Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * net/sched/gen_estimator.c Simple rate estimator. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public License | |
6 | * as published by the Free Software Foundation; either version | |
7 | * 2 of the License, or (at your option) any later version. | |
8 | * | |
9 | * Authors: Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru> | |
1c0d32fd | 10 | * Eric Dumazet <edumazet@google.com> |
1da177e4 LT |
11 | * |
12 | * Changes: | |
13 | * Jamal Hadi Salim - moved it to net/core and reshulfed | |
14 | * names to make it usable in general net subsystem. | |
15 | */ | |
16 | ||
7c0f6ba6 | 17 | #include <linux/uaccess.h> |
1977f032 | 18 | #include <linux/bitops.h> |
1da177e4 LT |
19 | #include <linux/module.h> |
20 | #include <linux/types.h> | |
21 | #include <linux/kernel.h> | |
22 | #include <linux/jiffies.h> | |
23 | #include <linux/string.h> | |
24 | #include <linux/mm.h> | |
25 | #include <linux/socket.h> | |
26 | #include <linux/sockios.h> | |
27 | #include <linux/in.h> | |
28 | #include <linux/errno.h> | |
29 | #include <linux/interrupt.h> | |
30 | #include <linux/netdevice.h> | |
31 | #include <linux/skbuff.h> | |
32 | #include <linux/rtnetlink.h> | |
33 | #include <linux/init.h> | |
5a0e3ad6 | 34 | #include <linux/slab.h> |
1c0d32fd | 35 | #include <linux/seqlock.h> |
1da177e4 LT |
36 | #include <net/sock.h> |
37 | #include <net/gen_stats.h> | |
38 | ||
1c0d32fd ED |
39 | /* This code is NOT intended to be used for statistics collection, |
40 | * its purpose is to provide a base for statistical multiplexing | |
41 | * for controlled load service. | |
42 | * If you need only statistics, run a user level daemon which | |
43 | * periodically reads byte counters. | |
1da177e4 LT |
44 | */ |
45 | ||
1c0d32fd | 46 | struct net_rate_estimator { |
c1a8f1f1 | 47 | struct gnet_stats_basic_packed *bstats; |
1da177e4 | 48 | spinlock_t *stats_lock; |
edb09eb1 | 49 | seqcount_t *running; |
1c0d32fd ED |
50 | struct gnet_stats_basic_cpu __percpu *cpu_bstats; |
51 | u8 ewma_log; | |
52 | u8 intvl_log; /* period : (250ms << intvl_log) */ | |
53 | ||
54 | seqcount_t seq; | |
32f675bb | 55 | u32 last_packets; |
1da177e4 | 56 | u64 last_bytes; |
1c0d32fd ED |
57 | |
58 | u64 avpps; | |
511e11e3 | 59 | u64 avbps; |
1da177e4 | 60 | |
1c0d32fd ED |
61 | unsigned long next_jiffies; |
62 | struct timer_list timer; | |
63 | struct rcu_head rcu; | |
1da177e4 LT |
64 | }; |
65 | ||
1c0d32fd ED |
66 | static void est_fetch_counters(struct net_rate_estimator *e, |
67 | struct gnet_stats_basic_packed *b) | |
1da177e4 | 68 | { |
1c0d32fd ED |
69 | if (e->stats_lock) |
70 | spin_lock(e->stats_lock); | |
1da177e4 | 71 | |
1c0d32fd | 72 | __gnet_stats_copy_basic(e->running, b, e->cpu_bstats, e->bstats); |
1da177e4 | 73 | |
1c0d32fd ED |
74 | if (e->stats_lock) |
75 | spin_unlock(e->stats_lock); | |
12efa1fa | 76 | |
1da177e4 LT |
77 | } |
78 | ||
e99e88a9 | 79 | static void est_timer(struct timer_list *t) |
4db0acf3 | 80 | { |
e99e88a9 | 81 | struct net_rate_estimator *est = from_timer(est, t, timer); |
1c0d32fd ED |
82 | struct gnet_stats_basic_packed b; |
83 | u64 rate, brate; | |
4db0acf3 | 84 | |
1c0d32fd | 85 | est_fetch_counters(est, &b); |
ca558e18 | 86 | brate = (b.bytes - est->last_bytes) << (10 - est->ewma_log - est->intvl_log); |
1c0d32fd | 87 | brate -= (est->avbps >> est->ewma_log); |
4db0acf3 | 88 | |
ca558e18 | 89 | rate = (u64)(b.packets - est->last_packets) << (10 - est->ewma_log - est->intvl_log); |
1c0d32fd | 90 | rate -= (est->avpps >> est->ewma_log); |
4db0acf3 | 91 | |
1c0d32fd ED |
92 | write_seqcount_begin(&est->seq); |
93 | est->avbps += brate; | |
94 | est->avpps += rate; | |
95 | write_seqcount_end(&est->seq); | |
4db0acf3 | 96 | |
1c0d32fd ED |
97 | est->last_bytes = b.bytes; |
98 | est->last_packets = b.packets; | |
4db0acf3 | 99 | |
1c0d32fd | 100 | est->next_jiffies += ((HZ/4) << est->intvl_log); |
4db0acf3 | 101 | |
1c0d32fd ED |
102 | if (unlikely(time_after_eq(jiffies, est->next_jiffies))) { |
103 | /* Ouch... timer was delayed. */ | |
104 | est->next_jiffies = jiffies + 1; | |
4db0acf3 | 105 | } |
1c0d32fd | 106 | mod_timer(&est->timer, est->next_jiffies); |
4db0acf3 JP |
107 | } |
108 | ||
1da177e4 LT |
109 | /** |
110 | * gen_new_estimator - create a new rate estimator | |
111 | * @bstats: basic statistics | |
e9fc2f05 | 112 | * @cpu_bstats: bstats per cpu |
1da177e4 LT |
113 | * @rate_est: rate estimator statistics |
114 | * @stats_lock: statistics lock | |
edb09eb1 | 115 | * @running: qdisc running seqcount |
1da177e4 LT |
116 | * @opt: rate estimator configuration TLV |
117 | * | |
118 | * Creates a new rate estimator with &bstats as source and &rate_est | |
119 | * as destination. A new timer with the interval specified in the | |
120 | * configuration TLV is created. Upon each interval, the latest statistics | |
121 | * will be read from &bstats and the estimated rate will be stored in | |
e793c0f7 | 122 | * &rate_est with the statistics lock grabbed during this period. |
4ec93edb | 123 | * |
1da177e4 | 124 | * Returns 0 on success or a negative error code. |
0929c2dd | 125 | * |
1da177e4 | 126 | */ |
c1a8f1f1 | 127 | int gen_new_estimator(struct gnet_stats_basic_packed *bstats, |
22e0f8b9 | 128 | struct gnet_stats_basic_cpu __percpu *cpu_bstats, |
1c0d32fd | 129 | struct net_rate_estimator __rcu **rate_est, |
0929c2dd | 130 | spinlock_t *stats_lock, |
edb09eb1 | 131 | seqcount_t *running, |
1e90474c | 132 | struct nlattr *opt) |
1da177e4 | 133 | { |
1e90474c | 134 | struct gnet_estimator *parm = nla_data(opt); |
1c0d32fd ED |
135 | struct net_rate_estimator *old, *est; |
136 | struct gnet_stats_basic_packed b; | |
137 | int intvl_log; | |
1da177e4 | 138 | |
1e90474c | 139 | if (nla_len(opt) < sizeof(*parm)) |
1da177e4 LT |
140 | return -EINVAL; |
141 | ||
1c0d32fd ED |
142 | /* allowed timer periods are : |
143 | * -2 : 250ms, -1 : 500ms, 0 : 1 sec | |
144 | * 1 : 2 sec, 2 : 4 sec, 3 : 8 sec | |
145 | */ | |
1da177e4 LT |
146 | if (parm->interval < -2 || parm->interval > 3) |
147 | return -EINVAL; | |
148 | ||
77d04bd9 | 149 | est = kzalloc(sizeof(*est), GFP_KERNEL); |
1c0d32fd | 150 | if (!est) |
1da177e4 LT |
151 | return -ENOBUFS; |
152 | ||
1c0d32fd ED |
153 | seqcount_init(&est->seq); |
154 | intvl_log = parm->interval + 2; | |
1da177e4 | 155 | est->bstats = bstats; |
1da177e4 | 156 | est->stats_lock = stats_lock; |
edb09eb1 | 157 | est->running = running; |
1da177e4 | 158 | est->ewma_log = parm->ewma_log; |
1c0d32fd | 159 | est->intvl_log = intvl_log; |
22e0f8b9 | 160 | est->cpu_bstats = cpu_bstats; |
1da177e4 | 161 | |
40ca54e3 ED |
162 | if (stats_lock) |
163 | local_bh_disable(); | |
1c0d32fd | 164 | est_fetch_counters(est, &b); |
40ca54e3 ED |
165 | if (stats_lock) |
166 | local_bh_enable(); | |
1c0d32fd ED |
167 | est->last_bytes = b.bytes; |
168 | est->last_packets = b.packets; | |
169 | old = rcu_dereference_protected(*rate_est, 1); | |
170 | if (old) { | |
171 | del_timer_sync(&old->timer); | |
172 | est->avbps = old->avbps; | |
173 | est->avpps = old->avpps; | |
1da177e4 | 174 | } |
0929c2dd | 175 | |
1c0d32fd | 176 | est->next_jiffies = jiffies + ((HZ/4) << intvl_log); |
e99e88a9 | 177 | timer_setup(&est->timer, est_timer, 0); |
1c0d32fd | 178 | mod_timer(&est->timer, est->next_jiffies); |
4db0acf3 | 179 | |
1c0d32fd ED |
180 | rcu_assign_pointer(*rate_est, est); |
181 | if (old) | |
182 | kfree_rcu(old, rcu); | |
1da177e4 LT |
183 | return 0; |
184 | } | |
c1b56878 | 185 | EXPORT_SYMBOL(gen_new_estimator); |
1da177e4 LT |
186 | |
187 | /** | |
188 | * gen_kill_estimator - remove a rate estimator | |
1c0d32fd | 189 | * @rate_est: rate estimator |
1da177e4 | 190 | * |
1c0d32fd | 191 | * Removes the rate estimator. |
0929c2dd | 192 | * |
1da177e4 | 193 | */ |
1c0d32fd | 194 | void gen_kill_estimator(struct net_rate_estimator __rcu **rate_est) |
1da177e4 | 195 | { |
1c0d32fd | 196 | struct net_rate_estimator *est; |
1da177e4 | 197 | |
1c0d32fd ED |
198 | est = xchg((__force struct net_rate_estimator **)rate_est, NULL); |
199 | if (est) { | |
200 | del_timer_sync(&est->timer); | |
201 | kfree_rcu(est, rcu); | |
1da177e4 LT |
202 | } |
203 | } | |
c1b56878 | 204 | EXPORT_SYMBOL(gen_kill_estimator); |
1da177e4 LT |
205 | |
206 | /** | |
96750162 | 207 | * gen_replace_estimator - replace rate estimator configuration |
1da177e4 | 208 | * @bstats: basic statistics |
e9fc2f05 | 209 | * @cpu_bstats: bstats per cpu |
1da177e4 LT |
210 | * @rate_est: rate estimator statistics |
211 | * @stats_lock: statistics lock | |
edb09eb1 | 212 | * @running: qdisc running seqcount (might be NULL) |
1da177e4 LT |
213 | * @opt: rate estimator configuration TLV |
214 | * | |
215 | * Replaces the configuration of a rate estimator by calling | |
216 | * gen_kill_estimator() and gen_new_estimator(). | |
4ec93edb | 217 | * |
1da177e4 LT |
218 | * Returns 0 on success or a negative error code. |
219 | */ | |
c1a8f1f1 | 220 | int gen_replace_estimator(struct gnet_stats_basic_packed *bstats, |
22e0f8b9 | 221 | struct gnet_stats_basic_cpu __percpu *cpu_bstats, |
1c0d32fd | 222 | struct net_rate_estimator __rcu **rate_est, |
edb09eb1 ED |
223 | spinlock_t *stats_lock, |
224 | seqcount_t *running, struct nlattr *opt) | |
1da177e4 | 225 | { |
1c0d32fd ED |
226 | return gen_new_estimator(bstats, cpu_bstats, rate_est, |
227 | stats_lock, running, opt); | |
1da177e4 | 228 | } |
c1b56878 SH |
229 | EXPORT_SYMBOL(gen_replace_estimator); |
230 | ||
231 | /** | |
232 | * gen_estimator_active - test if estimator is currently in use | |
1c0d32fd | 233 | * @rate_est: rate estimator |
c1b56878 | 234 | * |
244e6c2d | 235 | * Returns true if estimator is active, and false if not. |
c1b56878 | 236 | */ |
1c0d32fd | 237 | bool gen_estimator_active(struct net_rate_estimator __rcu **rate_est) |
c1b56878 | 238 | { |
1c0d32fd ED |
239 | return !!rcu_access_pointer(*rate_est); |
240 | } | |
241 | EXPORT_SYMBOL(gen_estimator_active); | |
ae638c47 | 242 | |
1c0d32fd ED |
243 | bool gen_estimator_read(struct net_rate_estimator __rcu **rate_est, |
244 | struct gnet_stats_rate_est64 *sample) | |
245 | { | |
246 | struct net_rate_estimator *est; | |
247 | unsigned seq; | |
248 | ||
249 | rcu_read_lock(); | |
250 | est = rcu_dereference(*rate_est); | |
251 | if (!est) { | |
252 | rcu_read_unlock(); | |
253 | return false; | |
254 | } | |
1da177e4 | 255 | |
1c0d32fd ED |
256 | do { |
257 | seq = read_seqcount_begin(&est->seq); | |
258 | sample->bps = est->avbps >> 8; | |
259 | sample->pps = est->avpps >> 8; | |
260 | } while (read_seqcount_retry(&est->seq, seq)); | |
ae638c47 | 261 | |
1c0d32fd ED |
262 | rcu_read_unlock(); |
263 | return true; | |
c1b56878 | 264 | } |
1c0d32fd | 265 | EXPORT_SYMBOL(gen_estimator_read); |