1 /* Copyright (c) 2010, 2012, 2013, 2014 Nicira, Inc.
3 * Licensed under the Apache License, Version 2.0 (the "License");
4 * you may not use this file except in compliance with the License.
5 * You may obtain a copy of the License at:
7 * http://www.apache.org/licenses/LICENSE-2.0
9 * Unless required by applicable law or agreed to in writing, software
10 * distributed under the License is distributed on an "AS IS" BASIS,
11 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 * See the License for the specific language governing permissions and
13 * limitations under the License.
18 #include "system-stats.h"
29 #if HAVE_SYS_STATVFS_H
30 #include <sys/statvfs.h>
36 #include "dynamic-string.h"
40 #include "ovs-thread.h"
41 #include "poll-loop.h"
45 #include "openvswitch/vlog.h"
47 VLOG_DEFINE_THIS_MODULE(system_stats
);
49 /* #ifdefs make it a pain to maintain code: you have to try to build both ways.
50 * Thus, this file tries to compile as much of the code as possible regardless
51 * of the target, by writing "if (LINUX)" instead of "#ifdef __linux__" where
52 * this is possible. */
55 #include <asm/param.h>
61 get_cpu_cores(struct smap
*stats
)
63 long int n_cores
= count_cpu_cores();
65 smap_add_format(stats
, "cpu", "%ld", n_cores
);
70 get_load_average(struct smap
*stats OVS_UNUSED
)
75 if (getloadavg(loadavg
, 3) == 3) {
76 smap_add_format(stats
, "load_average", "%.2f,%.2f,%.2f",
77 loadavg
[0], loadavg
[1], loadavg
[2]);
85 static unsigned int cached
;
89 long int value
= sysconf(_SC_PAGESIZE
);
93 GetSystemInfo(&sysinfo
);
94 value
= sysinfo
.dwPageSize
;
105 get_memory_stats(struct smap
*stats
)
108 unsigned int pagesize
= get_page_size();
109 #ifdef _SC_PHYS_PAGES
110 long int phys_pages
= sysconf(_SC_PHYS_PAGES
);
112 long int phys_pages
= 0;
114 #ifdef _SC_AVPHYS_PAGES
115 long int avphys_pages
= sysconf(_SC_AVPHYS_PAGES
);
117 long int avphys_pages
= 0;
119 int mem_total
, mem_used
;
122 if (pagesize
<= 0 || phys_pages
<= 0 || avphys_pages
<= 0) {
126 mem_total
= phys_pages
* (pagesize
/ 1024);
127 mem_used
= (phys_pages
- avphys_pages
) * (pagesize
/ 1024);
129 MEMORYSTATUS memory_status
;
130 GlobalMemoryStatus(&memory_status
);
132 mem_total
= memory_status
.dwTotalPhys
;
133 mem_used
= memory_status
.dwTotalPhys
- memory_status
.dwAvailPhys
;
135 smap_add_format(stats
, "memory", "%d,%d", mem_total
, mem_used
);
137 static const char file_name
[] = "/proc/meminfo";
138 int mem_used
, mem_cache
, swap_used
;
149 stream
= fopen(file_name
, "r");
151 VLOG_WARN_ONCE("%s: open failed (%s)",
152 file_name
, ovs_strerror(errno
));
157 shash_add(&dict
, "MemTotal", &mem_total
);
158 shash_add(&dict
, "MemFree", &mem_free
);
159 shash_add(&dict
, "Buffers", &buffers
);
160 shash_add(&dict
, "Cached", &cached
);
161 shash_add(&dict
, "SwapTotal", &swap_total
);
162 shash_add(&dict
, "SwapFree", &swap_free
);
163 while (fgets(line
, sizeof line
, stream
)) {
167 if (ovs_scan(line
, "%15[^:]: %u", key
, &value
)) {
168 int *valuep
= shash_find_data(&dict
, key
);
175 shash_destroy(&dict
);
177 mem_used
= mem_total
- mem_free
;
178 mem_cache
= buffers
+ cached
;
179 swap_used
= swap_total
- swap_free
;
180 smap_add_format(stats
, "memory", "%d,%d,%d,%d,%d",
181 mem_total
, mem_used
, mem_cache
, swap_total
, swap_used
);
185 /* Returns the time at which the system booted, as the number of milliseconds
186 * since the epoch, or 0 if the time of boot cannot be determined. */
190 static long long int cache_expiration
= LLONG_MIN
;
191 static long long int boot_time
;
195 if (time_msec() >= cache_expiration
) {
196 static const char stat_file
[] = "/proc/stat";
200 cache_expiration
= time_msec() + 5 * 1000;
202 stream
= fopen(stat_file
, "r");
204 VLOG_ERR_ONCE("%s: open failed (%s)",
205 stat_file
, ovs_strerror(errno
));
209 while (fgets(line
, sizeof line
, stream
)) {
211 if (ovs_scan(line
, "btime %lld", &btime
)) {
212 boot_time
= btime
* 1000;
216 VLOG_ERR_ONCE("%s: btime not found", stat_file
);
223 static unsigned long long int
224 ticks_to_ms(unsigned long long int ticks
)
232 #if USER_HZ == 100 /* Common case. */
233 return ticks
* (1000 / USER_HZ
);
234 #else /* Alpha and some other architectures. */
235 double factor
= 1000.0 / USER_HZ
;
236 return ticks
* factor
+ 0.5;
240 struct raw_process_info
{
241 unsigned long int vsz
; /* Virtual size, in kB. */
242 unsigned long int rss
; /* Resident set size, in kB. */
243 long long int uptime
; /* ms since started. */
244 long long int cputime
; /* ms of CPU used during 'uptime'. */
245 pid_t ppid
; /* Parent. */
246 char name
[18]; /* Name (surrounded by parentheses). */
250 get_raw_process_info(pid_t pid
, struct raw_process_info
*raw
)
252 unsigned long long int vsize
, rss
, start_time
, utime
, stime
;
253 long long int start_msec
;
261 sprintf(file_name
, "/proc/%lu/stat", (unsigned long int) pid
);
262 stream
= fopen(file_name
, "r");
264 VLOG_ERR_ONCE("%s: open failed (%s)",
265 file_name
, ovs_strerror(errno
));
270 "%*d " /* (1. pid) */
271 "%17s " /* 2. process name */
272 "%*c " /* (3. state) */
274 "%*d " /* (5. pgid) */
275 "%*d " /* (6. sid) */
276 "%*d " /* (7. tty_nr) */
277 "%*d " /* (8. tty_pgrp) */
278 "%*u " /* (9. flags) */
279 "%*u " /* (10. min_flt) */
280 "%*u " /* (11. cmin_flt) */
281 "%*u " /* (12. maj_flt) */
282 "%*u " /* (13. cmaj_flt) */
283 "%llu " /* 14. utime */
284 "%llu " /* 15. stime */
285 "%*d " /* (16. cutime) */
286 "%*d " /* (17. cstime) */
287 "%*d " /* (18. priority) */
288 "%*d " /* (19. nice) */
289 "%*d " /* (20. num_threads) */
290 "%*d " /* (21. always 0) */
291 "%llu " /* 22. start_time */
292 "%llu " /* 23. vsize */
293 "%llu " /* 24. rss */
295 /* These are here for documentation but #if'd out to save
296 * actually parsing them from the stream for no benefit. */
297 "%*lu " /* (25. rsslim) */
298 "%*lu " /* (26. start_code) */
299 "%*lu " /* (27. end_code) */
300 "%*lu " /* (28. start_stack) */
301 "%*lu " /* (29. esp) */
302 "%*lu " /* (30. eip) */
303 "%*lu " /* (31. pending signals) */
304 "%*lu " /* (32. blocked signals) */
305 "%*lu " /* (33. ignored signals) */
306 "%*lu " /* (34. caught signals) */
307 "%*lu " /* (35. whcan) */
308 "%*lu " /* (36. always 0) */
309 "%*lu " /* (37. always 0) */
310 "%*d " /* (38. exit_signal) */
311 "%*d " /* (39. task_cpu) */
312 "%*u " /* (40. rt_priority) */
313 "%*u " /* (41. policy) */
314 "%*llu " /* (42. blkio_ticks) */
315 "%*lu " /* (43. gtime) */
316 "%*ld" /* (44. cgtime) */
318 , raw
->name
, &ppid
, &utime
, &stime
, &start_time
, &vsize
, &rss
);
321 VLOG_ERR_ONCE("%s: fscanf failed", file_name
);
325 start_msec
= get_boot_time() + ticks_to_ms(start_time
);
327 raw
->vsz
= vsize
/ 1024;
328 raw
->rss
= rss
* (getpagesize() / 1024);
329 raw
->uptime
= time_wall_msec() - start_msec
;
330 raw
->cputime
= ticks_to_ms(utime
+ stime
);
337 count_crashes(pid_t pid
)
347 sprintf(file_name
, "/proc/%lu/cmdline", (unsigned long int) pid
);
348 stream
= fopen(file_name
, "r");
350 VLOG_WARN_ONCE("%s: open failed (%s)", file_name
, ovs_strerror(errno
));
354 if (!fgets(line
, sizeof line
, stream
)) {
355 VLOG_WARN_ONCE("%s: read failed (%s)", file_name
,
356 feof(stream
) ? "end of file" : ovs_strerror(errno
));
360 paren
= strchr(line
, '(');
363 if (ovs_scan(paren
+ 1, "%d", &x
)) {
374 struct process_info
{
375 unsigned long int vsz
; /* Virtual size, in kB. */
376 unsigned long int rss
; /* Resident set size, in kB. */
377 long long int booted
; /* ms since monitor started. */
378 int crashes
; /* # of crashes (usually 0). */
379 long long int uptime
; /* ms since last (re)started by monitor. */
380 long long int cputime
; /* ms of CPU used during 'uptime'. */
384 get_process_info(pid_t pid
, struct process_info
*pinfo
)
386 struct raw_process_info child
;
389 if (!get_raw_process_info(pid
, &child
)) {
393 pinfo
->vsz
= child
.vsz
;
394 pinfo
->rss
= child
.rss
;
395 pinfo
->booted
= child
.uptime
;
397 pinfo
->uptime
= child
.uptime
;
398 pinfo
->cputime
= child
.cputime
;
401 struct raw_process_info parent
;
403 get_raw_process_info(child
.ppid
, &parent
);
404 if (!strcmp(child
.name
, parent
.name
)) {
405 pinfo
->booted
= parent
.uptime
;
406 pinfo
->crashes
= count_crashes(child
.ppid
);
414 get_process_stats(struct smap
*stats
)
420 dir
= opendir(ovs_rundir());
422 VLOG_ERR_ONCE("%s: open failed (%s)",
423 ovs_rundir(), ovs_strerror(errno
));
427 while ((de
= readdir(dir
)) != NULL
) {
428 struct process_info pinfo
;
434 #ifdef _DIRENT_HAVE_D_TYPE
435 if (de
->d_type
!= DT_UNKNOWN
&& de
->d_type
!= DT_REG
) {
440 extension
= strrchr(de
->d_name
, '.');
441 if (!extension
|| strcmp(extension
, ".pid")) {
445 file_name
= xasprintf("%s/%s", ovs_rundir(), de
->d_name
);
446 pid
= read_pidfile(file_name
);
452 key
= xasprintf("process_%.*s",
453 (int) (extension
- de
->d_name
), de
->d_name
);
454 if (!smap_get(stats
, key
)) {
455 if (LINUX
&& get_process_info(pid
, &pinfo
)) {
456 smap_add_format(stats
, key
, "%lu,%lu,%lld,%d,%lld,%lld",
457 pinfo
.vsz
, pinfo
.rss
, pinfo
.cputime
,
458 pinfo
.crashes
, pinfo
.booted
, pinfo
.uptime
);
460 smap_add(stats
, key
, "");
471 get_filesys_stats(struct smap
*stats OVS_UNUSED
)
473 #if HAVE_GETMNTENT_R && HAVE_STATVFS
474 static const char file_name
[] = "/etc/mtab";
475 struct mntent mntent
;
481 stream
= setmntent(file_name
, "r");
483 VLOG_ERR_ONCE("%s: open failed (%s)", file_name
, ovs_strerror(errno
));
488 while ((me
= getmntent_r(stream
, &mntent
, buf
, sizeof buf
)) != NULL
) {
489 unsigned long long int total
, free
;
493 /* Skip non-local and read-only filesystems. */
494 if (strncmp(me
->mnt_fsname
, "/dev", 4)
495 || !strstr(me
->mnt_opts
, "rw")) {
499 /* Given the mount point we can stat the file system. */
500 if (statvfs(me
->mnt_dir
, &vfs
) && vfs
.f_flag
& ST_RDONLY
) {
505 /* Now format the data. */
507 ds_put_char(&s
, ' ');
509 for (p
= me
->mnt_dir
; *p
!= '\0'; p
++) {
510 ds_put_char(&s
, *p
== ' ' || *p
== ',' ? '_' : *p
);
512 total
= (unsigned long long int) vfs
.f_frsize
* vfs
.f_blocks
/ 1024;
513 free
= (unsigned long long int) vfs
.f_frsize
* vfs
.f_bfree
/ 1024;
514 ds_put_format(&s
, ",%llu,%llu", total
, total
- free
);
519 smap_add(stats
, "file_systems", ds_cstr(&s
));
522 #endif /* HAVE_GETMNTENT_R && HAVE_STATVFS */
525 #define SYSTEM_STATS_INTERVAL (5 * 1000) /* In milliseconds. */
527 static struct ovs_mutex mutex
= OVS_MUTEX_INITIALIZER
;
528 static pthread_cond_t cond
= PTHREAD_COND_INITIALIZER
;
529 static struct latch latch
OVS_GUARDED_BY(mutex
);
531 static bool started
OVS_GUARDED_BY(mutex
);
532 static struct smap
*system_stats
OVS_GUARDED_BY(mutex
);
534 OVS_NO_RETURN
static void *system_stats_thread_func(void *);
535 static void discard_stats(void);
537 /* Enables or disables system stats collection, according to 'enable'. */
539 system_stats_enable(bool enable
)
541 if (enabled
!= enable
) {
542 ovs_mutex_lock(&mutex
);
545 ovs_thread_create("system_stats",
546 system_stats_thread_func
, NULL
);
551 xpthread_cond_signal(&cond
);
554 ovs_mutex_unlock(&mutex
);
558 /* Tries to obtain a new snapshot of system stats every SYSTEM_STATS_INTERVAL
561 * When a new snapshot is available (which only occurs if system stats are
562 * enabled), returns it as an smap owned by the caller. The caller must use
563 * both smap_destroy() and free() to completely free the returned data.
565 * When no new snapshot is available, returns NULL. */
567 system_stats_run(void)
569 struct smap
*stats
= NULL
;
571 ovs_mutex_lock(&mutex
);
576 stats
= system_stats
;
582 ovs_mutex_unlock(&mutex
);
587 /* Causes poll_block() to wake up when system_stats_run() needs to be
590 system_stats_wait(void)
598 discard_stats(void) OVS_REQUIRES(mutex
)
601 smap_destroy(system_stats
);
608 system_stats_thread_func(void *arg OVS_UNUSED
)
610 pthread_detach(pthread_self());
613 long long int next_refresh
;
616 ovs_mutex_lock(&mutex
);
618 ovs_mutex_cond_wait(&cond
, &mutex
);
620 ovs_mutex_unlock(&mutex
);
622 stats
= xmalloc(sizeof *stats
);
624 get_cpu_cores(stats
);
625 get_load_average(stats
);
626 get_memory_stats(stats
);
627 get_process_stats(stats
);
628 get_filesys_stats(stats
);
630 ovs_mutex_lock(&mutex
);
632 system_stats
= stats
;
634 ovs_mutex_unlock(&mutex
);
636 next_refresh
= time_msec() + SYSTEM_STATS_INTERVAL
;
638 poll_timer_wait_until(next_refresh
);
640 } while (time_msec() < next_refresh
);