vswitchd/system-stats.c

   1 /* Copyright (c) 2010, 2012 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 <assert.h>
  21 #include <ctype.h>
  22 #include <dirent.h>
  23 #include <errno.h>
  24 #if HAVE_MNTENT_H
  25 #include <mntent.h>
  26 #endif
  27 #include <stdint.h>
  28 #include <stdio.h>
  29 #include <stdlib.h>
  30 #if HAVE_SYS_STATVFS_H
  31 #include <sys/statvfs.h>
  32 #endif
  33 #include <unistd.h>
  34
  35 #include "daemon.h"
  36 #include "dirs.h"
  37 #include "dynamic-string.h"
  38 #include "shash.h"
  39 #include "smap.h"
  40 #include "timeval.h"
  41 #include "vlog.h"
  42
  43 VLOG_DEFINE_THIS_MODULE(system_stats);
  44
  45 /* #ifdefs make it a pain to maintain code: you have to try to build both ways.
  46  * Thus, this file tries to compile as much of the code as possible regardless
  47  * of the target, by writing "if (LINUX)" instead of "#ifdef __linux__" where
  48  * this is possible. */
  49 #ifdef __linux__
  50 #include <asm/param.h>
  51 #define LINUX 1
  52 #else
  53 #define LINUX 0
  54 #endif
  55
  56 static void
  57 get_cpu_cores(struct smap *stats)
  58 {
  59     long int n_cores = sysconf(_SC_NPROCESSORS_ONLN);
  60     if (n_cores > 0) {
  61         smap_add_format(stats, "cpu", "%ld", n_cores);
  62     }
  63 }
  64
  65 static void
  66 get_load_average(struct smap *stats OVS_UNUSED)
  67 {
  68 #if HAVE_GETLOADAVG
  69     double loadavg[3];
  70
  71     if (getloadavg(loadavg, 3) == 3) {
  72         smap_add_format(stats, "load_average", "%.2f,%.2f,%.2f",
  73                         loadavg[0], loadavg[1], loadavg[2]);
  74     }
  75 #endif
  76 }
  77
  78 static unsigned int
  79 get_page_size(void)
  80 {
  81     static unsigned int cached;
  82
  83     if (!cached) {
  84         long int value = sysconf(_SC_PAGESIZE);
  85         if (value >= 0) {
  86             cached = value;
  87         }
  88     }
  89
  90     return cached;
  91 }
  92
  93 static void
  94 get_memory_stats(struct smap *stats)
  95 {
  96     if (!LINUX) {
  97         unsigned int pagesize = get_page_size();
  98         long int phys_pages = sysconf(_SC_PHYS_PAGES);
  99 #ifdef _SC_AVPHYS_PAGES
 100         long int avphys_pages = sysconf(_SC_AVPHYS_PAGES);
 101 #else
 102         long int avphys_pages = 0;
 103 #endif
 104         int mem_total, mem_used;
 105
 106         if (pagesize <= 0 || phys_pages <= 0 || avphys_pages <= 0) {
 107             return;
 108         }
 109
 110         mem_total = phys_pages * (pagesize / 1024);
 111         mem_used = (phys_pages - avphys_pages) * (pagesize / 1024);
 112         smap_add_format(stats, "memory", "%d,%d", mem_total, mem_used);
 113     } else {
 114         static const char file_name[] = "/proc/meminfo";
 115         int mem_used, mem_cache, swap_used;
 116         int mem_free = 0;
 117         int buffers = 0;
 118         int cached = 0;
 119         int swap_free = 0;
 120         int mem_total = 0;
 121         int swap_total = 0;
 122         struct shash dict;
 123         char line[128];
 124         FILE *stream;
 125
 126         stream = fopen(file_name, "r");
 127         if (!stream) {
 128             VLOG_WARN_ONCE("%s: open failed (%s)", file_name, strerror(errno));
 129             return;
 130         }
 131
 132         shash_init(&dict);
 133         shash_add(&dict, "MemTotal", &mem_total);
 134         shash_add(&dict, "MemFree", &mem_free);
 135         shash_add(&dict, "Buffers", &buffers);
 136         shash_add(&dict, "Cached", &cached);
 137         shash_add(&dict, "SwapTotal", &swap_total);
 138         shash_add(&dict, "SwapFree", &swap_free);
 139         while (fgets(line, sizeof line, stream)) {
 140             char key[16];
 141             int value;
 142
 143             if (sscanf(line, "%15[^:]: %u", key, &value) == 2) {
 144                 int *valuep = shash_find_data(&dict, key);
 145                 if (valuep) {
 146                     *valuep = value;
 147                 }
 148             }
 149         }
 150         fclose(stream);
 151         shash_destroy(&dict);
 152
 153         mem_used = mem_total - mem_free;
 154         mem_cache = buffers + cached;
 155         swap_used = swap_total - swap_free;
 156         smap_add_format(stats, "memory", "%d,%d,%d,%d,%d",
 157                         mem_total, mem_used, mem_cache, swap_total, swap_used);
 158     }
 159 }
 160
 161 /* Returns the time at which the system booted, as the number of milliseconds
 162  * since the epoch, or 0 if the time of boot cannot be determined. */
 163 static long long int
 164 get_boot_time(void)
 165 {
 166     static long long int cache_expiration = LLONG_MIN;
 167     static long long int boot_time;
 168
 169     assert(LINUX);
 170
 171     if (time_msec() >= cache_expiration) {
 172         static const char stat_file[] = "/proc/stat";
 173         char line[128];
 174         FILE *stream;
 175
 176         cache_expiration = time_msec() + 5 * 1000;
 177
 178         stream = fopen(stat_file, "r");
 179         if (!stream) {
 180             VLOG_ERR_ONCE("%s: open failed (%s)", stat_file, strerror(errno));
 181             return boot_time;
 182         }
 183
 184         while (fgets(line, sizeof line, stream)) {
 185             long long int btime;
 186             if (sscanf(line, "btime %lld", &btime) == 1) {
 187                 boot_time = btime * 1000;
 188                 goto done;
 189             }
 190         }
 191         VLOG_ERR_ONCE("%s: btime not found", stat_file);
 192     done:
 193         fclose(stream);
 194     }
 195     return boot_time;
 196 }
 197
 198 static unsigned long long int
 199 ticks_to_ms(unsigned long long int ticks)
 200 {
 201     assert(LINUX);
 202
 203 #ifndef USER_HZ
 204 #define USER_HZ 100
 205 #endif
 206
 207 #if USER_HZ == 100              /* Common case. */
 208     return ticks * (1000 / USER_HZ);
 209 #else  /* Alpha and some other architectures.  */
 210     double factor = 1000.0 / USER_HZ;
 211     return ticks * factor + 0.5;
 212 #endif
 213 }
 214
 215 struct raw_process_info {
 216     unsigned long int vsz;      /* Virtual size, in kB. */
 217     unsigned long int rss;      /* Resident set size, in kB. */
 218     long long int uptime;       /* ms since started. */
 219     long long int cputime;      /* ms of CPU used during 'uptime'. */
 220     pid_t ppid;                 /* Parent. */
 221     char name[18];              /* Name (surrounded by parentheses). */
 222 };
 223
 224 static bool
 225 get_raw_process_info(pid_t pid, struct raw_process_info *raw)
 226 {
 227     unsigned long long int vsize, rss, start_time, utime, stime;
 228     long long int start_msec;
 229     unsigned long ppid;
 230     char file_name[128];
 231     FILE *stream;
 232     int n;
 233
 234     assert(LINUX);
 235
 236     sprintf(file_name, "/proc/%lu/stat", (unsigned long int) pid);
 237     stream = fopen(file_name, "r");
 238     if (!stream) {
 239         VLOG_ERR_ONCE("%s: open failed (%s)", file_name, strerror(errno));
 240         return false;
 241     }
 242
 243     n = fscanf(stream,
 244                "%*d "           /* (1. pid) */
 245                "%17s "          /* 2. process name */
 246                "%*c "           /* (3. state) */
 247                "%lu "           /* 4. ppid */
 248                "%*d "           /* (5. pgid) */
 249                "%*d "           /* (6. sid) */
 250                "%*d "           /* (7. tty_nr) */
 251                "%*d "           /* (8. tty_pgrp) */
 252                "%*u "           /* (9. flags) */
 253                "%*u "           /* (10. min_flt) */
 254                "%*u "           /* (11. cmin_flt) */
 255                "%*u "           /* (12. maj_flt) */
 256                "%*u "           /* (13. cmaj_flt) */
 257                "%llu "          /* 14. utime */
 258                "%llu "          /* 15. stime */
 259                "%*d "           /* (16. cutime) */
 260                "%*d "           /* (17. cstime) */
 261                "%*d "           /* (18. priority) */
 262                "%*d "           /* (19. nice) */
 263                "%*d "           /* (20. num_threads) */
 264                "%*d "           /* (21. always 0) */
 265                "%llu "          /* 22. start_time */
 266                "%llu "          /* 23. vsize */
 267                "%llu "          /* 24. rss */
 268 #if 0
 269                /* These are here for documentation but #if'd out to save
 270                 * actually parsing them from the stream for no benefit. */
 271                "%*lu "          /* (25. rsslim) */
 272                "%*lu "          /* (26. start_code) */
 273                "%*lu "          /* (27. end_code) */
 274                "%*lu "          /* (28. start_stack) */
 275                "%*lu "          /* (29. esp) */
 276                "%*lu "          /* (30. eip) */
 277                "%*lu "          /* (31. pending signals) */
 278                "%*lu "          /* (32. blocked signals) */
 279                "%*lu "          /* (33. ignored signals) */
 280                "%*lu "          /* (34. caught signals) */
 281                "%*lu "          /* (35. whcan) */
 282                "%*lu "          /* (36. always 0) */
 283                "%*lu "          /* (37. always 0) */
 284                "%*d "           /* (38. exit_signal) */
 285                "%*d "           /* (39. task_cpu) */
 286                "%*u "           /* (40. rt_priority) */
 287                "%*u "           /* (41. policy) */
 288                "%*llu "         /* (42. blkio_ticks) */
 289                "%*lu "          /* (43. gtime) */
 290                "%*ld"           /* (44. cgtime) */
 291 #endif
 292                , raw->name, &ppid, &utime, &stime, &start_time, &vsize, &rss);
 293     fclose(stream);
 294     if (n != 7) {
 295         VLOG_ERR_ONCE("%s: fscanf failed", file_name);
 296         return false;
 297     }
 298
 299     start_msec = get_boot_time() + ticks_to_ms(start_time);
 300
 301     raw->vsz = vsize / 1024;
 302     raw->rss = rss * (getpagesize() / 1024);
 303     raw->uptime = time_wall_msec() - start_msec;
 304     raw->cputime = ticks_to_ms(utime + stime);
 305     raw->ppid = ppid;
 306
 307     return true;
 308 }
 309
 310 static int
 311 count_crashes(pid_t pid)
 312 {
 313     char file_name[128];
 314     const char *paren;
 315     char line[128];
 316     int crashes = 0;
 317     FILE *stream;
 318
 319     assert(LINUX);
 320
 321     sprintf(file_name, "/proc/%lu/cmdline", (unsigned long int) pid);
 322     stream = fopen(file_name, "r");
 323     if (!stream) {
 324         VLOG_WARN_ONCE("%s: open failed (%s)", file_name, strerror(errno));
 325         goto exit;
 326     }
 327
 328     if (!fgets(line, sizeof line, stream)) {
 329         VLOG_WARN_ONCE("%s: read failed (%s)", file_name,
 330                        feof(stream) ? "end of file" : strerror(errno));
 331         goto exit_close;
 332     }
 333
 334     paren = strchr(line, '(');
 335     if (paren) {
 336         int x;
 337         if (sscanf(paren + 1, "%d", &x) == 1) {
 338             crashes = x;
 339         }
 340     }
 341
 342 exit_close:
 343     fclose(stream);
 344 exit:
 345     return crashes;
 346 }
 347
 348 struct process_info {
 349     unsigned long int vsz;      /* Virtual size, in kB. */
 350     unsigned long int rss;      /* Resident set size, in kB. */
 351     long long int booted;       /* ms since monitor started. */
 352     int crashes;                /* # of crashes (usually 0). */
 353     long long int uptime;       /* ms since last (re)started by monitor. */
 354     long long int cputime;      /* ms of CPU used during 'uptime'. */
 355 };
 356
 357 static bool
 358 get_process_info(pid_t pid, struct process_info *pinfo)
 359 {
 360     struct raw_process_info child;
 361
 362     assert(LINUX);
 363     if (!get_raw_process_info(pid, &child)) {
 364         return false;
 365     }
 366
 367     pinfo->vsz = child.vsz;
 368     pinfo->rss = child.rss;
 369     pinfo->booted = child.uptime;
 370     pinfo->crashes = 0;
 371     pinfo->uptime = child.uptime;
 372     pinfo->cputime = child.cputime;
 373
 374     if (child.ppid) {
 375         struct raw_process_info parent;
 376
 377         get_raw_process_info(child.ppid, &parent);
 378         if (!strcmp(child.name, parent.name)) {
 379             pinfo->booted = parent.uptime;
 380             pinfo->crashes = count_crashes(child.ppid);
 381         }
 382     }
 383
 384     return true;
 385 }
 386
 387 static void
 388 get_process_stats(struct smap *stats)
 389 {
 390     struct dirent *de;
 391     DIR *dir;
 392
 393     dir = opendir(ovs_rundir());
 394     if (!dir) {
 395         VLOG_ERR_ONCE("%s: open failed (%s)", ovs_rundir(), strerror(errno));
 396         return;
 397     }
 398
 399     while ((de = readdir(dir)) != NULL) {
 400         struct process_info pinfo;
 401         char *file_name;
 402         char *extension;
 403         char *key;
 404         pid_t pid;
 405
 406 #ifdef _DIRENT_HAVE_D_TYPE
 407         if (de->d_type != DT_UNKNOWN && de->d_type != DT_REG) {
 408             continue;
 409         }
 410 #endif
 411
 412         extension = strrchr(de->d_name, '.');
 413         if (!extension || strcmp(extension, ".pid")) {
 414             continue;
 415         }
 416
 417         file_name = xasprintf("%s/%s", ovs_rundir(), de->d_name);
 418         pid = read_pidfile(file_name);
 419         free(file_name);
 420         if (pid < 0) {
 421             continue;
 422         }
 423
 424         key = xasprintf("process_%.*s",
 425                         (int) (extension - de->d_name), de->d_name);
 426         if (!smap_get(stats, key)) {
 427             if (LINUX && get_process_info(pid, &pinfo)) {
 428                 smap_add_format(stats, key, "%lu,%lu,%lld,%d,%lld,%lld",
 429                                 pinfo.vsz, pinfo.rss, pinfo.cputime,
 430                                 pinfo.crashes, pinfo.booted, pinfo.uptime);
 431             } else {
 432                 smap_add(stats, key, "");
 433             }
 434         }
 435         free(key);
 436     }
 437
 438     closedir(dir);
 439 }
 440
 441 static void
 442 get_filesys_stats(struct smap *stats OVS_UNUSED)
 443 {
 444 #if HAVE_SETMNTENT && HAVE_STATVFS
 445     static const char file_name[] = "/etc/mtab";
 446     struct mntent *me;
 447     FILE *stream;
 448     struct ds s;
 449
 450     stream = setmntent(file_name, "r");
 451     if (!stream) {
 452         VLOG_ERR_ONCE("%s: open failed (%s)", file_name, strerror(errno));
 453         return;
 454     }
 455
 456     ds_init(&s);
 457     while ((me = getmntent(stream)) != NULL) {
 458         unsigned long long int total, free;
 459         struct statvfs vfs;
 460         char *p;
 461
 462         /* Skip non-local and read-only filesystems. */
 463         if (strncmp(me->mnt_fsname, "/dev", 4)
 464             || !strstr(me->mnt_opts, "rw")) {
 465             continue;
 466         }
 467
 468         /* Given the mount point we can stat the file system. */
 469         if (statvfs(me->mnt_dir, &vfs) && vfs.f_flag & ST_RDONLY) {
 470             /* That's odd... */
 471             continue;
 472         }
 473
 474         /* Now format the data. */
 475         if (s.length) {
 476             ds_put_char(&s, ' ');
 477         }
 478         for (p = me->mnt_dir; *p != '\0'; p++) {
 479             ds_put_char(&s, *p == ' ' || *p == ',' ? '_' : *p);
 480         }
 481         total = (unsigned long long int) vfs.f_frsize * vfs.f_blocks / 1024;
 482         free = (unsigned long long int) vfs.f_frsize * vfs.f_bfree / 1024;
 483         ds_put_format(&s, ",%llu,%llu", total, total - free);
 484     }
 485     endmntent(stream);
 486
 487     if (s.length) {
 488         smap_add(stats, "file_systems", ds_cstr(&s));
 489     }
 490     ds_destroy(&s);
 491 #endif  /* HAVE_SETMNTENT && HAVE_STATVFS */
 492 }
 493
 494 void
 495 get_system_stats(struct smap *stats)
 496 {
 497     get_cpu_cores(stats);
 498     get_load_average(stats);
 499     get_memory_stats(stats);
 500     get_process_stats(stats);
 501     get_filesys_stats(stats);
 502 }