1 /* Copyright (c) 2010, 2012 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"
39 #include "poll-loop.h"
46 VLOG_DEFINE_THIS_MODULE(system_stats);
48 /* #ifdefs make it a pain to maintain code: you have to try to build both ways.
49 * Thus, this file tries to compile as much of the code as possible regardless
50 * of the target, by writing "if (LINUX_DATAPATH)" instead of "#ifdef
51 * __linux__" where this is possible. */
53 #include <asm/param.h>
55 #define LINUX_DATAPATH 0
59 get_cpu_cores(struct smap *stats)
61 long int n_cores = sysconf(_SC_NPROCESSORS_ONLN);
63 smap_add_format(stats, "cpu", "%ld", n_cores);
68 get_load_average(struct smap *stats OVS_UNUSED)
73 if (getloadavg(loadavg, 3) == 3) {
74 smap_add_format(stats, "load_average", "%.2f,%.2f,%.2f",
75 loadavg[0], loadavg[1], loadavg[2]);
83 static unsigned int cached;
86 long int value = sysconf(_SC_PAGESIZE);
96 get_memory_stats(struct smap *stats)
98 if (!LINUX_DATAPATH) {
99 unsigned int pagesize = get_page_size();
100 long int phys_pages = sysconf(_SC_PHYS_PAGES);
101 #ifdef _SC_AVPHYS_PAGES
102 long int avphys_pages = sysconf(_SC_AVPHYS_PAGES);
104 long int avphys_pages = 0;
106 int mem_total, mem_used;
108 if (pagesize <= 0 || phys_pages <= 0 || avphys_pages <= 0) {
112 mem_total = phys_pages * (pagesize / 1024);
113 mem_used = (phys_pages - avphys_pages) * (pagesize / 1024);
114 smap_add_format(stats, "memory", "%d,%d", mem_total, mem_used);
116 static const char file_name[] = "/proc/meminfo";
117 int mem_used, mem_cache, swap_used;
128 stream = fopen(file_name, "r");
130 VLOG_WARN_ONCE("%s: open failed (%s)", file_name, strerror(errno));
135 shash_add(&dict, "MemTotal", &mem_total);
136 shash_add(&dict, "MemFree", &mem_free);
137 shash_add(&dict, "Buffers", &buffers);
138 shash_add(&dict, "Cached", &cached);
139 shash_add(&dict, "SwapTotal", &swap_total);
140 shash_add(&dict, "SwapFree", &swap_free);
141 while (fgets(line, sizeof line, stream)) {
145 if (sscanf(line, "%15[^:]: %u", key, &value) == 2) {
146 int *valuep = shash_find_data(&dict, key);
153 shash_destroy(&dict);
155 mem_used = mem_total - mem_free;
156 mem_cache = buffers + cached;
157 swap_used = swap_total - swap_free;
158 smap_add_format(stats, "memory", "%d,%d,%d,%d,%d",
159 mem_total, mem_used, mem_cache, swap_total, swap_used);
163 /* Returns the time at which the system booted, as the number of milliseconds
164 * since the epoch, or 0 if the time of boot cannot be determined. */
168 static long long int cache_expiration = LLONG_MIN;
169 static long long int boot_time;
171 ovs_assert(LINUX_DATAPATH);
173 if (time_msec() >= cache_expiration) {
174 static const char stat_file[] = "/proc/stat";
178 cache_expiration = time_msec() + 5 * 1000;
180 stream = fopen(stat_file, "r");
182 VLOG_ERR_ONCE("%s: open failed (%s)", stat_file, strerror(errno));
186 while (fgets(line, sizeof line, stream)) {
188 if (sscanf(line, "btime %lld", &btime) == 1) {
189 boot_time = btime * 1000;
193 VLOG_ERR_ONCE("%s: btime not found", stat_file);
200 static unsigned long long int
201 ticks_to_ms(unsigned long long int ticks)
203 ovs_assert(LINUX_DATAPATH);
209 #if USER_HZ == 100 /* Common case. */
210 return ticks * (1000 / USER_HZ);
211 #else /* Alpha and some other architectures. */
212 double factor = 1000.0 / USER_HZ;
213 return ticks * factor + 0.5;
217 struct raw_process_info {
218 unsigned long int vsz; /* Virtual size, in kB. */
219 unsigned long int rss; /* Resident set size, in kB. */
220 long long int uptime; /* ms since started. */
221 long long int cputime; /* ms of CPU used during 'uptime'. */
222 pid_t ppid; /* Parent. */
223 char name[18]; /* Name (surrounded by parentheses). */
227 get_raw_process_info(pid_t pid, struct raw_process_info *raw)
229 unsigned long long int vsize, rss, start_time, utime, stime;
230 long long int start_msec;
236 ovs_assert(LINUX_DATAPATH);
238 sprintf(file_name, "/proc/%lu/stat", (unsigned long int) pid);
239 stream = fopen(file_name, "r");
241 VLOG_ERR_ONCE("%s: open failed (%s)", file_name, strerror(errno));
246 "%*d " /* (1. pid) */
247 "%17s " /* 2. process name */
248 "%*c " /* (3. state) */
250 "%*d " /* (5. pgid) */
251 "%*d " /* (6. sid) */
252 "%*d " /* (7. tty_nr) */
253 "%*d " /* (8. tty_pgrp) */
254 "%*u " /* (9. flags) */
255 "%*u " /* (10. min_flt) */
256 "%*u " /* (11. cmin_flt) */
257 "%*u " /* (12. maj_flt) */
258 "%*u " /* (13. cmaj_flt) */
259 "%llu " /* 14. utime */
260 "%llu " /* 15. stime */
261 "%*d " /* (16. cutime) */
262 "%*d " /* (17. cstime) */
263 "%*d " /* (18. priority) */
264 "%*d " /* (19. nice) */
265 "%*d " /* (20. num_threads) */
266 "%*d " /* (21. always 0) */
267 "%llu " /* 22. start_time */
268 "%llu " /* 23. vsize */
269 "%llu " /* 24. rss */
271 /* These are here for documentation but #if'd out to save
272 * actually parsing them from the stream for no benefit. */
273 "%*lu " /* (25. rsslim) */
274 "%*lu " /* (26. start_code) */
275 "%*lu " /* (27. end_code) */
276 "%*lu " /* (28. start_stack) */
277 "%*lu " /* (29. esp) */
278 "%*lu " /* (30. eip) */
279 "%*lu " /* (31. pending signals) */
280 "%*lu " /* (32. blocked signals) */
281 "%*lu " /* (33. ignored signals) */
282 "%*lu " /* (34. caught signals) */
283 "%*lu " /* (35. whcan) */
284 "%*lu " /* (36. always 0) */
285 "%*lu " /* (37. always 0) */
286 "%*d " /* (38. exit_signal) */
287 "%*d " /* (39. task_cpu) */
288 "%*u " /* (40. rt_priority) */
289 "%*u " /* (41. policy) */
290 "%*llu " /* (42. blkio_ticks) */
291 "%*lu " /* (43. gtime) */
292 "%*ld" /* (44. cgtime) */
294 , raw->name, &ppid, &utime, &stime, &start_time, &vsize, &rss);
297 VLOG_ERR_ONCE("%s: fscanf failed", file_name);
301 start_msec = get_boot_time() + ticks_to_ms(start_time);
303 raw->vsz = vsize / 1024;
304 raw->rss = rss * (getpagesize() / 1024);
305 raw->uptime = time_wall_msec() - start_msec;
306 raw->cputime = ticks_to_ms(utime + stime);
313 count_crashes(pid_t pid)
321 ovs_assert(LINUX_DATAPATH);
323 sprintf(file_name, "/proc/%lu/cmdline", (unsigned long int) pid);
324 stream = fopen(file_name, "r");
326 VLOG_WARN_ONCE("%s: open failed (%s)", file_name, strerror(errno));
330 if (!fgets(line, sizeof line, stream)) {
331 VLOG_WARN_ONCE("%s: read failed (%s)", file_name,
332 feof(stream) ? "end of file" : strerror(errno));
336 paren = strchr(line, '(');
339 if (sscanf(paren + 1, "%d", &x) == 1) {
350 struct process_info {
351 unsigned long int vsz; /* Virtual size, in kB. */
352 unsigned long int rss; /* Resident set size, in kB. */
353 long long int booted; /* ms since monitor started. */
354 int crashes; /* # of crashes (usually 0). */
355 long long int uptime; /* ms since last (re)started by monitor. */
356 long long int cputime; /* ms of CPU used during 'uptime'. */
360 get_process_info(pid_t pid, struct process_info *pinfo)
362 struct raw_process_info child;
364 ovs_assert(LINUX_DATAPATH);
365 if (!get_raw_process_info(pid, &child)) {
369 pinfo->vsz = child.vsz;
370 pinfo->rss = child.rss;
371 pinfo->booted = child.uptime;
373 pinfo->uptime = child.uptime;
374 pinfo->cputime = child.cputime;
377 struct raw_process_info parent;
379 get_raw_process_info(child.ppid, &parent);
380 if (!strcmp(child.name, parent.name)) {
381 pinfo->booted = parent.uptime;
382 pinfo->crashes = count_crashes(child.ppid);
390 get_process_stats(struct smap *stats)
395 dir = opendir(ovs_rundir());
397 VLOG_ERR_ONCE("%s: open failed (%s)", ovs_rundir(), strerror(errno));
401 while ((de = readdir(dir)) != NULL) {
402 struct process_info pinfo;
408 #ifdef _DIRENT_HAVE_D_TYPE
409 if (de->d_type != DT_UNKNOWN && de->d_type != DT_REG) {
414 extension = strrchr(de->d_name, '.');
415 if (!extension || strcmp(extension, ".pid")) {
419 file_name = xasprintf("%s/%s", ovs_rundir(), de->d_name);
420 pid = read_pidfile(file_name);
426 key = xasprintf("process_%.*s",
427 (int) (extension - de->d_name), de->d_name);
428 if (!smap_get(stats, key)) {
429 if (LINUX_DATAPATH && get_process_info(pid, &pinfo)) {
430 smap_add_format(stats, key, "%lu,%lu,%lld,%d,%lld,%lld",
431 pinfo.vsz, pinfo.rss, pinfo.cputime,
432 pinfo.crashes, pinfo.booted, pinfo.uptime);
434 smap_add(stats, key, "");
444 get_filesys_stats(struct smap *stats OVS_UNUSED)
446 #if HAVE_SETMNTENT && HAVE_STATVFS
447 static const char file_name[] = "/etc/mtab";
452 stream = setmntent(file_name, "r");
454 VLOG_ERR_ONCE("%s: open failed (%s)", file_name, strerror(errno));
459 while ((me = getmntent(stream)) != NULL) {
460 unsigned long long int total, free;
464 /* Skip non-local and read-only filesystems. */
465 if (strncmp(me->mnt_fsname, "/dev", 4)
466 || !strstr(me->mnt_opts, "rw")) {
470 /* Given the mount point we can stat the file system. */
471 if (statvfs(me->mnt_dir, &vfs) && vfs.f_flag & ST_RDONLY) {
476 /* Now format the data. */
478 ds_put_char(&s, ' ');
480 for (p = me->mnt_dir; *p != '\0'; p++) {
481 ds_put_char(&s, *p == ' ' || *p == ',' ? '_' : *p);
483 total = (unsigned long long int) vfs.f_frsize * vfs.f_blocks / 1024;
484 free = (unsigned long long int) vfs.f_frsize * vfs.f_bfree / 1024;
485 ds_put_format(&s, ",%llu,%llu", total, total - free);
490 smap_add(stats, "file_systems", ds_cstr(&s));
493 #endif /* HAVE_SETMNTENT && HAVE_STATVFS */
496 #define SYSTEM_STATS_INTERVAL (5 * 1000) /* In milliseconds. */
498 /* Whether the client wants us to report system stats. */
502 S_DISABLED, /* Not enabled, nothing going on. */
503 S_WAITING, /* Sleeping for SYSTEM_STATS_INTERVAL ms. */
504 S_REQUEST_SENT, /* Sent a request to worker. */
505 S_REPLY_RECEIVED /* Received a reply from worker. */
508 /* In S_WAITING state: the next time to wake up.
509 * In other states: not meaningful. */
510 static long long int next_refresh;
512 /* In S_REPLY_RECEIVED: the stats that have just been received.
513 * In other states: not meaningful. */
514 static struct smap *received_stats;
516 static worker_request_func system_stats_request_cb;
517 static worker_reply_func system_stats_reply_cb;
519 /* Enables or disables system stats collection, according to 'new_enable'.
521 * Even if system stats are disabled, the caller should still periodically call
522 * system_stats_run(). */
524 system_stats_enable(bool new_enable)
526 if (new_enable != enabled) {
528 if (state == S_DISABLED) {
530 next_refresh = time_msec();
533 if (state == S_WAITING) {
537 enabled = new_enable;
541 /* Tries to obtain a new snapshot of system stats every SYSTEM_STATS_INTERVAL
544 * When a new snapshot is available (which only occurs if system stats are
545 * enabled), returns it as an smap owned by the caller. The caller must use
546 * both smap_destroy() and free() to complete free the returned data.
548 * When no new snapshot is available, returns NULL. */
550 system_stats_run(void)
557 if (time_msec() >= next_refresh) {
558 worker_request(NULL, 0, NULL, 0, system_stats_request_cb,
559 system_stats_reply_cb, NULL);
560 state = S_REQUEST_SENT;
567 case S_REPLY_RECEIVED:
570 next_refresh = time_msec() + SYSTEM_STATS_INTERVAL;
571 return received_stats;
573 smap_destroy(received_stats);
574 free(received_stats);
583 /* Causes poll_block() to wake up when system_stats_run() needs to be
586 system_stats_wait(void)
593 poll_timer_wait_until(next_refresh);
597 /* Someone else should be calling worker_wait() to wake up when the
598 * reply arrives, otherwise there's a bug. */
601 case S_REPLY_RECEIVED:
602 poll_immediate_wake();
608 system_stats_request_cb(struct ofpbuf *request OVS_UNUSED,
609 const int fds[] OVS_UNUSED, size_t n_fds OVS_UNUSED)
616 get_cpu_cores(&stats);
617 get_load_average(&stats);
618 get_memory_stats(&stats);
619 get_process_stats(&stats);
620 get_filesys_stats(&stats);
622 json = smap_to_json(&stats);
623 s = json_to_string(json, 0);
624 worker_reply(s, strlen(s) + 1, NULL, 0);
628 smap_destroy(&stats);
632 system_stats_reply_cb(struct ofpbuf *reply,
633 const int fds[] OVS_UNUSED, size_t n_fds OVS_UNUSED,
634 void *aux OVS_UNUSED)
636 struct json *json = json_from_string(reply->data);
638 received_stats = xmalloc(sizeof *received_stats);
639 smap_init(received_stats);
640 smap_from_json(received_stats, json);
642 ovs_assert(state == S_REQUEST_SENT);
643 state = S_REPLY_RECEIVED;