vswitchd/system-stats.c

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