	spin_lock_bh(&svc->sched_lock);
170	dest = mark->cl;
171	/* No available dests? */
172	if (mark->mw == 0)
173		goto err_noavail;
174	last = dest;
175	/* Stop only after all dests were checked for weight >= 1 (last pass) */
176	while (1) {
177		list_for_each_entry_continue_rcu(dest,
178						 &svc->destinations,
179						 n_list) {
180			if (!(dest->flags & IP_VS_DEST_F_OVERLOAD) &&
181			    atomic_read(&dest->weight) >= mark->cw)
182				goto found;
183			if (dest == stop)
184				goto err_over;
185		}
186		mark->cw -= mark->di;
187		if (mark->cw <= 0) {
188			mark->cw = mark->mw;
189			/* Stop if we tried last pass from first dest:
190			 * 1. last_pass: we started checks when cw > di but
191			 *	then all dests were checked for w >= 1
192			 * 2. last was head: the first and only traversal
193			 *	was for weight >= 1, for all dests.
194			 */
195			if (last_pass ||
196			    &last->n_list == &svc->destinations)
197				goto err_over;
198			restarted = true;
199		}
200		last_pass = mark->cw <= mark->di;
201		if (last_pass && restarted &&
202		    &last->n_list != &svc->destinations) {
203			/* First traversal was for w >= 1 but only
204			 * for dests after 'last', now do the same
205			 * for all dests up to 'last'.
206			 */
207			stop = last;
208		}
209	}
210
211found:
212	IP_VS_DBG_BUF(6, "WRR: server %s:%u "
213		      "activeconns %d refcnt %d weight %d\n",
214		      IP_VS_DBG_ADDR(dest->af, &dest->addr), ntohs(dest->port),
215		      atomic_read(&dest->activeconns),
216		      refcount_read(&dest->refcnt),
217		      atomic_read(&dest->weight));
218	mark->cl = dest;
219
220  out:
221	spin_unlock_bh(&svc->sched_lock);
222	return dest;
223
224err_noavail:
225	mark->cl = dest;
226	dest = NULL;
227	ip_vs_scheduler_err(svc, "no destination available");
228	goto out;
229
230err_over:
231	mark->cl = dest;
232	dest = NULL;
233	ip_vs_scheduler_err(svc, "no destination available: "
234			    "all destinations are overloaded");
235	goto out;
236}
237
238
239static struct ip_vs_scheduler ip_vs_wrr_scheduler = {
240	.name =			"wrr",
241	.refcnt =		ATOMIC_INIT(0),
242	.module =		THIS_MODULE,
243	.n_list =		LIST_HEAD_INIT(ip_vs_wrr_scheduler.n_list),
244	.init_service =		ip_vs_wrr_init_svc,
245	.done_service =		ip_vs_wrr_done_svc,
246	.add_dest =		ip_vs_wrr_dest_changed,
247	.del_dest =		ip_vs_wrr_dest_changed,
248	.upd_dest =		ip_vs_wrr_dest_changed,
249	.schedule =		ip_vs_wrr_schedule,
250};
251
252static int __init ip_vs_wrr_init(void)
253{
254	return register_ip_vs_scheduler(&ip_vs_wrr_scheduler) ;
255}
256
257static void __exit ip_vs_wrr_cleanup(void)
258{
259	unregister_ip_vs_scheduler(&ip_vs_wrr_scheduler);
260	synchronize_rcu();
261}
262
263module_init(ip_vs_wrr_init);
264module_exit(ip_vs_wrr_cleanup);
265MODULE_LICENSE("GPL");