vswitchd/system-stats.c

   1 /* Copyright (c) 2010, 2012, 2013, 2014 Nicira, Inc.
   2  *
   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:
   6  *
   7  *     http://www.apache.org/licenses/LICENSE-2.0
   8  *
   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.
  14  */
  15
  16 #include <config.h>
  17
  18 #include "system-stats.h"
  19
  20 #include <ctype.h>
  21 #include <dirent.h>
  22 #include <errno.h>
  23 #if HAVE_MNTENT_H
  24 #include <mntent.h>
  25 #endif
  26 #include <stdint.h>
  27 #include <stdio.h>
  28 #include <stdlib.h>
  29 #if HAVE_SYS_STATVFS_H
  30 #include <sys/statvfs.h>
  31 #endif
  32 #include <unistd.h>
  33
  34 #include "daemon.h"
  35 #include "dirs.h"
  36 #include "dynamic-string.h"
  37 #include "json.h"
  38 #include "latch.h"
  39 #include "ofpbuf.h"
  40 #include "ovs-thread.h"
  41 #include "poll-loop.h"
  42 #include "shash.h"
  43 #include "smap.h"
  44 #include "timeval.h"
  45 #include "vlog.h"
  46
  47 VLOG_DEFINE_THIS_MODULE(system_stats);
  48
  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. */
  53 #ifdef __linux__
  54 #define LINUX 1
  55 #include <asm/param.h>
  56 #else
  57 #define LINUX 0
  58 #endif
  59
  60 static void
  61 get_cpu_cores(struct smap *stats)
  62 {
  63     long int n_cores = count_cpu_cores();
  64     if (n_cores > 0) {
  65         smap_add_format(stats, "cpu", "%ld", n_cores);
  66     }
  67 }
  68
  69 static void
  70 get_load_average(struct smap *stats OVS_UNUSED)
  71 {
  72 #if HAVE_GETLOADAVG
  73     double loadavg[3];
  74
  75     if (getloadavg(loadavg, 3) == 3) {
  76         smap_add_format(stats, "load_average", "%.2f,%.2f,%.2f",
  77                         loadavg[0], loadavg[1], loadavg[2]);
  78     }
  79 #endif
  80 }
  81
  82 static unsigned int
  83 get_page_size(void)
  84 {
  85     static unsigned int cached;
  86
  87     if (!cached) {
  88         long int value = sysconf(_SC_PAGESIZE);
  89         if (value >= 0) {
  90             cached = value;
  91         }
  92     }
  93
  94     return cached;
  95 }
  96
  97 static void
  98 get_memory_stats(struct smap *stats)
  99 {
 100     if (!LINUX) {
 101         unsigned int pagesize = get_page_size();
 102 #ifdef _SC_PHYS_PAGES
 103         long int phys_pages = sysconf(_SC_PHYS_PAGES);
 104 #else
 105         long int phys_pages = 0;
 106 #endif
 107 #ifdef _SC_AVPHYS_PAGES
 108         long int avphys_pages = sysconf(_SC_AVPHYS_PAGES);
 109 #else
 110         long int avphys_pages = 0;
 111 #endif
 112         int mem_total, mem_used;
 113
 114         if (pagesize <= 0 || phys_pages <= 0 || avphys_pages <= 0) {
 115             return;
 116         }
 117
 118         mem_total = phys_pages * (pagesize / 1024);
 119         mem_used = (phys_pages - avphys_pages) * (pagesize / 1024);
 120         smap_add_format(stats, "memory", "%d,%d", mem_total, mem_used);
 121     } else {
 122         static const char file_name[] = "/proc/meminfo";
 123         int mem_used, mem_cache, swap_used;
 124         int mem_free = 0;
 125         int buffers = 0;
 126         int cached = 0;
 127         int swap_free = 0;
 128         int mem_total = 0;
 129         int swap_total = 0;
 130         struct shash dict;
 131         char line[128];
 132         FILE *stream;
 133
 134         stream = fopen(file_name, "r");
 135         if (!stream) {
 136             VLOG_WARN_ONCE("%s: open failed (%s)",
 137                            file_name, ovs_strerror(errno));
 138             return;
 139         }
 140
 141         shash_init(&dict);
 142         shash_add(&dict, "MemTotal", &mem_total);
 143         shash_add(&dict, "MemFree", &mem_free);
 144         shash_add(&dict, "Buffers", &buffers);
 145         shash_add(&dict, "Cached", &cached);
 146         shash_add(&dict, "SwapTotal", &swap_total);
 147         shash_add(&dict, "SwapFree", &swap_free);
 148         while (fgets(line, sizeof line, stream)) {
 149             char key[16];
 150             int value;
 151
 152             if (ovs_scan(line, "%15[^:]: %u", key, &value)) {
 153                 int *valuep = shash_find_data(&dict, key);
 154                 if (valuep) {
 155                     *valuep = value;
 156                 }
 157             }
 158         }
 159         fclose(stream);
 160         shash_destroy(&dict);
 161
 162         mem_used = mem_total - mem_free;
 163         mem_cache = buffers + cached;
 164         swap_used = swap_total - swap_free;
 165         smap_add_format(stats, "memory", "%d,%d,%d,%d,%d",
 166                         mem_total, mem_used, mem_cache, swap_total, swap_used);
 167     }
 168 }
 169
 170 /* Returns the time at which the system booted, as the number of milliseconds
 171  * since the epoch, or 0 if the time of boot cannot be determined. */
 172 static long long int
 173 get_boot_time(void)
 174 {
 175     static long long int cache_expiration = LLONG_MIN;
 176     static long long int boot_time;
 177
 178     ovs_assert(LINUX);
 179
 180     if (time_msec() >= cache_expiration) {
 181         static const char stat_file[] = "/proc/stat";
 182         char line[128];
 183         FILE *stream;
 184
 185         cache_expiration = time_msec() + 5 * 1000;
 186
 187         stream = fopen(stat_file, "r");
 188         if (!stream) {
 189             VLOG_ERR_ONCE("%s: open failed (%s)",
 190                           stat_file, ovs_strerror(errno));
 191             return boot_time;
 192         }
 193
 194         while (fgets(line, sizeof line, stream)) {
 195             long long int btime;
 196             if (ovs_scan(line, "btime %lld", &btime)) {
 197                 boot_time = btime * 1000;
 198                 goto done;
 199             }
 200         }
 201         VLOG_ERR_ONCE("%s: btime not found", stat_file);
 202     done:
 203         fclose(stream);
 204     }
 205     return boot_time;
 206 }
 207
 208 static unsigned long long int
 209 ticks_to_ms(unsigned long long int ticks)
 210 {
 211     ovs_assert(LINUX);
 212
 213 #ifndef USER_HZ
 214 #define USER_HZ 100
 215 #endif
 216
 217 #if USER_HZ == 100              /* Common case. */
 218     return ticks * (1000 / USER_HZ);
 219 #else  /* Alpha and some other architectures.  */
 220     double factor = 1000.0 / USER_HZ;
 221     return ticks * factor + 0.5;
 222 #endif
 223 }
 224
 225 struct raw_process_info {
 226     unsigned long int vsz;      /* Virtual size, in kB. */
 227     unsigned long int rss;      /* Resident set size, in kB. */
 228     long long int uptime;       /* ms since started. */
 229     long long int cputime;      /* ms of CPU used during 'uptime'. */
 230     pid_t ppid;                 /* Parent. */
 231     char name[18];              /* Name (surrounded by parentheses). */
 232 };
 233
 234 static bool
 235 get_raw_process_info(pid_t pid, struct raw_process_info *raw)
 236 {
 237     unsigned long long int vsize, rss, start_time, utime, stime;
 238     long long int start_msec;
 239     unsigned long ppid;
 240     char file_name[128];
 241     FILE *stream;
 242     int n;
 243
 244     ovs_assert(LINUX);
 245
 246     sprintf(file_name, "/proc/%lu/stat", (unsigned long int) pid);
 247     stream = fopen(file_name, "r");
 248     if (!stream) {
 249         VLOG_ERR_ONCE("%s: open failed (%s)",
 250                       file_name, ovs_strerror(errno));
 251         return false;
 252     }
 253
 254     n = fscanf(stream,
 255                "%*d "           /* (1. pid) */
 256                "%17s "          /* 2. process name */
 257                "%*c "           /* (3. state) */
 258                "%lu "           /* 4. ppid */
 259                "%*d "           /* (5. pgid) */
 260                "%*d "           /* (6. sid) */
 261                "%*d "           /* (7. tty_nr) */
 262                "%*d "           /* (8. tty_pgrp) */
 263                "%*u "           /* (9. flags) */
 264                "%*u "           /* (10. min_flt) */
 265                "%*u "           /* (11. cmin_flt) */
 266                "%*u "           /* (12. maj_flt) */
 267                "%*u "           /* (13. cmaj_flt) */
 268                "%llu "          /* 14. utime */
 269                "%llu "          /* 15. stime */
 270                "%*d "           /* (16. cutime) */
 271                "%*d "           /* (17. cstime) */
 272                "%*d "           /* (18. priority) */
 273                "%*d "           /* (19. nice) */
 274                "%*d "           /* (20. num_threads) */
 275                "%*d "           /* (21. always 0) */
 276                "%llu "          /* 22. start_time */
 277                "%llu "          /* 23. vsize */
 278                "%llu "          /* 24. rss */
 279 #if 0
 280                /* These are here for documentation but #if'd out to save
 281                 * actually parsing them from the stream for no benefit. */
 282                "%*lu "          /* (25. rsslim) */
 283                "%*lu "          /* (26. start_code) */
 284                "%*lu "          /* (27. end_code) */
 285                "%*lu "          /* (28. start_stack) */
 286                "%*lu "          /* (29. esp) */
 287                "%*lu "          /* (30. eip) */
 288                "%*lu "          /* (31. pending signals) */
 289                "%*lu "          /* (32. blocked signals) */
 290                "%*lu "          /* (33. ignored signals) */
 291                "%*lu "          /* (34. caught signals) */
 292                "%*lu "          /* (35. whcan) */
 293                "%*lu "          /* (36. always 0) */
 294                "%*lu "          /* (37. always 0) */
 295                "%*d "           /* (38. exit_signal) */
 296                "%*d "           /* (39. task_cpu) */
 297                "%*u "           /* (40. rt_priority) */
 298                "%*u "           /* (41. policy) */
 299                "%*llu "         /* (42. blkio_ticks) */
 300                "%*lu "          /* (43. gtime) */
 301                "%*ld"           /* (44. cgtime) */
 302 #endif
 303                , raw->name, &ppid, &utime, &stime, &start_time, &vsize, &rss);
 304     fclose(stream);
 305     if (n != 7) {
 306         VLOG_ERR_ONCE("%s: fscanf failed", file_name);
 307         return false;
 308     }
 309
 310     start_msec = get_boot_time() + ticks_to_ms(start_time);
 311
 312     raw->vsz = vsize / 1024;
 313     raw->rss = rss * (getpagesize() / 1024);
 314     raw->uptime = time_wall_msec() - start_msec;
 315     raw->cputime = ticks_to_ms(utime + stime);
 316     raw->ppid = ppid;
 317
 318     return true;
 319 }
 320
 321 static int
 322 count_crashes(pid_t pid)
 323 {
 324     char file_name[128];
 325     const char *paren;
 326     char line[128];
 327     int crashes = 0;
 328     FILE *stream;
 329
 330     ovs_assert(LINUX);
 331
 332     sprintf(file_name, "/proc/%lu/cmdline", (unsigned long int) pid);
 333     stream = fopen(file_name, "r");
 334     if (!stream) {
 335         VLOG_WARN_ONCE("%s: open failed (%s)", file_name, ovs_strerror(errno));
 336         goto exit;
 337     }
 338
 339     if (!fgets(line, sizeof line, stream)) {
 340         VLOG_WARN_ONCE("%s: read failed (%s)", file_name,
 341                        feof(stream) ? "end of file" : ovs_strerror(errno));
 342         goto exit_close;
 343     }
 344
 345     paren = strchr(line, '(');
 346     if (paren) {
 347         int x;
 348         if (ovs_scan(paren + 1, "%d", &x)) {
 349             crashes = x;
 350         }
 351     }
 352
 353 exit_close:
 354     fclose(stream);
 355 exit:
 356     return crashes;
 357 }
 358
 359 struct process_info {
 360     unsigned long int vsz;      /* Virtual size, in kB. */
 361     unsigned long int rss;      /* Resident set size, in kB. */
 362     long long int booted;       /* ms since monitor started. */
 363     int crashes;                /* # of crashes (usually 0). */
 364     long long int uptime;       /* ms since last (re)started by monitor. */
 365     long long int cputime;      /* ms of CPU used during 'uptime'. */
 366 };
 367
 368 static bool
 369 get_process_info(pid_t pid, struct process_info *pinfo)
 370 {
 371     struct raw_process_info child;
 372
 373     ovs_assert(LINUX);
 374     if (!get_raw_process_info(pid, &child)) {
 375         return false;
 376     }
 377
 378     pinfo->vsz = child.vsz;
 379     pinfo->rss = child.rss;
 380     pinfo->booted = child.uptime;
 381     pinfo->crashes = 0;
 382     pinfo->uptime = child.uptime;
 383     pinfo->cputime = child.cputime;
 384
 385     if (child.ppid) {
 386         struct raw_process_info parent;
 387
 388         get_raw_process_info(child.ppid, &parent);
 389         if (!strcmp(child.name, parent.name)) {
 390             pinfo->booted = parent.uptime;
 391             pinfo->crashes = count_crashes(child.ppid);
 392         }
 393     }
 394
 395     return true;
 396 }
 397
 398 static void
 399 get_process_stats(struct smap *stats)
 400 {
 401     struct dirent *de;
 402     DIR *dir;
 403
 404     dir = opendir(ovs_rundir());
 405     if (!dir) {
 406         VLOG_ERR_ONCE("%s: open failed (%s)",
 407                       ovs_rundir(), ovs_strerror(errno));
 408         return;
 409     }
 410
 411     while ((de = readdir(dir)) != NULL) {
 412         struct process_info pinfo;
 413         char *file_name;
 414         char *extension;
 415         char *key;
 416         pid_t pid;
 417
 418 #ifdef _DIRENT_HAVE_D_TYPE
 419         if (de->d_type != DT_UNKNOWN && de->d_type != DT_REG) {
 420             continue;
 421         }
 422 #endif
 423
 424         extension = strrchr(de->d_name, '.');
 425         if (!extension || strcmp(extension, ".pid")) {
 426             continue;
 427         }
 428
 429         file_name = xasprintf("%s/%s", ovs_rundir(), de->d_name);
 430         pid = read_pidfile(file_name);
 431         free(file_name);
 432         if (pid < 0) {
 433             continue;
 434         }
 435
 436         key = xasprintf("process_%.*s",
 437                         (int) (extension - de->d_name), de->d_name);
 438         if (!smap_get(stats, key)) {
 439             if (LINUX && get_process_info(pid, &pinfo)) {
 440                 smap_add_format(stats, key, "%lu,%lu,%lld,%d,%lld,%lld",
 441                                 pinfo.vsz, pinfo.rss, pinfo.cputime,
 442                                 pinfo.crashes, pinfo.booted, pinfo.uptime);
 443             } else {
 444                 smap_add(stats, key, "");
 445             }
 446         }
 447         free(key);
 448     }
 449
 450     closedir(dir);
 451 }
 452
 453 static void
 454 get_filesys_stats(struct smap *stats OVS_UNUSED)
 455 {
 456 #if HAVE_GETMNTENT_R && HAVE_STATVFS
 457     static const char file_name[] = "/etc/mtab";
 458     struct mntent mntent;
 459     struct mntent *me;
 460     char buf[4096];
 461     FILE *stream;
 462     struct ds s;
 463
 464     stream = setmntent(file_name, "r");
 465     if (!stream) {
 466         VLOG_ERR_ONCE("%s: open failed (%s)", file_name, ovs_strerror(errno));
 467         return;
 468     }
 469
 470     ds_init(&s);
 471     while ((me = getmntent_r(stream, &mntent, buf, sizeof buf)) != NULL) {
 472         unsigned long long int total, free;
 473         struct statvfs vfs;
 474         char *p;
 475
 476         /* Skip non-local and read-only filesystems. */
 477         if (strncmp(me->mnt_fsname, "/dev", 4)
 478             || !strstr(me->mnt_opts, "rw")) {
 479             continue;
 480         }
 481
 482         /* Given the mount point we can stat the file system. */
 483         if (statvfs(me->mnt_dir, &vfs) && vfs.f_flag & ST_RDONLY) {
 484             /* That's odd... */
 485             continue;
 486         }
 487
 488         /* Now format the data. */
 489         if (s.length) {
 490             ds_put_char(&s, ' ');
 491         }
 492         for (p = me->mnt_dir; *p != '\0'; p++) {
 493             ds_put_char(&s, *p == ' ' || *p == ',' ? '_' : *p);
 494         }
 495         total = (unsigned long long int) vfs.f_frsize * vfs.f_blocks / 1024;
 496         free = (unsigned long long int) vfs.f_frsize * vfs.f_bfree / 1024;
 497         ds_put_format(&s, ",%llu,%llu", total, total - free);
 498     }
 499     endmntent(stream);
 500
 501     if (s.length) {
 502         smap_add(stats, "file_systems", ds_cstr(&s));
 503     }
 504     ds_destroy(&s);
 505 #endif  /* HAVE_GETMNTENT_R && HAVE_STATVFS */
 506 }
 507 \f
 508 #define SYSTEM_STATS_INTERVAL (5 * 1000) /* In milliseconds. */
 509
 510 static struct ovs_mutex mutex = OVS_MUTEX_INITIALIZER;
 511 static pthread_cond_t cond = PTHREAD_COND_INITIALIZER;
 512 static struct latch latch OVS_GUARDED_BY(mutex);
 513 static bool enabled;
 514 static bool started OVS_GUARDED_BY(mutex);
 515 static struct smap *system_stats OVS_GUARDED_BY(mutex);
 516
 517 static void *system_stats_thread_func(void *);
 518 static void discard_stats(void);
 519
 520 /* Enables or disables system stats collection, according to 'enable'. */
 521 void
 522 system_stats_enable(bool enable)
 523 {
 524     if (enabled != enable) {
 525         ovs_mutex_lock(&mutex);
 526         if (enable) {
 527             if (!started) {
 528                 xpthread_create(NULL, NULL, system_stats_thread_func, NULL);
 529                 latch_init(&latch);
 530                 started = true;
 531             }
 532             discard_stats();
 533             xpthread_cond_signal(&cond);
 534         }
 535         enabled = enable;
 536         ovs_mutex_unlock(&mutex);
 537     }
 538 }
 539
 540 /* Tries to obtain a new snapshot of system stats every SYSTEM_STATS_INTERVAL
 541  * milliseconds.
 542  *
 543  * When a new snapshot is available (which only occurs if system stats are
 544  * enabled), returns it as an smap owned by the caller.  The caller must use
 545  * both smap_destroy() and free() to completely free the returned data.
 546  *
 547  * When no new snapshot is available, returns NULL. */
 548 struct smap *
 549 system_stats_run(void)
 550 {
 551     struct smap *stats = NULL;
 552
 553     ovs_mutex_lock(&mutex);
 554     if (system_stats) {
 555         latch_poll(&latch);
 556
 557         if (enabled) {
 558             stats = system_stats;
 559             system_stats = NULL;
 560         } else {
 561             discard_stats();
 562         }
 563     }
 564     ovs_mutex_unlock(&mutex);
 565
 566     return stats;
 567 }
 568
 569 /* Causes poll_block() to wake up when system_stats_run() needs to be
 570  * called. */
 571 void
 572 system_stats_wait(void)
 573 {
 574     if (enabled) {
 575         latch_wait(&latch);
 576     }
 577 }
 578
 579 static void
 580 discard_stats(void) OVS_REQUIRES(mutex)
 581 {
 582     if (system_stats) {
 583         smap_destroy(system_stats);
 584         free(system_stats);
 585         system_stats = NULL;
 586     }
 587 }
 588
 589 static void * NO_RETURN
 590 system_stats_thread_func(void *arg OVS_UNUSED)
 591 {
 592     pthread_detach(pthread_self());
 593
 594     for (;;) {
 595         long long int next_refresh;
 596         struct smap *stats;
 597
 598         ovs_mutex_lock(&mutex);
 599         while (!enabled) {
 600             ovs_mutex_cond_wait(&cond, &mutex);
 601         }
 602         ovs_mutex_unlock(&mutex);
 603
 604         stats = xmalloc(sizeof *stats);
 605         smap_init(stats);
 606         get_cpu_cores(stats);
 607         get_load_average(stats);
 608         get_memory_stats(stats);
 609         get_process_stats(stats);
 610         get_filesys_stats(stats);
 611
 612         ovs_mutex_lock(&mutex);
 613         discard_stats();
 614         system_stats = stats;
 615         latch_set(&latch);
 616         ovs_mutex_unlock(&mutex);
 617
 618         next_refresh = time_msec() + SYSTEM_STATS_INTERVAL;
 619         do {
 620             poll_timer_wait_until(next_refresh);
 621             poll_block();
 622         } while (time_msec() < next_refresh);
 623     }
 624 }