Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * ip_vs_est.c: simple rate estimator for IPVS | |
3 | * | |
1da177e4 LT |
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 | * | |
29c2026f HS |
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. | |
1da177e4 | 17 | */ |
9aada7ac HE |
18 | |
19 | #define KMSG_COMPONENT "IPVS" | |
20 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt | |
21 | ||
1da177e4 | 22 | #include <linux/kernel.h> |
14c85021 | 23 | #include <linux/jiffies.h> |
1da177e4 | 24 | #include <linux/types.h> |
4ffd2e49 | 25 | #include <linux/interrupt.h> |
90754f8e | 26 | #include <linux/sysctl.h> |
3a14a313 | 27 | #include <linux/list.h> |
1da177e4 LT |
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 | * The stored value for average bps is scaled by 2^5, so that maximal | |
49 | rate is ~2.15Gbits/s, average pps and cps are scaled by 2^10. | |
50 | ||
51 | * A lot code is taken from net/sched/estimator.c | |
52 | */ | |
53 | ||
54 | ||
b17fc996 HS |
55 | /* |
56 | * Make a summary from each cpu | |
57 | */ | |
58 | static void ip_vs_read_cpu_stats(struct ip_vs_stats_user *sum, | |
b962abdc | 59 | struct ip_vs_cpu_stats __percpu *stats) |
b17fc996 HS |
60 | { |
61 | int i; | |
d1ee4fea | 62 | bool add = false; |
b17fc996 HS |
63 | |
64 | for_each_possible_cpu(i) { | |
65 | struct ip_vs_cpu_stats *s = per_cpu_ptr(stats, i); | |
66 | unsigned int start; | |
67 | __u64 inbytes, outbytes; | |
d1ee4fea | 68 | if (add) { |
b17fc996 HS |
69 | sum->conns += s->ustats.conns; |
70 | sum->inpkts += s->ustats.inpkts; | |
71 | sum->outpkts += s->ustats.outpkts; | |
72 | do { | |
4a569c0c | 73 | start = u64_stats_fetch_begin(&s->syncp); |
b17fc996 HS |
74 | inbytes = s->ustats.inbytes; |
75 | outbytes = s->ustats.outbytes; | |
4a569c0c | 76 | } while (u64_stats_fetch_retry(&s->syncp, start)); |
b17fc996 HS |
77 | sum->inbytes += inbytes; |
78 | sum->outbytes += outbytes; | |
79 | } else { | |
d1ee4fea | 80 | add = true; |
b17fc996 HS |
81 | sum->conns = s->ustats.conns; |
82 | sum->inpkts = s->ustats.inpkts; | |
83 | sum->outpkts = s->ustats.outpkts; | |
84 | do { | |
4a569c0c | 85 | start = u64_stats_fetch_begin(&s->syncp); |
b17fc996 HS |
86 | sum->inbytes = s->ustats.inbytes; |
87 | sum->outbytes = s->ustats.outbytes; | |
4a569c0c | 88 | } while (u64_stats_fetch_retry(&s->syncp, start)); |
b17fc996 HS |
89 | } |
90 | } | |
91 | } | |
92 | ||
93 | ||
1da177e4 LT |
94 | static void estimation_timer(unsigned long arg) |
95 | { | |
96 | struct ip_vs_estimator *e; | |
97 | struct ip_vs_stats *s; | |
98 | u32 n_conns; | |
99 | u32 n_inpkts, n_outpkts; | |
100 | u64 n_inbytes, n_outbytes; | |
101 | u32 rate; | |
29c2026f HS |
102 | struct net *net = (struct net *)arg; |
103 | struct netns_ipvs *ipvs; | |
1da177e4 | 104 | |
29c2026f HS |
105 | ipvs = net_ipvs(net); |
106 | spin_lock(&ipvs->est_lock); | |
107 | list_for_each_entry(e, &ipvs->est_list, list) { | |
3a14a313 | 108 | s = container_of(e, struct ip_vs_stats, est); |
1da177e4 LT |
109 | |
110 | spin_lock(&s->lock); | |
2a0751af | 111 | ip_vs_read_cpu_stats(&s->ustats, s->cpustats); |
e9c0ce23 SW |
112 | n_conns = s->ustats.conns; |
113 | n_inpkts = s->ustats.inpkts; | |
114 | n_outpkts = s->ustats.outpkts; | |
115 | n_inbytes = s->ustats.inbytes; | |
116 | n_outbytes = s->ustats.outbytes; | |
1da177e4 LT |
117 | |
118 | /* scaled by 2^10, but divided 2 seconds */ | |
29c2026f | 119 | rate = (n_conns - e->last_conns) << 9; |
1da177e4 | 120 | e->last_conns = n_conns; |
29c2026f | 121 | e->cps += ((long)rate - (long)e->cps) >> 2; |
1da177e4 | 122 | |
29c2026f | 123 | rate = (n_inpkts - e->last_inpkts) << 9; |
1da177e4 | 124 | e->last_inpkts = n_inpkts; |
29c2026f | 125 | e->inpps += ((long)rate - (long)e->inpps) >> 2; |
1da177e4 | 126 | |
29c2026f | 127 | rate = (n_outpkts - e->last_outpkts) << 9; |
1da177e4 | 128 | e->last_outpkts = n_outpkts; |
29c2026f | 129 | e->outpps += ((long)rate - (long)e->outpps) >> 2; |
1da177e4 | 130 | |
29c2026f | 131 | rate = (n_inbytes - e->last_inbytes) << 4; |
1da177e4 | 132 | e->last_inbytes = n_inbytes; |
29c2026f | 133 | e->inbps += ((long)rate - (long)e->inbps) >> 2; |
1da177e4 | 134 | |
29c2026f | 135 | rate = (n_outbytes - e->last_outbytes) << 4; |
1da177e4 | 136 | e->last_outbytes = n_outbytes; |
29c2026f | 137 | e->outbps += ((long)rate - (long)e->outbps) >> 2; |
1da177e4 LT |
138 | spin_unlock(&s->lock); |
139 | } | |
29c2026f HS |
140 | spin_unlock(&ipvs->est_lock); |
141 | mod_timer(&ipvs->est_timer, jiffies + 2*HZ); | |
1da177e4 LT |
142 | } |
143 | ||
6ef757f9 | 144 | void ip_vs_start_estimator(struct net *net, struct ip_vs_stats *stats) |
1da177e4 | 145 | { |
29c2026f | 146 | struct netns_ipvs *ipvs = net_ipvs(net); |
3a14a313 | 147 | struct ip_vs_estimator *est = &stats->est; |
1da177e4 | 148 | |
3a14a313 | 149 | INIT_LIST_HEAD(&est->list); |
1da177e4 | 150 | |
29c2026f HS |
151 | spin_lock_bh(&ipvs->est_lock); |
152 | list_add(&est->list, &ipvs->est_list); | |
153 | spin_unlock_bh(&ipvs->est_lock); | |
1da177e4 LT |
154 | } |
155 | ||
6ef757f9 | 156 | void ip_vs_stop_estimator(struct net *net, struct ip_vs_stats *stats) |
1da177e4 | 157 | { |
29c2026f | 158 | struct netns_ipvs *ipvs = net_ipvs(net); |
3a14a313 SW |
159 | struct ip_vs_estimator *est = &stats->est; |
160 | ||
29c2026f | 161 | spin_lock_bh(&ipvs->est_lock); |
3a14a313 | 162 | list_del(&est->list); |
29c2026f | 163 | spin_unlock_bh(&ipvs->est_lock); |
1da177e4 LT |
164 | } |
165 | ||
166 | void ip_vs_zero_estimator(struct ip_vs_stats *stats) | |
167 | { | |
3a14a313 | 168 | struct ip_vs_estimator *est = &stats->est; |
55a3d4e1 JA |
169 | struct ip_vs_stats_user *u = &stats->ustats; |
170 | ||
171 | /* reset counters, caller must hold the stats->lock lock */ | |
172 | est->last_inbytes = u->inbytes; | |
173 | est->last_outbytes = u->outbytes; | |
174 | est->last_conns = u->conns; | |
175 | est->last_inpkts = u->inpkts; | |
176 | est->last_outpkts = u->outpkts; | |
3a14a313 SW |
177 | est->cps = 0; |
178 | est->inpps = 0; | |
179 | est->outpps = 0; | |
180 | est->inbps = 0; | |
181 | est->outbps = 0; | |
1da177e4 | 182 | } |
a919cf4b | 183 | |
ea9f22cc JA |
184 | /* Get decoded rates */ |
185 | void ip_vs_read_estimator(struct ip_vs_stats_user *dst, | |
186 | struct ip_vs_stats *stats) | |
187 | { | |
188 | struct ip_vs_estimator *e = &stats->est; | |
189 | ||
190 | dst->cps = (e->cps + 0x1FF) >> 10; | |
191 | dst->inpps = (e->inpps + 0x1FF) >> 10; | |
192 | dst->outpps = (e->outpps + 0x1FF) >> 10; | |
193 | dst->inbps = (e->inbps + 0xF) >> 5; | |
194 | dst->outbps = (e->outbps + 0xF) >> 5; | |
195 | } | |
196 | ||
503cf15a | 197 | int __net_init ip_vs_estimator_net_init(struct net *net) |
61b1ab45 | 198 | { |
29c2026f HS |
199 | struct netns_ipvs *ipvs = net_ipvs(net); |
200 | ||
29c2026f HS |
201 | INIT_LIST_HEAD(&ipvs->est_list); |
202 | spin_lock_init(&ipvs->est_lock); | |
203 | setup_timer(&ipvs->est_timer, estimation_timer, (unsigned long)net); | |
204 | mod_timer(&ipvs->est_timer, jiffies + 2 * HZ); | |
61b1ab45 HS |
205 | return 0; |
206 | } | |
207 | ||
503cf15a | 208 | void __net_exit ip_vs_estimator_net_cleanup(struct net *net) |
29c2026f HS |
209 | { |
210 | del_timer_sync(&net_ipvs(net)->est_timer); | |
211 | } |