src/proc_diskstats.c

   1 #include "common.h"
   2
   3 #define RRD_TYPE_DISK "disk"
   4
   5 #define DISK_TYPE_PHYSICAL  1
   6 #define DISK_TYPE_PARTITION 2
   7 #define DISK_TYPE_CONTAINER 3
   8
   9 #ifndef NETDATA_RELOAD_MOUNTINFO_EVERY
  10 #define NETDATA_RELOAD_MOUNTINFO_EVERY 10
  11 #endif
  12
  13 static struct disk {
  14     char *disk;             // the name of the disk (sda, sdb, etc)
  15     unsigned long major;
  16     unsigned long minor;
  17     int sector_size;
  18     int type;
  19
  20     char *mount_point;
  21
  22     // disk options caching
  23     int configured;
  24     int do_io;
  25     int do_ops;
  26     int do_mops;
  27     int do_iotime;
  28     int do_qops;
  29     int do_util;
  30     int do_backlog;
  31
  32     struct disk *next;
  33 } *disk_root = NULL;
  34
  35 static struct mountinfo *disk_mountinfo_root = NULL;
  36
  37 static inline void mountinfo_reload(int force) {
  38     static time_t last_loaded = 0;
  39     time_t now = now_realtime_sec();
  40
  41     if(force || now - last_loaded >= NETDATA_RELOAD_MOUNTINFO_EVERY) {
  42 //#ifdef NETDATA_INTERNAL_CHECKS
  43 //        info("Reloading mountinfo");
  44 //#endif
  45
  46         // mountinfo_free() can be called with NULL disk_mountinfo_root
  47         mountinfo_free(disk_mountinfo_root);
  48
  49         // re-read mountinfo in case something changed
  50         disk_mountinfo_root = mountinfo_read();
  51
  52         last_loaded = now;
  53     }
  54 }
  55
  56
  57 // linked list of mount points that are by default disabled
  58 static struct excluded_mount_point {
  59     const char *prefix;
  60     size_t len;
  61     struct excluded_mount_point *next;
  62 } *excluded_mount_points = NULL;
  63
  64 static inline int is_mount_point_excluded(const char *mount_point) {
  65     static int initialized = 0;
  66
  67     if(unlikely(!initialized)) {
  68         initialized = 1;
  69
  70         char *a = config_get("plugin:proc:/proc/diskstats", "exclude space metrics on paths", "/proc/ /sys/ /var/run/user/ /run/user/");
  71         if(a && *a) {
  72             char *s = a;
  73
  74             while(s && *s) {
  75                 // skip all spaces
  76                 while(isspace(*s)) s++;
  77
  78                 // empty string
  79                 if(unlikely(!*s)) break;
  80
  81                 // find the next space
  82                 char *c = s;
  83                 while(*c && !isspace(*c)) c++;
  84
  85                 char *n;
  86                 if(likely(*c)) n = c + 1;
  87                 else n = NULL;
  88
  89                 // terminate our string
  90                 *c = '\0';
  91
  92                 // allocate the structure
  93                 struct excluded_mount_point *m = mallocz(sizeof(struct excluded_mount_point));
  94                 m->prefix = strdup(s);
  95                 m->len = strlen(m->prefix);
  96                 m->next = excluded_mount_points;
  97                 excluded_mount_points = m;
  98
  99                 // prepare for next loop
 100                 s = n;
 101                 if(likely(n)) *c = ' ';
 102             }
 103         }
 104     }
 105
 106     size_t len = strlen(mount_point);
 107     struct excluded_mount_point *m;
 108     for(m = excluded_mount_points; m ; m = m->next) {
 109         if(m->len <= len) {
 110             // fprintf(stderr, "SPACE: comparing '%s' with '%s'\n", mount_point, m->prefix);
 111             if(unlikely(strncmp(m->prefix, mount_point, m->len) == 0)) {
 112                 // fprintf(stderr, "SPACE: excluded '%s'\n", mount_point);
 113                 return 1;
 114             }
 115         }
 116     }
 117
 118     // fprintf(stderr, "SPACE: included '%s'\n", mount_point);
 119     return 0;
 120 }
 121
 122 // Data to be stored in DICTIONARY mount_points used by do_disk_space_stats().
 123 // This DICTIONARY is used to lookup the settings of the mount point on each iteration.
 124 struct mount_point_metadata {
 125     int do_space;
 126     int do_inodes;
 127 };
 128
 129 static inline void do_disk_space_stats(struct mountinfo *mi, int update_every, usec_t dt) {
 130     (void)dt;
 131
 132     const char *family = mi->mount_point;
 133     const char *disk = mi->persistent_id;
 134
 135     static DICTIONARY *mount_points = NULL;
 136     int do_space, do_inodes;
 137
 138     if(unlikely(!mount_points)) {
 139         mount_points = dictionary_create(DICTIONARY_FLAG_SINGLE_THREADED);
 140     }
 141
 142     struct mount_point_metadata *m = dictionary_get(mount_points, mi->mount_point);
 143     if(unlikely(!m)) {
 144         char var_name[4096 + 1];
 145         snprintfz(var_name, 4096, "plugin:proc:/proc/diskstats:%s", mi->mount_point);
 146
 147         int def_space = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "space usage for all disks", CONFIG_ONDEMAND_ONDEMAND);
 148         int def_inodes = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "inodes usage for all disks", CONFIG_ONDEMAND_ONDEMAND);
 149
 150         if(unlikely(is_mount_point_excluded(mi->mount_point))) {
 151             def_space = CONFIG_ONDEMAND_NO;
 152             def_inodes = CONFIG_ONDEMAND_NO;
 153         }
 154
 155         do_space = config_get_boolean_ondemand(var_name, "space usage", def_space);
 156         do_inodes = config_get_boolean_ondemand(var_name, "inodes usage", def_inodes);
 157
 158         struct mount_point_metadata mp = {
 159             .do_space = do_space,
 160             .do_inodes = do_inodes
 161         };
 162
 163         dictionary_set(mount_points, mi->mount_point, &mp, sizeof(struct mount_point_metadata));
 164     }
 165     else {
 166         do_space = m->do_space;
 167         do_inodes = m->do_inodes;
 168     }
 169
 170     if(unlikely(do_space == CONFIG_ONDEMAND_NO && do_inodes == CONFIG_ONDEMAND_NO))
 171         return;
 172
 173     struct statvfs buff_statvfs;
 174     if (statvfs(mi->mount_point, &buff_statvfs) < 0) {
 175         error("Failed statvfs() for '%s' (disk '%s')", mi->mount_point, disk);
 176         return;
 177     }
 178
 179     // taken from get_fs_usage() found in coreutils
 180     unsigned long bsize = (buff_statvfs.f_frsize) ? buff_statvfs.f_frsize : buff_statvfs.f_bsize;
 181
 182     fsblkcnt_t bavail         = buff_statvfs.f_bavail;
 183     fsblkcnt_t btotal         = buff_statvfs.f_blocks;
 184     fsblkcnt_t bavail_root    = buff_statvfs.f_bfree;
 185     fsblkcnt_t breserved_root = bavail_root - bavail;
 186     fsblkcnt_t bused;
 187     if(likely(btotal >= bavail_root))
 188         bused = btotal - bavail_root;
 189     else
 190         bused = bavail_root - btotal;
 191
 192 #ifdef NETDATA_INTERNAL_CHECKS
 193     if(unlikely(btotal != bavail + breserved_root + bused))
 194         error("Disk block statistics for '%s' (disk '%s') do not sum up: total = %llu, available = %llu, reserved = %llu, used = %llu", mi->mount_point, disk, (unsigned long long)btotal, (unsigned long long)bavail, (unsigned long long)breserved_root, (unsigned long long)bused);
 195 #endif
 196
 197     // --------------------------------------------------------------------------
 198
 199     fsfilcnt_t favail         = buff_statvfs.f_favail;
 200     fsfilcnt_t ftotal         = buff_statvfs.f_files;
 201     fsfilcnt_t favail_root    = buff_statvfs.f_ffree;
 202     fsfilcnt_t freserved_root = favail_root - favail;
 203     fsfilcnt_t fused          = ftotal - favail_root;
 204
 205 #ifdef NETDATA_INTERNAL_CHECKS
 206     if(unlikely(btotal != bavail + breserved_root + bused))
 207         error("Disk inode statistics for '%s' (disk '%s') do not sum up: total = %llu, available = %llu, reserved = %llu, used = %llu", mi->mount_point, disk, (unsigned long long)ftotal, (unsigned long long)favail, (unsigned long long)freserved_root, (unsigned long long)fused);
 208 #endif
 209
 210     // --------------------------------------------------------------------------
 211
 212     RRDSET *st;
 213
 214     if(do_space == CONFIG_ONDEMAND_YES || (do_space == CONFIG_ONDEMAND_ONDEMAND && (bavail || breserved_root || bused))) {
 215         st = rrdset_find_bytype("disk_space", disk);
 216         if(unlikely(!st)) {
 217             char title[4096 + 1];
 218             snprintfz(title, 4096, "Disk Space Usage for %s [%s]", family, mi->mount_source);
 219             st = rrdset_create("disk_space", disk, NULL, family, "disk.space", title, "GB", 2023, update_every, RRDSET_TYPE_STACKED);
 220
 221             rrddim_add(st, "avail", NULL, bsize, 1024*1024*1024, RRDDIM_ABSOLUTE);
 222             rrddim_add(st, "used" , NULL, bsize, 1024*1024*1024, RRDDIM_ABSOLUTE);
 223             rrddim_add(st, "reserved_for_root", "reserved for root", bsize, 1024*1024*1024, RRDDIM_ABSOLUTE);
 224         }
 225         else rrdset_next(st);
 226
 227         rrddim_set(st, "avail", (collected_number)bavail);
 228         rrddim_set(st, "used", (collected_number)bused);
 229         rrddim_set(st, "reserved_for_root", (collected_number)breserved_root);
 230         rrdset_done(st);
 231     }
 232
 233     // --------------------------------------------------------------------------
 234
 235     if(do_inodes == CONFIG_ONDEMAND_YES || (do_inodes == CONFIG_ONDEMAND_ONDEMAND && (favail || freserved_root || fused))) {
 236         st = rrdset_find_bytype("disk_inodes", disk);
 237         if(unlikely(!st)) {
 238             char title[4096 + 1];
 239             snprintfz(title, 4096, "Disk Files (inodes) Usage for %s [%s]", family, mi->mount_source);
 240             st = rrdset_create("disk_inodes", disk, NULL, family, "disk.inodes", title, "Inodes", 2024, update_every, RRDSET_TYPE_STACKED);
 241
 242             rrddim_add(st, "avail", NULL, 1, 1, RRDDIM_ABSOLUTE);
 243             rrddim_add(st, "used" , NULL, 1, 1, RRDDIM_ABSOLUTE);
 244             rrddim_add(st, "reserved_for_root", "reserved for root", 1, 1, RRDDIM_ABSOLUTE);
 245         }
 246         else rrdset_next(st);
 247
 248         rrddim_set(st, "avail", (collected_number)favail);
 249         rrddim_set(st, "used", (collected_number)fused);
 250         rrddim_set(st, "reserved_for_root", (collected_number)freserved_root);
 251         rrdset_done(st);
 252     }
 253 }
 254
 255 static struct disk *get_disk(unsigned long major, unsigned long minor, char *disk) {
 256     static char path_to_get_hw_sector_size[FILENAME_MAX + 1] = "";
 257     static char path_to_get_hw_sector_size_partitions[FILENAME_MAX + 1] = "";
 258     static char path_find_block_device[FILENAME_MAX + 1] = "";
 259     struct disk *d;
 260
 261     // search for it in our RAM list.
 262     // this is sequential, but since we just walk through
 263     // and the number of disks / partitions in a system
 264     // should not be that many, it should be acceptable
 265     for(d = disk_root; d ; d = d->next)
 266         if(unlikely(d->major == major && d->minor == minor))
 267             break;
 268
 269     // if we found it, return it
 270     if(likely(d))
 271         return d;
 272
 273     // not found
 274     // create a new disk structure
 275     d = (struct disk *)mallocz(sizeof(struct disk));
 276
 277     d->disk = strdupz(disk);
 278     d->major = major;
 279     d->minor = minor;
 280     d->type = DISK_TYPE_PHYSICAL; // Default type. Changed later if not correct.
 281     d->configured = 0;
 282     d->sector_size = 512; // the default, will be changed below
 283     d->next = NULL;
 284
 285     // append it to the list
 286     if(unlikely(!disk_root))
 287         disk_root = d;
 288     else {
 289         struct disk *last;
 290         for(last = disk_root; last->next ;last = last->next);
 291         last->next = d;
 292     }
 293
 294     // ------------------------------------------------------------------------
 295     // find the type of the device
 296
 297     char buffer[FILENAME_MAX + 1];
 298
 299     // get the default path for finding info about the block device
 300     if(unlikely(!path_find_block_device[0])) {
 301         snprintfz(buffer, FILENAME_MAX, "%s%s", global_host_prefix, "/sys/dev/block/%lu:%lu/%s");
 302         snprintfz(path_find_block_device, FILENAME_MAX, "%s", config_get("plugin:proc:/proc/diskstats", "path to get block device infos", buffer));
 303     }
 304
 305     // find if it is a partition
 306     // by checking if /sys/dev/block/MAJOR:MINOR/partition is readable.
 307     snprintfz(buffer, FILENAME_MAX, path_find_block_device, major, minor, "partition");
 308     if(likely(access(buffer, R_OK) == 0)) {
 309         d->type = DISK_TYPE_PARTITION;
 310     }
 311     else {
 312         // find if it is a container
 313         // by checking if /sys/dev/block/MAJOR:MINOR/slaves has entries
 314         snprintfz(buffer, FILENAME_MAX, path_find_block_device, major, minor, "slaves/");
 315         DIR *dirp = opendir(buffer);
 316         if(likely(dirp != NULL)) {
 317             struct dirent *dp;
 318             while( (dp = readdir(dirp)) ) {
 319                 // . and .. are also files in empty folders.
 320                 if(unlikely(strcmp(dp->d_name, ".") == 0 || strcmp(dp->d_name, "..") == 0)) {
 321                     continue;
 322                 }
 323
 324                 d->type = DISK_TYPE_CONTAINER;
 325
 326                 // Stop the loop after we found one file.
 327                 break;
 328             }
 329             if(unlikely(closedir(dirp) == -1))
 330                 error("Unable to close dir %s", buffer);
 331         }
 332     }
 333
 334     // ------------------------------------------------------------------------
 335     // check if we can find its mount point
 336
 337     // mountinfo_find() can be called with NULL disk_mountinfo_root
 338     struct mountinfo *mi = mountinfo_find(disk_mountinfo_root, d->major, d->minor);
 339     if(unlikely(mi))
 340         d->mount_point = strdupz(mi->mount_point);
 341     else
 342         d->mount_point = NULL;
 343
 344     // ------------------------------------------------------------------------
 345     // find the disk sector size
 346
 347     if(unlikely(!path_to_get_hw_sector_size[0])) {
 348         snprintfz(buffer, FILENAME_MAX, "%s%s", global_host_prefix, "/sys/block/%s/queue/hw_sector_size");
 349         snprintfz(path_to_get_hw_sector_size, FILENAME_MAX, "%s", config_get("plugin:proc:/proc/diskstats", "path to get h/w sector size", buffer));
 350     }
 351     if(unlikely(!path_to_get_hw_sector_size_partitions[0])) {
 352         snprintfz(buffer, FILENAME_MAX, "%s%s", global_host_prefix, "/sys/dev/block/%lu:%lu/subsystem/%s/../queue/hw_sector_size");
 353         snprintfz(path_to_get_hw_sector_size_partitions, FILENAME_MAX, "%s", config_get("plugin:proc:/proc/diskstats", "path to get h/w sector size for partitions", buffer));
 354     }
 355
 356     {
 357         char tf[FILENAME_MAX + 1], *t;
 358         strncpyz(tf, d->disk, FILENAME_MAX);
 359
 360         // replace all / with !
 361         for(t = tf; *t ;t++)
 362             if(unlikely(*t == '/')) *t = '!';
 363
 364         if(likely(d->type == DISK_TYPE_PARTITION))
 365             snprintfz(buffer, FILENAME_MAX, path_to_get_hw_sector_size_partitions, d->major, d->minor, tf);
 366         else
 367             snprintfz(buffer, FILENAME_MAX, path_to_get_hw_sector_size, tf);
 368
 369         FILE *fpss = fopen(buffer, "r");
 370         if(likely(fpss)) {
 371             char buffer2[1024 + 1];
 372             char *tmp = fgets(buffer2, 1024, fpss);
 373
 374             if(likely(tmp)) {
 375                 d->sector_size = atoi(tmp);
 376                 if(unlikely(d->sector_size <= 0)) {
 377                     error("Invalid sector size %d for device %s in %s. Assuming 512.", d->sector_size, d->disk, buffer);
 378                     d->sector_size = 512;
 379                 }
 380             }
 381             else error("Cannot read data for sector size for device %s from %s. Assuming 512.", d->disk, buffer);
 382
 383             fclose(fpss);
 384         }
 385         else error("Cannot read sector size for device %s from %s. Assuming 512.", d->disk, buffer);
 386     }
 387
 388     return d;
 389 }
 390
 391 static inline int select_positive_option(int option1, int option2) {
 392     if(unlikely(option1 == CONFIG_ONDEMAND_YES || option2 == CONFIG_ONDEMAND_YES))
 393         return CONFIG_ONDEMAND_YES;
 394     else if(unlikely(option1 == CONFIG_ONDEMAND_ONDEMAND || option2 == CONFIG_ONDEMAND_ONDEMAND))
 395         return CONFIG_ONDEMAND_ONDEMAND;
 396
 397     return CONFIG_ONDEMAND_NO;
 398 }
 399
 400 static inline int is_major_enabled(int major) {
 401     static char *major_configs = NULL;
 402     static size_t major_size = 0;
 403
 404     if(major < 0) return 1;
 405
 406     size_t wanted_size = (size_t)major + 1;
 407
 408     if(major_size < wanted_size) {
 409         major_configs = reallocz(major_configs, wanted_size);
 410
 411         size_t i;
 412         for(i = major_size; i < wanted_size ; i++)
 413             major_configs[i] = -1;
 414
 415         major_size = wanted_size;
 416     }
 417
 418     if(major_configs[major] == -1) {
 419         char buffer[CONFIG_MAX_NAME + 1];
 420         snprintfz(buffer, CONFIG_MAX_NAME, "performance metrics for disks with major %d", major);
 421         major_configs[major] = (char)config_get_boolean("plugin:proc:/proc/diskstats", buffer, 1);
 422     }
 423
 424     return major_configs[major];
 425 }
 426
 427 int do_proc_diskstats(int update_every, usec_t dt) {
 428     (void)dt;
 429
 430     static procfile *ff = NULL;
 431     static int  global_enable_new_disks_detected_at_runtime = CONFIG_ONDEMAND_YES,
 432                 global_enable_performance_for_physical_disks = CONFIG_ONDEMAND_ONDEMAND,
 433                 global_enable_performance_for_virtual_disks = CONFIG_ONDEMAND_ONDEMAND,
 434                 global_enable_performance_for_partitions = CONFIG_ONDEMAND_NO,
 435                 global_do_io = CONFIG_ONDEMAND_ONDEMAND,
 436                 global_do_ops = CONFIG_ONDEMAND_ONDEMAND,
 437                 global_do_mops = CONFIG_ONDEMAND_ONDEMAND,
 438                 global_do_iotime = CONFIG_ONDEMAND_ONDEMAND,
 439                 global_do_qops = CONFIG_ONDEMAND_ONDEMAND,
 440                 global_do_util = CONFIG_ONDEMAND_ONDEMAND,
 441                 global_do_backlog = CONFIG_ONDEMAND_ONDEMAND,
 442                 globals_initialized = 0;
 443
 444     if(unlikely(!globals_initialized)) {
 445         global_enable_new_disks_detected_at_runtime = config_get_boolean("plugin:proc:/proc/diskstats", "enable new disks detected at runtime", global_enable_new_disks_detected_at_runtime);
 446
 447         global_enable_performance_for_physical_disks = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "performance metrics for physical disks", global_enable_performance_for_physical_disks);
 448         global_enable_performance_for_virtual_disks = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "performance metrics for virtual disks", global_enable_performance_for_virtual_disks);
 449         global_enable_performance_for_partitions = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "performance metrics for partitions", global_enable_performance_for_partitions);
 450
 451         global_do_io      = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "bandwidth for all disks", global_do_io);
 452         global_do_ops     = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "operations for all disks", global_do_ops);
 453         global_do_mops    = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "merged operations for all disks", global_do_mops);
 454         global_do_iotime  = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "i/o time for all disks", global_do_iotime);
 455         global_do_qops    = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "queued operations for all disks", global_do_qops);
 456         global_do_util    = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "utilization percentage for all disks", global_do_util);
 457         global_do_backlog = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "backlog for all disks", global_do_backlog);
 458
 459         globals_initialized = 1;
 460     }
 461
 462     // --------------------------------------------------------------------------
 463     // this is smart enough not to reload it every time
 464
 465     mountinfo_reload(0);
 466
 467     // --------------------------------------------------------------------------
 468     // disk space metrics
 469
 470     struct mountinfo *mi;
 471     for(mi = disk_mountinfo_root; mi ;mi = mi->next) {
 472         if(unlikely(mi->flags & (MOUNTINFO_IS_DUMMY|MOUNTINFO_IS_BIND|MOUNTINFO_IS_SAME_DEV|MOUNTINFO_NO_STAT|MOUNTINFO_NO_SIZE|MOUNTINFO_READONLY)))
 473             continue;
 474
 475         do_disk_space_stats(mi, update_every, dt);
 476     }
 477
 478     // --------------------------------------------------------------------------
 479
 480     if(unlikely(!ff)) {
 481         char filename[FILENAME_MAX + 1];
 482         snprintfz(filename, FILENAME_MAX, "%s%s", global_host_prefix, "/proc/diskstats");
 483         ff = procfile_open(config_get("plugin:proc:/proc/diskstats", "filename to monitor", filename), " \t", PROCFILE_FLAG_DEFAULT);
 484     }
 485     if(unlikely(!ff)) return 0;
 486
 487     ff = procfile_readall(ff);
 488     if(unlikely(!ff)) return 0; // we return 0, so that we will retry to open it next time
 489
 490     uint32_t lines = procfile_lines(ff), l;
 491
 492     for(l = 0; l < lines ;l++) {
 493         // --------------------------------------------------------------------------
 494         // Read parameters
 495
 496         char *disk;
 497         unsigned long       major = 0, minor = 0;
 498
 499         collected_number    reads = 0,  mreads = 0,  readsectors = 0,  readms = 0,
 500                             writes = 0, mwrites = 0, writesectors = 0, writems = 0,
 501                             queued_ios = 0, busy_ms = 0, backlog_ms = 0;
 502
 503         collected_number    last_reads = 0,  last_readsectors = 0,  last_readms = 0,
 504                             last_writes = 0, last_writesectors = 0, last_writems = 0,
 505                             last_busy_ms = 0;
 506
 507         uint32_t words = procfile_linewords(ff, l);
 508         if(unlikely(words < 14)) continue;
 509
 510         major           = strtoul(procfile_lineword(ff, l, 0), NULL, 10);
 511         minor           = strtoul(procfile_lineword(ff, l, 1), NULL, 10);
 512         disk            = procfile_lineword(ff, l, 2);
 513
 514         // # of reads completed # of writes completed
 515         // This is the total number of reads or writes completed successfully.
 516         reads           = strtoull(procfile_lineword(ff, l, 3), NULL, 10);  // rd_ios
 517         writes          = strtoull(procfile_lineword(ff, l, 7), NULL, 10);  // wr_ios
 518
 519         // # of reads merged # of writes merged
 520         // Reads and writes which are adjacent to each other may be merged for
 521         // efficiency.  Thus two 4K reads may become one 8K read before it is
 522         // ultimately handed to the disk, and so it will be counted (and queued)
 523         mreads          = strtoull(procfile_lineword(ff, l, 4), NULL, 10);  // rd_merges_or_rd_sec
 524         mwrites         = strtoull(procfile_lineword(ff, l, 8), NULL, 10);  // wr_merges
 525
 526         // # of sectors read # of sectors written
 527         // This is the total number of sectors read or written successfully.
 528         readsectors     = strtoull(procfile_lineword(ff, l, 5), NULL, 10);  // rd_sec_or_wr_ios
 529         writesectors    = strtoull(procfile_lineword(ff, l, 9), NULL, 10);  // wr_sec
 530
 531         // # of milliseconds spent reading # of milliseconds spent writing
 532         // This is the total number of milliseconds spent by all reads or writes (as
 533         // measured from __make_request() to end_that_request_last()).
 534         readms          = strtoull(procfile_lineword(ff, l, 6), NULL, 10);  // rd_ticks_or_wr_sec
 535         writems         = strtoull(procfile_lineword(ff, l, 10), NULL, 10); // wr_ticks
 536
 537         // # of I/Os currently in progress
 538         // The only field that should go to zero. Incremented as requests are
 539         // given to appropriate struct request_queue and decremented as they finish.
 540         queued_ios      = strtoull(procfile_lineword(ff, l, 11), NULL, 10); // ios_pgr
 541
 542         // # of milliseconds spent doing I/Os
 543         // This field increases so long as field queued_ios is nonzero.
 544         busy_ms         = strtoull(procfile_lineword(ff, l, 12), NULL, 10); // tot_ticks
 545
 546         // weighted # of milliseconds spent doing I/Os
 547         // This field is incremented at each I/O start, I/O completion, I/O
 548         // merge, or read of these stats by the number of I/Os in progress
 549         // (field queued_ios) times the number of milliseconds spent doing I/O since the
 550         // last update of this field.  This can provide an easy measure of both
 551         // I/O completion time and the backlog that may be accumulating.
 552         backlog_ms      = strtoull(procfile_lineword(ff, l, 13), NULL, 10); // rq_ticks
 553
 554
 555         // --------------------------------------------------------------------------
 556         // remove slashes from disk names
 557         char *s;
 558         for(s = disk; *s ;s++)
 559             if(*s == '/') *s = '_';
 560
 561         // --------------------------------------------------------------------------
 562         // get a disk structure for the disk
 563
 564         struct disk *d = get_disk(major, minor, disk);
 565
 566
 567         // --------------------------------------------------------------------------
 568         // Set its family based on mount point
 569
 570         char *family = d->mount_point;
 571         if(!family) family = disk;
 572
 573
 574         // --------------------------------------------------------------------------
 575         // Check the configuration for the device
 576
 577         if(unlikely(!d->configured)) {
 578             char var_name[4096 + 1];
 579             snprintfz(var_name, 4096, "plugin:proc:/proc/diskstats:%s", disk);
 580
 581             int def_enable = config_get_boolean_ondemand(var_name, "enable", global_enable_new_disks_detected_at_runtime);
 582             if(unlikely(def_enable == CONFIG_ONDEMAND_NO)) {
 583                 // the user does not want any metrics for this disk
 584                 d->do_io = CONFIG_ONDEMAND_NO;
 585                 d->do_ops = CONFIG_ONDEMAND_NO;
 586                 d->do_mops = CONFIG_ONDEMAND_NO;
 587                 d->do_iotime = CONFIG_ONDEMAND_NO;
 588                 d->do_qops = CONFIG_ONDEMAND_NO;
 589                 d->do_util = CONFIG_ONDEMAND_NO;
 590                 d->do_backlog = CONFIG_ONDEMAND_NO;
 591             }
 592             else {
 593                 // this disk is enabled
 594                 // check its direct settings
 595
 596                 int def_performance = CONFIG_ONDEMAND_ONDEMAND;
 597
 598                 // since this is 'on demand' we can figure the performance settings
 599                 // based on the type of disk
 600
 601                 switch(d->type) {
 602                     case DISK_TYPE_PHYSICAL:
 603                         def_performance = global_enable_performance_for_physical_disks;
 604                         break;
 605
 606                     case DISK_TYPE_PARTITION:
 607                         def_performance = global_enable_performance_for_partitions;
 608                         break;
 609
 610                     case DISK_TYPE_CONTAINER:
 611                         def_performance = global_enable_performance_for_virtual_disks;
 612                         break;
 613                 }
 614
 615                 // check if we have to disable performance for this disk
 616                 if(def_performance)
 617                     def_performance = is_major_enabled((int)major);
 618
 619                 // ------------------------------------------------------------
 620                 // now we have def_performance and def_space
 621                 // to work further
 622
 623                 // def_performance
 624                 // check the user configuration (this will also show our 'on demand' decision)
 625                 def_performance = config_get_boolean_ondemand(var_name, "enable performance metrics", def_performance);
 626
 627                 int ddo_io = CONFIG_ONDEMAND_NO,
 628                     ddo_ops = CONFIG_ONDEMAND_NO,
 629                     ddo_mops = CONFIG_ONDEMAND_NO,
 630                     ddo_iotime = CONFIG_ONDEMAND_NO,
 631                     ddo_qops = CONFIG_ONDEMAND_NO,
 632                     ddo_util = CONFIG_ONDEMAND_NO,
 633                     ddo_backlog = CONFIG_ONDEMAND_NO;
 634
 635                 // we enable individual performance charts only when def_performance is not disabled
 636                 if(unlikely(def_performance != CONFIG_ONDEMAND_NO)) {
 637                     ddo_io = global_do_io,
 638                     ddo_ops = global_do_ops,
 639                     ddo_mops = global_do_mops,
 640                     ddo_iotime = global_do_iotime,
 641                     ddo_qops = global_do_qops,
 642                     ddo_util = global_do_util,
 643                     ddo_backlog = global_do_backlog;
 644                 }
 645
 646                 d->do_io      = config_get_boolean_ondemand(var_name, "bandwidth", ddo_io);
 647                 d->do_ops     = config_get_boolean_ondemand(var_name, "operations", ddo_ops);
 648                 d->do_mops    = config_get_boolean_ondemand(var_name, "merged operations", ddo_mops);
 649                 d->do_iotime  = config_get_boolean_ondemand(var_name, "i/o time", ddo_iotime);
 650                 d->do_qops    = config_get_boolean_ondemand(var_name, "queued operations", ddo_qops);
 651                 d->do_util    = config_get_boolean_ondemand(var_name, "utilization percentage", ddo_util);
 652                 d->do_backlog = config_get_boolean_ondemand(var_name, "backlog", ddo_backlog);
 653             }
 654
 655             d->configured = 1;
 656         }
 657
 658         RRDSET *st;
 659
 660         // --------------------------------------------------------------------------
 661         // Do performance metrics
 662
 663         if(d->do_io == CONFIG_ONDEMAND_YES || (d->do_io == CONFIG_ONDEMAND_ONDEMAND && (readsectors || writesectors))) {
 664             d->do_io = CONFIG_ONDEMAND_YES;
 665
 666             st = rrdset_find_bytype(RRD_TYPE_DISK, disk);
 667             if(unlikely(!st)) {
 668                 st = rrdset_create(RRD_TYPE_DISK, disk, NULL, family, "disk.io", "Disk I/O Bandwidth", "kilobytes/s", 2000, update_every, RRDSET_TYPE_AREA);
 669
 670                 rrddim_add(st, "reads", NULL, d->sector_size, 1024, RRDDIM_INCREMENTAL);
 671                 rrddim_add(st, "writes", NULL, d->sector_size * -1, 1024, RRDDIM_INCREMENTAL);
 672             }
 673             else rrdset_next(st);
 674
 675             last_readsectors  = rrddim_set(st, "reads", readsectors);
 676             last_writesectors = rrddim_set(st, "writes", writesectors);
 677             rrdset_done(st);
 678         }
 679
 680         // --------------------------------------------------------------------
 681
 682         if(d->do_ops == CONFIG_ONDEMAND_YES || (d->do_ops == CONFIG_ONDEMAND_ONDEMAND && (reads || writes))) {
 683             d->do_ops = CONFIG_ONDEMAND_YES;
 684
 685             st = rrdset_find_bytype("disk_ops", disk);
 686             if(unlikely(!st)) {
 687                 st = rrdset_create("disk_ops", disk, NULL, family, "disk.ops", "Disk Completed I/O Operations", "operations/s", 2001, update_every, RRDSET_TYPE_LINE);
 688                 st->isdetail = 1;
 689
 690                 rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_INCREMENTAL);
 691                 rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_INCREMENTAL);
 692             }
 693             else rrdset_next(st);
 694
 695             last_reads  = rrddim_set(st, "reads", reads);
 696             last_writes = rrddim_set(st, "writes", writes);
 697             rrdset_done(st);
 698         }
 699
 700         // --------------------------------------------------------------------
 701
 702         if(d->do_qops == CONFIG_ONDEMAND_YES || (d->do_qops == CONFIG_ONDEMAND_ONDEMAND && queued_ios)) {
 703             d->do_qops = CONFIG_ONDEMAND_YES;
 704
 705             st = rrdset_find_bytype("disk_qops", disk);
 706             if(unlikely(!st)) {
 707                 st = rrdset_create("disk_qops", disk, NULL, family, "disk.qops", "Disk Current I/O Operations", "operations", 2002, update_every, RRDSET_TYPE_LINE);
 708                 st->isdetail = 1;
 709
 710                 rrddim_add(st, "operations", NULL, 1, 1, RRDDIM_ABSOLUTE);
 711             }
 712             else rrdset_next(st);
 713
 714             rrddim_set(st, "operations", queued_ios);
 715             rrdset_done(st);
 716         }
 717
 718         // --------------------------------------------------------------------
 719
 720         if(d->do_backlog == CONFIG_ONDEMAND_YES || (d->do_backlog == CONFIG_ONDEMAND_ONDEMAND && backlog_ms)) {
 721             d->do_backlog = CONFIG_ONDEMAND_YES;
 722
 723             st = rrdset_find_bytype("disk_backlog", disk);
 724             if(unlikely(!st)) {
 725                 st = rrdset_create("disk_backlog", disk, NULL, family, "disk.backlog", "Disk Backlog", "backlog (ms)", 2003, update_every, RRDSET_TYPE_AREA);
 726                 st->isdetail = 1;
 727
 728                 rrddim_add(st, "backlog", NULL, 1, 10, RRDDIM_INCREMENTAL);
 729             }
 730             else rrdset_next(st);
 731
 732             rrddim_set(st, "backlog", backlog_ms);
 733             rrdset_done(st);
 734         }
 735
 736         // --------------------------------------------------------------------
 737
 738         if(d->do_util == CONFIG_ONDEMAND_YES || (d->do_util == CONFIG_ONDEMAND_ONDEMAND && busy_ms)) {
 739             d->do_util = CONFIG_ONDEMAND_YES;
 740
 741             st = rrdset_find_bytype("disk_util", disk);
 742             if(unlikely(!st)) {
 743                 st = rrdset_create("disk_util", disk, NULL, family, "disk.util", "Disk Utilization Time", "% of time working", 2004, update_every, RRDSET_TYPE_AREA);
 744                 st->isdetail = 1;
 745
 746                 rrddim_add(st, "utilization", NULL, 1, 10, RRDDIM_INCREMENTAL);
 747             }
 748             else rrdset_next(st);
 749
 750             last_busy_ms = rrddim_set(st, "utilization", busy_ms);
 751             rrdset_done(st);
 752         }
 753
 754         // --------------------------------------------------------------------
 755
 756         if(d->do_mops == CONFIG_ONDEMAND_YES || (d->do_mops == CONFIG_ONDEMAND_ONDEMAND && (mreads || mwrites))) {
 757             d->do_mops = CONFIG_ONDEMAND_YES;
 758
 759             st = rrdset_find_bytype("disk_mops", disk);
 760             if(unlikely(!st)) {
 761                 st = rrdset_create("disk_mops", disk, NULL, family, "disk.mops", "Disk Merged Operations", "merged operations/s", 2021, update_every, RRDSET_TYPE_LINE);
 762                 st->isdetail = 1;
 763
 764                 rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_INCREMENTAL);
 765                 rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_INCREMENTAL);
 766             }
 767             else rrdset_next(st);
 768
 769             rrddim_set(st, "reads", mreads);
 770             rrddim_set(st, "writes", mwrites);
 771             rrdset_done(st);
 772         }
 773
 774         // --------------------------------------------------------------------
 775
 776         if(d->do_iotime == CONFIG_ONDEMAND_YES || (d->do_iotime == CONFIG_ONDEMAND_ONDEMAND && (readms || writems))) {
 777             d->do_iotime = CONFIG_ONDEMAND_YES;
 778
 779             st = rrdset_find_bytype("disk_iotime", disk);
 780             if(unlikely(!st)) {
 781                 st = rrdset_create("disk_iotime", disk, NULL, family, "disk.iotime", "Disk Total I/O Time", "milliseconds/s", 2022, update_every, RRDSET_TYPE_LINE);
 782                 st->isdetail = 1;
 783
 784                 rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_INCREMENTAL);
 785                 rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_INCREMENTAL);
 786             }
 787             else rrdset_next(st);
 788
 789             last_readms  = rrddim_set(st, "reads", readms);
 790             last_writems = rrddim_set(st, "writes", writems);
 791             rrdset_done(st);
 792         }
 793
 794         // --------------------------------------------------------------------
 795         // calculate differential charts
 796         // only if this is not the first time we run
 797
 798         if(likely(dt)) {
 799             if( (d->do_iotime == CONFIG_ONDEMAND_YES || (d->do_iotime == CONFIG_ONDEMAND_ONDEMAND && (readms || writems))) &&
 800                 (d->do_ops    == CONFIG_ONDEMAND_YES || (d->do_ops    == CONFIG_ONDEMAND_ONDEMAND && (reads || writes)))) {
 801                 st = rrdset_find_bytype("disk_await", disk);
 802                 if(unlikely(!st)) {
 803                     st = rrdset_create("disk_await", disk, NULL, family, "disk.await", "Average Completed I/O Operation Time", "ms per operation", 2005, update_every, RRDSET_TYPE_LINE);
 804                     st->isdetail = 1;
 805
 806                     rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_ABSOLUTE);
 807                     rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_ABSOLUTE);
 808                 }
 809                 else rrdset_next(st);
 810
 811                 rrddim_set(st, "reads", (reads - last_reads) ? (readms - last_readms) / (reads - last_reads) : 0);
 812                 rrddim_set(st, "writes", (writes - last_writes) ? (writems - last_writems) / (writes - last_writes) : 0);
 813                 rrdset_done(st);
 814             }
 815
 816             if( (d->do_io  == CONFIG_ONDEMAND_YES || (d->do_io  == CONFIG_ONDEMAND_ONDEMAND && (readsectors || writesectors))) &&
 817                 (d->do_ops == CONFIG_ONDEMAND_YES || (d->do_ops == CONFIG_ONDEMAND_ONDEMAND && (reads || writes)))) {
 818                 st = rrdset_find_bytype("disk_avgsz", disk);
 819                 if(unlikely(!st)) {
 820                     st = rrdset_create("disk_avgsz", disk, NULL, family, "disk.avgsz", "Average Completed I/O Operation Bandwidth", "kilobytes per operation", 2006, update_every, RRDSET_TYPE_AREA);
 821                     st->isdetail = 1;
 822
 823                     rrddim_add(st, "reads", NULL, d->sector_size, 1024, RRDDIM_ABSOLUTE);
 824                     rrddim_add(st, "writes", NULL, d->sector_size * -1, 1024, RRDDIM_ABSOLUTE);
 825                 }
 826                 else rrdset_next(st);
 827
 828                 rrddim_set(st, "reads", (reads - last_reads) ? (readsectors - last_readsectors) / (reads - last_reads) : 0);
 829                 rrddim_set(st, "writes", (writes - last_writes) ? (writesectors - last_writesectors) / (writes - last_writes) : 0);
 830                 rrdset_done(st);
 831             }
 832
 833             if( (d->do_util == CONFIG_ONDEMAND_YES || (d->do_util == CONFIG_ONDEMAND_ONDEMAND && busy_ms)) &&
 834                 (d->do_ops  == CONFIG_ONDEMAND_YES || (d->do_ops  == CONFIG_ONDEMAND_ONDEMAND && (reads || writes)))) {
 835                 st = rrdset_find_bytype("disk_svctm", disk);
 836                 if(unlikely(!st)) {
 837                     st = rrdset_create("disk_svctm", disk, NULL, family, "disk.svctm", "Average Service Time", "ms per operation", 2007, update_every, RRDSET_TYPE_LINE);
 838                     st->isdetail = 1;
 839
 840                     rrddim_add(st, "svctm", NULL, 1, 1, RRDDIM_ABSOLUTE);
 841                 }
 842                 else rrdset_next(st);
 843
 844                 rrddim_set(st, "svctm", ((reads - last_reads) + (writes - last_writes)) ? (busy_ms - last_busy_ms) / ((reads - last_reads) + (writes - last_writes)) : 0);
 845                 rrdset_done(st);
 846             }
 847         }
 848     }
 849
 850     return 0;
 851 }