net/netfilter/ipvs/ip_vs_est.c

   1 /*
   2  * ip_vs_est.c: simple rate estimator for IPVS
   3  *
   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:     Hans Schillstrom <hans.schillstrom@ericsson.com>
  12  *              Network name space (netns) aware.
  13  *              Global data moved to netns i.e struct netns_ipvs
  14  *              Affected data: est_list and est_lock.
  15  *              estimation_timer() runs with timer per netns.
  16  *              get_stats()) do the per cpu summing.
  17  */
  18
  19 #define KMSG_COMPONENT "IPVS"
  20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  21
  22 #include <linux/kernel.h>
  23 #include <linux/jiffies.h>
  24 #include <linux/types.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/sysctl.h>
  27 #include <linux/list.h>
  28
  29 #include <net/ip_vs.h>
  30
  31 /*
  32   This code is to estimate rate in a shorter interval (such as 8
  33   seconds) for virtual services and real servers. For measure rate in a
  34   long interval, it is easy to implement a user level daemon which
  35   periodically reads those statistical counters and measure rate.
  36
  37   Currently, the measurement is activated by slow timer handler. Hope
  38   this measurement will not introduce too much load.
  39
  40   We measure rate during the last 8 seconds every 2 seconds:
  41
  42     avgrate = avgrate*(1-W) + rate*W
  43
  44     where W = 2^(-2)
  45
  46   NOTES.
  47
  48   * Average bps is scaled by 2^5, while average pps and cps are scaled by 2^10.
  49
  50   * Netlink users can see 64-bit values but sockopt users are restricted
  51     to 32-bit values for conns, packets, bps, cps and pps.
  52
  53   * A lot of code is taken from net/core/gen_estimator.c
  54  */
  55
  56
  57 /*
  58  * Make a summary from each cpu
  59  */
  60 static void ip_vs_read_cpu_stats(struct ip_vs_kstats *sum,
  61                                  struct ip_vs_cpu_stats __percpu *stats)
  62 {
  63         int i;
  64         bool add = false;
  65
  66         for_each_possible_cpu(i) {
  67                 struct ip_vs_cpu_stats *s = per_cpu_ptr(stats, i);
  68                 unsigned int start;
  69                 u64 conns, inpkts, outpkts, inbytes, outbytes;
  70
  71                 if (add) {
  72                         do {
  73                                 start = u64_stats_fetch_begin(&s->syncp);
  74                                 conns = s->cnt.conns;
  75                                 inpkts = s->cnt.inpkts;
  76                                 outpkts = s->cnt.outpkts;
  77                                 inbytes = s->cnt.inbytes;
  78                                 outbytes = s->cnt.outbytes;
  79                         } while (u64_stats_fetch_retry(&s->syncp, start));
  80                         sum->conns += conns;
  81                         sum->inpkts += inpkts;
  82                         sum->outpkts += outpkts;
  83                         sum->inbytes += inbytes;
  84                         sum->outbytes += outbytes;
  85                 } else {
  86                         add = true;
  87                         do {
  88                                 start = u64_stats_fetch_begin(&s->syncp);
  89                                 sum->conns = s->cnt.conns;
  90                                 sum->inpkts = s->cnt.inpkts;
  91                                 sum->outpkts = s->cnt.outpkts;
  92                                 sum->inbytes = s->cnt.inbytes;
  93                                 sum->outbytes = s->cnt.outbytes;
  94                         } while (u64_stats_fetch_retry(&s->syncp, start));
  95                 }
  96         }
  97 }
  98
  99
 100 static void estimation_timer(unsigned long arg)
 101 {
 102         struct ip_vs_estimator *e;
 103         struct ip_vs_stats *s;
 104         u64 rate;
 105         struct netns_ipvs *ipvs = (struct netns_ipvs *)arg;
 106
 107         spin_lock(&ipvs->est_lock);
 108         list_for_each_entry(e, &ipvs->est_list, list) {
 109                 s = container_of(e, struct ip_vs_stats, est);
 110
 111                 spin_lock(&s->lock);
 112                 ip_vs_read_cpu_stats(&s->kstats, s->cpustats);
 113
 114                 /* scaled by 2^10, but divided 2 seconds */
 115                 rate = (s->kstats.conns - e->last_conns) << 9;
 116                 e->last_conns = s->kstats.conns;
 117                 e->cps += ((s64)rate - (s64)e->cps) >> 2;
 118
 119                 rate = (s->kstats.inpkts - e->last_inpkts) << 9;
 120                 e->last_inpkts = s->kstats.inpkts;
 121                 e->inpps += ((s64)rate - (s64)e->inpps) >> 2;
 122
 123                 rate = (s->kstats.outpkts - e->last_outpkts) << 9;
 124                 e->last_outpkts = s->kstats.outpkts;
 125                 e->outpps += ((s64)rate - (s64)e->outpps) >> 2;
 126
 127                 /* scaled by 2^5, but divided 2 seconds */
 128                 rate = (s->kstats.inbytes - e->last_inbytes) << 4;
 129                 e->last_inbytes = s->kstats.inbytes;
 130                 e->inbps += ((s64)rate - (s64)e->inbps) >> 2;
 131
 132                 rate = (s->kstats.outbytes - e->last_outbytes) << 4;
 133                 e->last_outbytes = s->kstats.outbytes;
 134                 e->outbps += ((s64)rate - (s64)e->outbps) >> 2;
 135                 spin_unlock(&s->lock);
 136         }
 137         spin_unlock(&ipvs->est_lock);
 138         mod_timer(&ipvs->est_timer, jiffies + 2*HZ);
 139 }
 140
 141 void ip_vs_start_estimator(struct netns_ipvs *ipvs, struct ip_vs_stats *stats)
 142 {
 143         struct ip_vs_estimator *est = &stats->est;
 144
 145         INIT_LIST_HEAD(&est->list);
 146
 147         spin_lock_bh(&ipvs->est_lock);
 148         list_add(&est->list, &ipvs->est_list);
 149         spin_unlock_bh(&ipvs->est_lock);
 150 }
 151
 152 void ip_vs_stop_estimator(struct netns_ipvs *ipvs, struct ip_vs_stats *stats)
 153 {
 154         struct ip_vs_estimator *est = &stats->est;
 155
 156         spin_lock_bh(&ipvs->est_lock);
 157         list_del(&est->list);
 158         spin_unlock_bh(&ipvs->est_lock);
 159 }
 160
 161 void ip_vs_zero_estimator(struct ip_vs_stats *stats)
 162 {
 163         struct ip_vs_estimator *est = &stats->est;
 164         struct ip_vs_kstats *k = &stats->kstats;
 165
 166         /* reset counters, caller must hold the stats->lock lock */
 167         est->last_inbytes = k->inbytes;
 168         est->last_outbytes = k->outbytes;
 169         est->last_conns = k->conns;
 170         est->last_inpkts = k->inpkts;
 171         est->last_outpkts = k->outpkts;
 172         est->cps = 0;
 173         est->inpps = 0;
 174         est->outpps = 0;
 175         est->inbps = 0;
 176         est->outbps = 0;
 177 }
 178
 179 /* Get decoded rates */
 180 void ip_vs_read_estimator(struct ip_vs_kstats *dst, struct ip_vs_stats *stats)
 181 {
 182         struct ip_vs_estimator *e = &stats->est;
 183
 184         dst->cps = (e->cps + 0x1FF) >> 10;
 185         dst->inpps = (e->inpps + 0x1FF) >> 10;
 186         dst->outpps = (e->outpps + 0x1FF) >> 10;
 187         dst->inbps = (e->inbps + 0xF) >> 5;
 188         dst->outbps = (e->outbps + 0xF) >> 5;
 189 }
 190
 191 int __net_init ip_vs_estimator_net_init(struct netns_ipvs *ipvs)
 192 {
 193         INIT_LIST_HEAD(&ipvs->est_list);
 194         spin_lock_init(&ipvs->est_lock);
 195         setup_timer(&ipvs->est_timer, estimation_timer, (unsigned long)ipvs);
 196         mod_timer(&ipvs->est_timer, jiffies + 2 * HZ);
 197         return 0;
 198 }
 199
 200 void __net_exit ip_vs_estimator_net_cleanup(struct netns_ipvs *ipvs)
 201 {
 202         del_timer_sync(&ipvs->est_timer);
 203 }