1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * net/sched/gen_estimator.c Simple rate estimator.
4 *
5 * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
6 * Eric Dumazet <edumazet@google.com>
7 *
8 * Changes:
9 * Jamal Hadi Salim - moved it to net/core and reshulfed
10 * names to make it usable in general net subsystem.
11 */
12
13 #include <linux/uaccess.h>
14 #include <linux/bitops.h>
15 #include <linux/module.h>
16 #include <linux/types.h>
17 #include <linux/kernel.h>
18 #include <linux/jiffies.h>
19 #include <linux/string.h>
20 #include <linux/mm.h>
21 #include <linux/socket.h>
22 #include <linux/sockios.h>
23 #include <linux/in.h>
24 #include <linux/errno.h>
25 #include <linux/interrupt.h>
26 #include <linux/netdevice.h>
27 #include <linux/skbuff.h>
28 #include <linux/rtnetlink.h>
29 #include <linux/init.h>
30 #include <linux/slab.h>
31 #include <linux/seqlock.h>
32 #include <net/sock.h>
33 #include <net/gen_stats.h>
34
35 /* This code is NOT intended to be used for statistics collection,
36 * its purpose is to provide a base for statistical multiplexing
37 * for controlled load service.
38 * If you need only statistics, run a user level daemon which
39 * periodically reads byte counters.
40 */
41
42 struct net_rate_estimator {
43 struct gnet_stats_basic_packed *bstats;
44 spinlock_t *stats_lock;
45 seqcount_t *running;
46 struct gnet_stats_basic_cpu __percpu *cpu_bstats;
47 u8 ewma_log;
48 u8 intvl_log; /* period : (250ms << intvl_log) */
49
50 seqcount_t seq;
51 u64 last_packets;
52 u64 last_bytes;
53
54 u64 avpps;
55 u64 avbps;
56
57 unsigned long next_jiffies;
58 struct timer_list timer;
59 struct rcu_head rcu;
60 };
61
est_fetch_counters(struct net_rate_estimator * e,struct gnet_stats_basic_packed * b)62 static void est_fetch_counters(struct net_rate_estimator *e,
63 struct gnet_stats_basic_packed *b)
64 {
65 memset(b, 0, sizeof(*b));
66 if (e->stats_lock)
67 spin_lock(e->stats_lock);
68
69 __gnet_stats_copy_basic(e->running, b, e->cpu_bstats, e->bstats);
70
71 if (e->stats_lock)
72 spin_unlock(e->stats_lock);
73
74 }
75
est_timer(struct timer_list * t)76 static void est_timer(struct timer_list *t)
77 {
78 struct net_rate_estimator *est = from_timer(est, t, timer);
79 struct gnet_stats_basic_packed b;
80 u64 rate, brate;
81
82 est_fetch_counters(est, &b);
83 brate = (b.bytes - est->last_bytes) << (10 - est->intvl_log);
84 brate = (brate >> est->ewma_log) - (est->avbps >> est->ewma_log);
85
86 rate = (b.packets - est->last_packets) << (10 - est->intvl_log);
87 rate = (rate >> est->ewma_log) - (est->avpps >> est->ewma_log);
88
89 write_seqcount_begin(&est->seq);
90 est->avbps += brate;
91 est->avpps += rate;
92 write_seqcount_end(&est->seq);
93
94 est->last_bytes = b.bytes;
95 est->last_packets = b.packets;
96
97 est->next_jiffies += ((HZ/4) << est->intvl_log);
98
99 if (unlikely(time_after_eq(jiffies, est->next_jiffies))) {
100 /* Ouch... timer was delayed. */
101 est->next_jiffies = jiffies + 1;
102 }
103 mod_timer(&est->timer, est->next_jiffies);
104 }
105
106 /**
107 * gen_new_estimator - create a new rate estimator
108 * @bstats: basic statistics
109 * @cpu_bstats: bstats per cpu
110 * @rate_est: rate estimator statistics
111 * @lock: lock for statistics and control path
112 * @running: qdisc running seqcount
113 * @opt: rate estimator configuration TLV
114 *
115 * Creates a new rate estimator with &bstats as source and &rate_est
116 * as destination. A new timer with the interval specified in the
117 * configuration TLV is created. Upon each interval, the latest statistics
118 * will be read from &bstats and the estimated rate will be stored in
119 * &rate_est with the statistics lock grabbed during this period.
120 *
121 * Returns 0 on success or a negative error code.
122 *
123 */
gen_new_estimator(struct gnet_stats_basic_packed * bstats,struct gnet_stats_basic_cpu __percpu * cpu_bstats,struct net_rate_estimator __rcu ** rate_est,spinlock_t * lock,seqcount_t * running,struct nlattr * opt)124 int gen_new_estimator(struct gnet_stats_basic_packed *bstats,
125 struct gnet_stats_basic_cpu __percpu *cpu_bstats,
126 struct net_rate_estimator __rcu **rate_est,
127 spinlock_t *lock,
128 seqcount_t *running,
129 struct nlattr *opt)
130 {
131 struct gnet_estimator *parm = nla_data(opt);
132 struct net_rate_estimator *old, *est;
133 struct gnet_stats_basic_packed b;
134 int intvl_log;
135
136 if (nla_len(opt) < sizeof(*parm))
137 return -EINVAL;
138
139 /* allowed timer periods are :
140 * -2 : 250ms, -1 : 500ms, 0 : 1 sec
141 * 1 : 2 sec, 2 : 4 sec, 3 : 8 sec
142 */
143 if (parm->interval < -2 || parm->interval > 3)
144 return -EINVAL;
145
146 if (parm->ewma_log == 0 || parm->ewma_log >= 31)
147 return -EINVAL;
148
149 est = kzalloc(sizeof(*est), GFP_KERNEL);
150 if (!est)
151 return -ENOBUFS;
152
153 seqcount_init(&est->seq);
154 intvl_log = parm->interval + 2;
155 est->bstats = bstats;
156 est->stats_lock = lock;
157 est->running = running;
158 est->ewma_log = parm->ewma_log;
159 est->intvl_log = intvl_log;
160 est->cpu_bstats = cpu_bstats;
161
162 if (lock)
163 local_bh_disable();
164 est_fetch_counters(est, &b);
165 if (lock)
166 local_bh_enable();
167 est->last_bytes = b.bytes;
168 est->last_packets = b.packets;
169
170 if (lock)
171 spin_lock_bh(lock);
172 old = rcu_dereference_protected(*rate_est, 1);
173 if (old) {
174 del_timer_sync(&old->timer);
175 est->avbps = old->avbps;
176 est->avpps = old->avpps;
177 }
178
179 est->next_jiffies = jiffies + ((HZ/4) << intvl_log);
180 timer_setup(&est->timer, est_timer, 0);
181 mod_timer(&est->timer, est->next_jiffies);
182
183 rcu_assign_pointer(*rate_est, est);
184 if (lock)
185 spin_unlock_bh(lock);
186 if (old)
187 kfree_rcu(old, rcu);
188 return 0;
189 }
190 EXPORT_SYMBOL(gen_new_estimator);
191
192 /**
193 * gen_kill_estimator - remove a rate estimator
194 * @rate_est: rate estimator
195 *
196 * Removes the rate estimator.
197 *
198 */
gen_kill_estimator(struct net_rate_estimator __rcu ** rate_est)199 void gen_kill_estimator(struct net_rate_estimator __rcu **rate_est)
200 {
201 struct net_rate_estimator *est;
202
203 est = xchg((__force struct net_rate_estimator **)rate_est, NULL);
204 if (est) {
205 del_timer_sync(&est->timer);
206 kfree_rcu(est, rcu);
207 }
208 }
209 EXPORT_SYMBOL(gen_kill_estimator);
210
211 /**
212 * gen_replace_estimator - replace rate estimator configuration
213 * @bstats: basic statistics
214 * @cpu_bstats: bstats per cpu
215 * @rate_est: rate estimator statistics
216 * @lock: lock for statistics and control path
217 * @running: qdisc running seqcount (might be NULL)
218 * @opt: rate estimator configuration TLV
219 *
220 * Replaces the configuration of a rate estimator by calling
221 * gen_kill_estimator() and gen_new_estimator().
222 *
223 * Returns 0 on success or a negative error code.
224 */
gen_replace_estimator(struct gnet_stats_basic_packed * bstats,struct gnet_stats_basic_cpu __percpu * cpu_bstats,struct net_rate_estimator __rcu ** rate_est,spinlock_t * lock,seqcount_t * running,struct nlattr * opt)225 int gen_replace_estimator(struct gnet_stats_basic_packed *bstats,
226 struct gnet_stats_basic_cpu __percpu *cpu_bstats,
227 struct net_rate_estimator __rcu **rate_est,
228 spinlock_t *lock,
229 seqcount_t *running, struct nlattr *opt)
230 {
231 return gen_new_estimator(bstats, cpu_bstats, rate_est,
232 lock, running, opt);
233 }
234 EXPORT_SYMBOL(gen_replace_estimator);
235
236 /**
237 * gen_estimator_active - test if estimator is currently in use
238 * @rate_est: rate estimator
239 *
240 * Returns true if estimator is active, and false if not.
241 */
gen_estimator_active(struct net_rate_estimator __rcu ** rate_est)242 bool gen_estimator_active(struct net_rate_estimator __rcu **rate_est)
243 {
244 return !!rcu_access_pointer(*rate_est);
245 }
246 EXPORT_SYMBOL(gen_estimator_active);
247
gen_estimator_read(struct net_rate_estimator __rcu ** rate_est,struct gnet_stats_rate_est64 * sample)248 bool gen_estimator_read(struct net_rate_estimator __rcu **rate_est,
249 struct gnet_stats_rate_est64 *sample)
250 {
251 struct net_rate_estimator *est;
252 unsigned seq;
253
254 rcu_read_lock();
255 est = rcu_dereference(*rate_est);
256 if (!est) {
257 rcu_read_unlock();
258 return false;
259 }
260
261 do {
262 seq = read_seqcount_begin(&est->seq);
263 sample->bps = est->avbps >> 8;
264 sample->pps = est->avpps >> 8;
265 } while (read_seqcount_retry(&est->seq, seq));
266
267 rcu_read_unlock();
268 return true;
269 }
270 EXPORT_SYMBOL(gen_estimator_read);
271
