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_sync *bstats; 44 spinlock_t *stats_lock; 45 seqcount_t *running; 46 struct gnet_stats_basic_sync __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 62 static void est_fetch_counters(struct net_rate_estimator *e, 63 struct gnet_stats_basic_sync *b) 64 { 65 gnet_stats_basic_sync_init(b); 66 if (e->stats_lock) 67 spin_lock(e->stats_lock); 68 69 gnet_stats_add_basic(e->running, b, e->cpu_bstats, e->bstats); 70 71 if (e->stats_lock) 72 spin_unlock(e->stats_lock); 73 74 } 75 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_sync b; 80 u64 b_bytes, b_packets; 81 u64 rate, brate; 82 83 est_fetch_counters(est, &b); 84 b_bytes = u64_stats_read(&b.bytes); 85 b_packets = u64_stats_read(&b.packets); 86 87 brate = (b_bytes - est->last_bytes) << (10 - est->intvl_log); 88 brate = (brate >> est->ewma_log) - (est->avbps >> est->ewma_log); 89 90 rate = (b_packets - est->last_packets) << (10 - est->intvl_log); 91 rate = (rate >> est->ewma_log) - (est->avpps >> est->ewma_log); 92 93 write_seqcount_begin(&est->seq); 94 est->avbps += brate; 95 est->avpps += rate; 96 write_seqcount_end(&est->seq); 97 98 est->last_bytes = b_bytes; 99 est->last_packets = b_packets; 100 101 est->next_jiffies += ((HZ/4) << est->intvl_log); 102 103 if (unlikely(time_after_eq(jiffies, est->next_jiffies))) { 104 /* Ouch... timer was delayed. */ 105 est->next_jiffies = jiffies + 1; 106 } 107 mod_timer(&est->timer, est->next_jiffies); 108 } 109 110 /** 111 * gen_new_estimator - create a new rate estimator 112 * @bstats: basic statistics 113 * @cpu_bstats: bstats per cpu 114 * @rate_est: rate estimator statistics 115 * @lock: lock for statistics and control path 116 * @running: qdisc running seqcount 117 * @opt: rate estimator configuration TLV 118 * 119 * Creates a new rate estimator with &bstats as source and &rate_est 120 * as destination. A new timer with the interval specified in the 121 * configuration TLV is created. Upon each interval, the latest statistics 122 * will be read from &bstats and the estimated rate will be stored in 123 * &rate_est with the statistics lock grabbed during this period. 124 * 125 * Returns 0 on success or a negative error code. 126 * 127 */ 128 int gen_new_estimator(struct gnet_stats_basic_sync *bstats, 129 struct gnet_stats_basic_sync __percpu *cpu_bstats, 130 struct net_rate_estimator __rcu **rate_est, 131 spinlock_t *lock, 132 seqcount_t *running, 133 struct nlattr *opt) 134 { 135 struct gnet_estimator *parm = nla_data(opt); 136 struct net_rate_estimator *old, *est; 137 struct gnet_stats_basic_sync b; 138 int intvl_log; 139 140 if (nla_len(opt) < sizeof(*parm)) 141 return -EINVAL; 142 143 /* allowed timer periods are : 144 * -2 : 250ms, -1 : 500ms, 0 : 1 sec 145 * 1 : 2 sec, 2 : 4 sec, 3 : 8 sec 146 */ 147 if (parm->interval < -2 || parm->interval > 3) 148 return -EINVAL; 149 150 if (parm->ewma_log == 0 || parm->ewma_log >= 31) 151 return -EINVAL; 152 153 est = kzalloc(sizeof(*est), GFP_KERNEL); 154 if (!est) 155 return -ENOBUFS; 156 157 seqcount_init(&est->seq); 158 intvl_log = parm->interval + 2; 159 est->bstats = bstats; 160 est->stats_lock = lock; 161 est->running = running; 162 est->ewma_log = parm->ewma_log; 163 est->intvl_log = intvl_log; 164 est->cpu_bstats = cpu_bstats; 165 166 if (lock) 167 local_bh_disable(); 168 est_fetch_counters(est, &b); 169 if (lock) 170 local_bh_enable(); 171 est->last_bytes = u64_stats_read(&b.bytes); 172 est->last_packets = u64_stats_read(&b.packets); 173 174 if (lock) 175 spin_lock_bh(lock); 176 old = rcu_dereference_protected(*rate_est, 1); 177 if (old) { 178 del_timer_sync(&old->timer); 179 est->avbps = old->avbps; 180 est->avpps = old->avpps; 181 } 182 183 est->next_jiffies = jiffies + ((HZ/4) << intvl_log); 184 timer_setup(&est->timer, est_timer, 0); 185 mod_timer(&est->timer, est->next_jiffies); 186 187 rcu_assign_pointer(*rate_est, est); 188 if (lock) 189 spin_unlock_bh(lock); 190 if (old) 191 kfree_rcu(old, rcu); 192 return 0; 193 } 194 EXPORT_SYMBOL(gen_new_estimator); 195 196 /** 197 * gen_kill_estimator - remove a rate estimator 198 * @rate_est: rate estimator 199 * 200 * Removes the rate estimator. 201 * 202 */ 203 void gen_kill_estimator(struct net_rate_estimator __rcu **rate_est) 204 { 205 struct net_rate_estimator *est; 206 207 est = xchg((__force struct net_rate_estimator **)rate_est, NULL); 208 if (est) { 209 del_timer_sync(&est->timer); 210 kfree_rcu(est, rcu); 211 } 212 } 213 EXPORT_SYMBOL(gen_kill_estimator); 214 215 /** 216 * gen_replace_estimator - replace rate estimator configuration 217 * @bstats: basic statistics 218 * @cpu_bstats: bstats per cpu 219 * @rate_est: rate estimator statistics 220 * @lock: lock for statistics and control path 221 * @running: qdisc running seqcount (might be NULL) 222 * @opt: rate estimator configuration TLV 223 * 224 * Replaces the configuration of a rate estimator by calling 225 * gen_kill_estimator() and gen_new_estimator(). 226 * 227 * Returns 0 on success or a negative error code. 228 */ 229 int gen_replace_estimator(struct gnet_stats_basic_sync *bstats, 230 struct gnet_stats_basic_sync __percpu *cpu_bstats, 231 struct net_rate_estimator __rcu **rate_est, 232 spinlock_t *lock, 233 seqcount_t *running, struct nlattr *opt) 234 { 235 return gen_new_estimator(bstats, cpu_bstats, rate_est, 236 lock, running, opt); 237 } 238 EXPORT_SYMBOL(gen_replace_estimator); 239 240 /** 241 * gen_estimator_active - test if estimator is currently in use 242 * @rate_est: rate estimator 243 * 244 * Returns true if estimator is active, and false if not. 245 */ 246 bool gen_estimator_active(struct net_rate_estimator __rcu **rate_est) 247 { 248 return !!rcu_access_pointer(*rate_est); 249 } 250 EXPORT_SYMBOL(gen_estimator_active); 251 252 bool gen_estimator_read(struct net_rate_estimator __rcu **rate_est, 253 struct gnet_stats_rate_est64 *sample) 254 { 255 struct net_rate_estimator *est; 256 unsigned seq; 257 258 rcu_read_lock(); 259 est = rcu_dereference(*rate_est); 260 if (!est) { 261 rcu_read_unlock(); 262 return false; 263 } 264 265 do { 266 seq = read_seqcount_begin(&est->seq); 267 sample->bps = est->avbps >> 8; 268 sample->pps = est->avpps >> 8; 269 } while (read_seqcount_retry(&est->seq, seq)); 270 271 rcu_read_unlock(); 272 return true; 273 } 274 EXPORT_SYMBOL(gen_estimator_read); 275