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     uint32_t mount_point_hash;
  22
  23     // disk options caching
  24     int configured;
  25     int do_io;
  26     int do_ops;
  27     int do_mops;
  28     int do_iotime;
  29     int do_qops;
  30     int do_util;
  31     int do_backlog;
  32     int do_space;
  33     int do_inodes;
  34
  35     struct disk *next;
  36 } *disk_root = NULL;
  37
  38 static struct mountinfo *disk_mountinfo_root = NULL;
  39
  40 static inline void mountinfo_reload(int force) {
  41     static time_t last_loaded = 0;
  42     time_t now = time(NULL);
  43
  44     if(force || now - last_loaded >= NETDATA_RELOAD_MOUNTINFO_EVERY) {
  45 //#ifdef NETDATA_INTERNAL_CHECKS
  46 //        info("Reloading mountinfo");
  47 //#endif
  48
  49         // mountinfo_free() can be called with NULL disk_mountinfo_root
  50         mountinfo_free(disk_mountinfo_root);
  51
  52         // re-read mountinfo in case something changed
  53         disk_mountinfo_root = mountinfo_read();
  54
  55         last_loaded = now;
  56     }
  57 }
  58
  59
  60 // linked list of mount points that are by default disabled
  61 static struct excluded_mount_point {
  62     const char *prefix;
  63     size_t len;
  64     struct excluded_mount_point *next;
  65 } *excluded_mount_points = NULL;
  66
  67 static inline int is_mount_point_excluded(const char *mount_point) {
  68     static int initialized = 0;
  69
  70     if(unlikely(!initialized)) {
  71         initialized = 1;
  72
  73         char *a = config_get("plugin:proc:/proc/diskstats", "exclude space metrics on paths", "/var/run/user/ /run/user/");
  74         if(a && *a) {
  75             char *s = a;
  76
  77             while(s && *s) {
  78                 // skip all spaces
  79                 while(isspace(*s)) s++;
  80
  81                 // empty string
  82                 if(unlikely(!*s)) break;
  83
  84                 // find the next space
  85                 char *c = s;
  86                 while(*c && !isspace(*c)) c++;
  87
  88                 char *n;
  89                 if(likely(*c)) n = c + 1;
  90                 else n = NULL;
  91
  92                 // terminate our string
  93                 *c = '\0';
  94
  95                 // allocate the structure
  96                 struct excluded_mount_point *m = mallocz(sizeof(struct excluded_mount_point));
  97                 m->prefix = strdup(s);
  98                 m->len = strlen(m->prefix);
  99                 m->next = excluded_mount_points;
 100                 excluded_mount_points = m;
 101
 102                 // prepare for next loop
 103                 s = n;
 104                 if(likely(n)) *c = ' ';
 105             }
 106         }
 107     }
 108
 109     size_t len = strlen(mount_point);
 110     struct excluded_mount_point *m;
 111     for(m = excluded_mount_points; m ; m = m->next) {
 112         if(m->len <= len) {
 113             // fprintf(stderr, "SPACE: comparing '%s' with '%s'\n", mount_point, m->prefix);
 114             if(unlikely(strncmp(m->prefix, mount_point, m->len) == 0)) {
 115                 // fprintf(stderr, "SPACE: excluded '%s'\n", mount_point);
 116                 return 1;
 117             }
 118         }
 119     }
 120
 121     // fprintf(stderr, "SPACE: included '%s'\n", mount_point);
 122     return 0;
 123 }
 124
 125 // Data to be stored in DICTIONARY mount_points used by do_disk_space_stats().
 126 // This DICTIONARY is used to lookup the settings of the mount point on each iteration.
 127 struct mount_point_metadata {
 128     int do_space;
 129     int do_inodes;
 130 };
 131
 132 static inline void do_disk_space_stats(struct disk *d, const char *mount_point, const char *mount_source, const char *disk, const char *family, int update_every, unsigned long long dt) {
 133     (void)dt;
 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     if(unlikely(d)) {
 143         // verify we collected the metrics for the right disk.
 144         // if not the mountpoint has changed.
 145
 146         struct stat buff_stat;
 147         if(unlikely(stat(mount_point, &buff_stat) == -1)) {
 148             error("Failed to stat() for '%s' (disk '%s')", mount_point, disk);
 149             return;
 150         }
 151         else if(unlikely(major(buff_stat.st_dev) != d->major || minor(buff_stat.st_dev) != d->minor)) {
 152             error("Disk '%s' (disk '%s') switched major:minor", mount_point, disk);
 153             freez(d->mount_point);
 154             d->mount_point = NULL;
 155             d->mount_point_hash = 0;
 156             return;
 157         }
 158
 159         do_space = d->do_space;
 160         do_inodes = d->do_inodes;
 161     }
 162     else {
 163         struct mount_point_metadata *m = dictionary_get(mount_points, mount_point);
 164         if(unlikely(!m)) {
 165             char var_name[4096 + 1];
 166             snprintfz(var_name, 4096, "plugin:proc:/proc/diskstats:%s", mount_point);
 167
 168             int def_space = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "space usage for all disks", CONFIG_ONDEMAND_ONDEMAND);
 169             int def_inodes = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "inodes usage for all disks", CONFIG_ONDEMAND_ONDEMAND);
 170
 171             if(unlikely(is_mount_point_excluded(mount_point))) {
 172                 def_space = CONFIG_ONDEMAND_NO;
 173                 def_inodes = CONFIG_ONDEMAND_NO;
 174             }
 175
 176             do_space = config_get_boolean_ondemand(var_name, "space usage", def_space);
 177             do_inodes = config_get_boolean_ondemand(var_name, "inodes usage", def_inodes);
 178
 179             struct mount_point_metadata mp = {
 180                 .do_space = do_space,
 181                 .do_inodes = do_inodes
 182             };
 183
 184             dictionary_set(mount_points, mount_point, &mp, sizeof(struct mount_point_metadata));
 185         }
 186         else {
 187             do_space = m->do_space;
 188             do_inodes = m->do_inodes;
 189         }
 190     }
 191
 192     if(unlikely(do_space == CONFIG_ONDEMAND_NO && do_inodes == CONFIG_ONDEMAND_NO))
 193         return;
 194
 195     struct statvfs buff_statvfs;
 196     if (statvfs(mount_point, &buff_statvfs) < 0) {
 197         error("Failed statvfs() for '%s' (disk '%s')", mount_point, disk);
 198         return;
 199     }
 200
 201     // taken from get_fs_usage() found in coreutils
 202     unsigned long bsize = (buff_statvfs.f_frsize) ? buff_statvfs.f_frsize : buff_statvfs.f_bsize;
 203
 204     fsblkcnt_t bavail         = buff_statvfs.f_bavail;
 205     fsblkcnt_t btotal         = buff_statvfs.f_blocks;
 206     fsblkcnt_t bavail_root    = buff_statvfs.f_bfree;
 207     fsblkcnt_t breserved_root = bavail_root - bavail;
 208     fsblkcnt_t bused;
 209     if(likely(btotal >= bavail_root))
 210         bused = btotal - bavail_root;
 211     else
 212         bused = bavail_root - btotal;
 213
 214 #ifdef NETDATA_INTERNAL_CHECKS
 215     if(unlikely(btotal != bavail + breserved_root + bused))
 216         error("Disk block statistics for '%s' (disk '%s') do not sum up: total = %llu, available = %llu, reserved = %llu, used = %llu", mount_point, disk, (unsigned long long)btotal, (unsigned long long)bavail, (unsigned long long)breserved_root, (unsigned long long)bused);
 217 #endif
 218
 219     // --------------------------------------------------------------------------
 220
 221     fsfilcnt_t favail         = buff_statvfs.f_favail;
 222     fsfilcnt_t ftotal         = buff_statvfs.f_files;
 223     fsfilcnt_t favail_root    = buff_statvfs.f_ffree;
 224     fsfilcnt_t freserved_root = favail_root - favail;
 225     fsfilcnt_t fused          = ftotal - favail_root;
 226
 227 #ifdef NETDATA_INTERNAL_CHECKS
 228     if(unlikely(btotal != bavail + breserved_root + bused))
 229         error("Disk inode statistics for '%s' (disk '%s') do not sum up: total = %llu, available = %llu, reserved = %llu, used = %llu", mount_point, disk, (unsigned long long)ftotal, (unsigned long long)favail, (unsigned long long)freserved_root, (unsigned long long)fused);
 230 #endif
 231
 232     // --------------------------------------------------------------------------
 233
 234     RRDSET *st;
 235
 236     if(do_space == CONFIG_ONDEMAND_YES || (do_space == CONFIG_ONDEMAND_ONDEMAND && (bavail || breserved_root || bused))) {
 237         st = rrdset_find_bytype("disk_space", disk);
 238         if(unlikely(!st)) {
 239             char title[4096 + 1];
 240             snprintfz(title, 4096, "Disk Space Usage for %s [%s]", family, mount_source);
 241             st = rrdset_create("disk_space", disk, NULL, family, "disk.space", title, "GB", 2023, update_every, RRDSET_TYPE_STACKED);
 242
 243             rrddim_add(st, "avail", NULL, bsize, 1024*1024*1024, RRDDIM_ABSOLUTE);
 244             rrddim_add(st, "used" , NULL, bsize, 1024*1024*1024, RRDDIM_ABSOLUTE);
 245             rrddim_add(st, "reserved_for_root", "reserved for root", bsize, 1024*1024*1024, RRDDIM_ABSOLUTE);
 246         }
 247         else rrdset_next(st);
 248
 249         rrddim_set(st, "avail", bavail);
 250         rrddim_set(st, "used", bused);
 251         rrddim_set(st, "reserved_for_root", breserved_root);
 252         rrdset_done(st);
 253     }
 254
 255     // --------------------------------------------------------------------------
 256
 257     if(do_inodes == CONFIG_ONDEMAND_YES || (do_inodes == CONFIG_ONDEMAND_ONDEMAND && (favail || freserved_root || fused))) {
 258         st = rrdset_find_bytype("disk_inodes", disk);
 259         if(unlikely(!st)) {
 260             char title[4096 + 1];
 261             snprintfz(title, 4096, "Disk Files (inodes) Usage for %s [%s]", family, mount_source);
 262             st = rrdset_create("disk_inodes", disk, NULL, family, "disk.inodes", title, "Inodes", 2024, update_every, RRDSET_TYPE_STACKED);
 263
 264             rrddim_add(st, "avail", NULL, 1, 1, RRDDIM_ABSOLUTE);
 265             rrddim_add(st, "used" , NULL, 1, 1, RRDDIM_ABSOLUTE);
 266             rrddim_add(st, "reserved_for_root", "reserved for root", 1, 1, RRDDIM_ABSOLUTE);
 267         }
 268         else rrdset_next(st);
 269
 270         rrddim_set(st, "avail", favail);
 271         rrddim_set(st, "used", fused);
 272         rrddim_set(st, "reserved_for_root", freserved_root);
 273         rrdset_done(st);
 274     }
 275 }
 276
 277 static struct disk *get_disk(unsigned long major, unsigned long minor, char *disk) {
 278     static char path_to_get_hw_sector_size[FILENAME_MAX + 1] = "";
 279     static char path_to_get_hw_sector_size_partitions[FILENAME_MAX + 1] = "";
 280     static char path_find_block_device[FILENAME_MAX + 1] = "";
 281     struct disk *d;
 282
 283     // search for it in our RAM list.
 284     // this is sequential, but since we just walk through
 285     // and the number of disks / partitions in a system
 286     // should not be that many, it should be acceptable
 287     for(d = disk_root; d ; d = d->next)
 288         if(unlikely(d->major == major && d->minor == minor))
 289             break;
 290
 291     // if we found it, return it
 292     if(likely(d))
 293         return d;
 294
 295     // not found
 296     // create a new disk structure
 297     d = (struct disk *)mallocz(sizeof(struct disk));
 298
 299     d->disk = strdupz(disk);
 300     d->major = major;
 301     d->minor = minor;
 302     d->type = DISK_TYPE_PHYSICAL; // Default type. Changed later if not correct.
 303     d->configured = 0;
 304     d->sector_size = 512; // the default, will be changed below
 305     d->next = NULL;
 306
 307     // append it to the list
 308     if(unlikely(!disk_root))
 309         disk_root = d;
 310     else {
 311         struct disk *last;
 312         for(last = disk_root; last->next ;last = last->next);
 313         last->next = d;
 314     }
 315
 316     // ------------------------------------------------------------------------
 317     // find the type of the device
 318
 319     char buffer[FILENAME_MAX + 1];
 320
 321     // get the default path for finding info about the block device
 322     if(unlikely(!path_find_block_device[0])) {
 323         snprintfz(buffer, FILENAME_MAX, "%s%s", global_host_prefix, "/sys/dev/block/%lu:%lu/%s");
 324         snprintfz(path_find_block_device, FILENAME_MAX, "%s", config_get("plugin:proc:/proc/diskstats", "path to get block device infos", buffer));
 325     }
 326
 327     // find if it is a partition
 328     // by checking if /sys/dev/block/MAJOR:MINOR/partition is readable.
 329     snprintfz(buffer, FILENAME_MAX, path_find_block_device, major, minor, "partition");
 330     if(likely(access(buffer, R_OK) == 0)) {
 331         d->type = DISK_TYPE_PARTITION;
 332     }
 333     else {
 334         // find if it is a container
 335         // by checking if /sys/dev/block/MAJOR:MINOR/slaves has entries
 336         snprintfz(buffer, FILENAME_MAX, path_find_block_device, major, minor, "slaves/");
 337         DIR *dirp = opendir(buffer);
 338         if(likely(dirp != NULL)) {
 339             struct dirent *dp;
 340             while( (dp = readdir(dirp)) ) {
 341                 // . and .. are also files in empty folders.
 342                 if(unlikely(strcmp(dp->d_name, ".") == 0 || strcmp(dp->d_name, "..") == 0)) {
 343                     continue;
 344                 }
 345
 346                 d->type = DISK_TYPE_CONTAINER;
 347
 348                 // Stop the loop after we found one file.
 349                 break;
 350             }
 351             if(unlikely(closedir(dirp) == -1))
 352                 error("Unable to close dir %s", buffer);
 353         }
 354     }
 355
 356     // ------------------------------------------------------------------------
 357     // check if we can find its mount point
 358
 359     // mountinfo_find() can be called with NULL disk_mountinfo_root
 360     struct mountinfo *mi = mountinfo_find(disk_mountinfo_root, d->major, d->minor);
 361 /*    if(unlikely(!mi)) {
 362         mountinfo_reload(1);
 363
 364         // search again for this disk
 365         mi = mountinfo_find(disk_mountinfo_root, d->major, d->minor);
 366     }
 367 */
 368     if(unlikely(mi)) {
 369         d->mount_point = strdupz(mi->mount_point);
 370         d->mount_point_hash = mi->mount_point_hash;
 371     }
 372     else {
 373         d->mount_point = NULL;
 374         d->mount_point_hash = 0;
 375     }
 376
 377     // ------------------------------------------------------------------------
 378     // find the disk sector size
 379
 380     if(unlikely(!path_to_get_hw_sector_size[0])) {
 381         snprintfz(buffer, FILENAME_MAX, "%s%s", global_host_prefix, "/sys/block/%s/queue/hw_sector_size");
 382         snprintfz(path_to_get_hw_sector_size, FILENAME_MAX, "%s", config_get("plugin:proc:/proc/diskstats", "path to get h/w sector size", buffer));
 383     }
 384     if(unlikely(!path_to_get_hw_sector_size_partitions[0])) {
 385         snprintfz(buffer, FILENAME_MAX, "%s%s", global_host_prefix, "/sys/dev/block/%lu:%lu/subsystem/%s/../queue/hw_sector_size");
 386         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));
 387     }
 388
 389     {
 390         char tf[FILENAME_MAX + 1], *t;
 391         strncpyz(tf, d->disk, FILENAME_MAX);
 392
 393         // replace all / with !
 394         for(t = tf; *t ;t++)
 395             if(unlikely(*t == '/')) *t = '!';
 396
 397         if(likely(d->type == DISK_TYPE_PARTITION))
 398             snprintfz(buffer, FILENAME_MAX, path_to_get_hw_sector_size_partitions, d->major, d->minor, tf);
 399         else
 400             snprintfz(buffer, FILENAME_MAX, path_to_get_hw_sector_size, tf);
 401
 402         FILE *fpss = fopen(buffer, "r");
 403         if(likely(fpss)) {
 404             char buffer2[1024 + 1];
 405             char *tmp = fgets(buffer2, 1024, fpss);
 406
 407             if(likely(tmp)) {
 408                 d->sector_size = atoi(tmp);
 409                 if(unlikely(d->sector_size <= 0)) {
 410                     error("Invalid sector size %d for device %s in %s. Assuming 512.", d->sector_size, d->disk, buffer);
 411                     d->sector_size = 512;
 412                 }
 413             }
 414             else error("Cannot read data for sector size for device %s from %s. Assuming 512.", d->disk, buffer);
 415
 416             fclose(fpss);
 417         }
 418         else error("Cannot read sector size for device %s from %s. Assuming 512.", d->disk, buffer);
 419     }
 420
 421     return d;
 422 }
 423
 424 static inline int select_positive_option(int option1, int option2) {
 425     if(unlikely(option1 == CONFIG_ONDEMAND_YES || option2 == CONFIG_ONDEMAND_YES))
 426         return CONFIG_ONDEMAND_YES;
 427     else if(unlikely(option1 == CONFIG_ONDEMAND_ONDEMAND || option2 == CONFIG_ONDEMAND_ONDEMAND))
 428         return CONFIG_ONDEMAND_ONDEMAND;
 429
 430     return CONFIG_ONDEMAND_NO;
 431 }
 432
 433 int do_proc_diskstats(int update_every, unsigned long long dt) {
 434     (void)dt;
 435
 436     static procfile *ff = NULL;
 437     static int  global_enable_new_disks_detected_at_runtime = CONFIG_ONDEMAND_YES,
 438                 global_enable_performance_for_physical_disks = CONFIG_ONDEMAND_ONDEMAND,
 439                 global_enable_performance_for_virtual_disks = CONFIG_ONDEMAND_NO,
 440                 global_enable_performance_for_partitions = CONFIG_ONDEMAND_NO,
 441                 global_enable_performance_for_mountpoints = CONFIG_ONDEMAND_NO,
 442                 global_enable_performance_for_virtual_mountpoints = CONFIG_ONDEMAND_ONDEMAND,
 443                 global_do_io = CONFIG_ONDEMAND_ONDEMAND,
 444                 global_do_ops = CONFIG_ONDEMAND_ONDEMAND,
 445                 global_do_mops = CONFIG_ONDEMAND_ONDEMAND,
 446                 global_do_iotime = CONFIG_ONDEMAND_ONDEMAND,
 447                 global_do_qops = CONFIG_ONDEMAND_ONDEMAND,
 448                 global_do_util = CONFIG_ONDEMAND_ONDEMAND,
 449                 global_do_backlog = CONFIG_ONDEMAND_ONDEMAND,
 450                 globals_initialized = 0;
 451
 452     if(unlikely(!globals_initialized)) {
 453         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);
 454
 455         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);
 456         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);
 457         global_enable_performance_for_partitions = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "performance metrics for partitions", global_enable_performance_for_partitions);
 458         global_enable_performance_for_mountpoints = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "performance metrics for mounted filesystems", global_enable_performance_for_mountpoints);
 459         global_enable_performance_for_virtual_mountpoints = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "performance metrics for mounted virtual disks", global_enable_performance_for_virtual_mountpoints);
 460
 461         global_do_io      = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "bandwidth for all disks", global_do_io);
 462         global_do_ops     = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "operations for all disks", global_do_ops);
 463         global_do_mops    = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "merged operations for all disks", global_do_mops);
 464         global_do_iotime  = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "i/o time for all disks", global_do_iotime);
 465         global_do_qops    = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "queued operations for all disks", global_do_qops);
 466         global_do_util    = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "utilization percentage for all disks", global_do_util);
 467         global_do_backlog = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "backlog for all disks", global_do_backlog);
 468
 469         globals_initialized = 1;
 470     }
 471
 472     if(unlikely(!ff)) {
 473         char filename[FILENAME_MAX + 1];
 474         snprintfz(filename, FILENAME_MAX, "%s%s", global_host_prefix, "/proc/diskstats");
 475         ff = procfile_open(config_get("plugin:proc:/proc/diskstats", "filename to monitor", filename), " \t", PROCFILE_FLAG_DEFAULT);
 476     }
 477     if(unlikely(!ff)) return 1;
 478
 479     ff = procfile_readall(ff);
 480     if(unlikely(!ff)) return 0; // we return 0, so that we will retry to open it next time
 481
 482     uint32_t lines = procfile_lines(ff), l;
 483     uint32_t words;
 484
 485     // this is smart enough not to reload it every time
 486     mountinfo_reload(0);
 487
 488     for(l = 0; l < lines ;l++) {
 489         // --------------------------------------------------------------------------
 490         // Read parameters
 491
 492         char *disk;
 493         unsigned long       major = 0, minor = 0;
 494
 495         collected_number    reads = 0,  mreads = 0,  readsectors = 0,  readms = 0,
 496                             writes = 0, mwrites = 0, writesectors = 0, writems = 0,
 497                             queued_ios = 0, busy_ms = 0, backlog_ms = 0;
 498
 499         collected_number    last_reads = 0,  last_readsectors = 0,  last_readms = 0,
 500                             last_writes = 0, last_writesectors = 0, last_writems = 0,
 501                             last_busy_ms = 0;
 502
 503         words = procfile_linewords(ff, l);
 504         if(unlikely(words < 14)) continue;
 505
 506         major           = strtoul(procfile_lineword(ff, l, 0), NULL, 10);
 507         minor           = strtoul(procfile_lineword(ff, l, 1), NULL, 10);
 508         disk            = procfile_lineword(ff, l, 2);
 509
 510         // # of reads completed # of writes completed
 511         // This is the total number of reads or writes completed successfully.
 512         reads           = strtoull(procfile_lineword(ff, l, 3), NULL, 10);  // rd_ios
 513         writes          = strtoull(procfile_lineword(ff, l, 7), NULL, 10);  // wr_ios
 514
 515         // # of reads merged # of writes merged
 516         // Reads and writes which are adjacent to each other may be merged for
 517         // efficiency.  Thus two 4K reads may become one 8K read before it is
 518         // ultimately handed to the disk, and so it will be counted (and queued)
 519         mreads          = strtoull(procfile_lineword(ff, l, 4), NULL, 10);  // rd_merges_or_rd_sec
 520         mwrites         = strtoull(procfile_lineword(ff, l, 8), NULL, 10);  // wr_merges
 521
 522         // # of sectors read # of sectors written
 523         // This is the total number of sectors read or written successfully.
 524         readsectors     = strtoull(procfile_lineword(ff, l, 5), NULL, 10);  // rd_sec_or_wr_ios
 525         writesectors    = strtoull(procfile_lineword(ff, l, 9), NULL, 10);  // wr_sec
 526
 527         // # of milliseconds spent reading # of milliseconds spent writing
 528         // This is the total number of milliseconds spent by all reads or writes (as
 529         // measured from __make_request() to end_that_request_last()).
 530         readms          = strtoull(procfile_lineword(ff, l, 6), NULL, 10);  // rd_ticks_or_wr_sec
 531         writems         = strtoull(procfile_lineword(ff, l, 10), NULL, 10); // wr_ticks
 532
 533         // # of I/Os currently in progress
 534         // The only field that should go to zero. Incremented as requests are
 535         // given to appropriate struct request_queue and decremented as they finish.
 536         queued_ios      = strtoull(procfile_lineword(ff, l, 11), NULL, 10); // ios_pgr
 537
 538         // # of milliseconds spent doing I/Os
 539         // This field increases so long as field queued_ios is nonzero.
 540         busy_ms         = strtoull(procfile_lineword(ff, l, 12), NULL, 10); // tot_ticks
 541
 542         // weighted # of milliseconds spent doing I/Os
 543         // This field is incremented at each I/O start, I/O completion, I/O
 544         // merge, or read of these stats by the number of I/Os in progress
 545         // (field queued_ios) times the number of milliseconds spent doing I/O since the
 546         // last update of this field.  This can provide an easy measure of both
 547         // I/O completion time and the backlog that may be accumulating.
 548         backlog_ms      = strtoull(procfile_lineword(ff, l, 13), NULL, 10); // rq_ticks
 549
 550
 551         // --------------------------------------------------------------------------
 552         // remove slashes from disk names
 553         char *s;
 554         for(s = disk; *s ;s++)
 555             if(*s == '/') *s = '_';
 556
 557         // --------------------------------------------------------------------------
 558         // get a disk structure for the disk
 559
 560         struct disk *d = get_disk(major, minor, disk);
 561
 562
 563         // --------------------------------------------------------------------------
 564         // Set its family based on mount point
 565
 566         char *family = d->mount_point;
 567         if(!family) family = disk;
 568
 569
 570         // --------------------------------------------------------------------------
 571         // Check the configuration for the device
 572
 573         if(unlikely(!d->configured)) {
 574             char var_name[4096 + 1];
 575             snprintfz(var_name, 4096, "plugin:proc:/proc/diskstats:%s", disk);
 576
 577             int def_enable = config_get_boolean_ondemand(var_name, "enable", global_enable_new_disks_detected_at_runtime);
 578             if(unlikely(def_enable == CONFIG_ONDEMAND_NO)) {
 579                 // the user does not want any metrics for this disk
 580                 d->do_io = CONFIG_ONDEMAND_NO;
 581                 d->do_ops = CONFIG_ONDEMAND_NO;
 582                 d->do_mops = CONFIG_ONDEMAND_NO;
 583                 d->do_iotime = CONFIG_ONDEMAND_NO;
 584                 d->do_qops = CONFIG_ONDEMAND_NO;
 585                 d->do_util = CONFIG_ONDEMAND_NO;
 586                 d->do_backlog = CONFIG_ONDEMAND_NO;
 587                 d->do_space = CONFIG_ONDEMAND_NO;
 588                 d->do_inodes = CONFIG_ONDEMAND_NO;
 589             }
 590             else {
 591                 // this disk is enabled
 592                 // check its direct settings
 593
 594                 int def_performance = CONFIG_ONDEMAND_ONDEMAND;
 595                 int def_space = (d->mount_point)?CONFIG_ONDEMAND_ONDEMAND:CONFIG_ONDEMAND_NO;
 596
 597                 // since this is 'on demand' we can figure the performance settings
 598                 // based on the type of disk
 599
 600                 switch(d->type) {
 601                     case DISK_TYPE_PHYSICAL:
 602                         def_performance = global_enable_performance_for_physical_disks;
 603                         break;
 604
 605                     case DISK_TYPE_PARTITION:
 606                         def_performance = global_enable_performance_for_partitions;
 607                         break;
 608
 609                     case DISK_TYPE_CONTAINER:
 610                         def_performance = global_enable_performance_for_virtual_disks;
 611
 612                         if(d->mount_point)
 613                             def_performance = select_positive_option(def_performance, global_enable_performance_for_virtual_mountpoints);
 614                         break;
 615                 }
 616
 617                 if(d->mount_point)
 618                     def_performance = select_positive_option(def_performance, global_enable_performance_for_mountpoints);
 619
 620                 // ------------------------------------------------------------
 621                 // now we have def_performance and def_space
 622                 // to work further
 623
 624                 // def_performance
 625                 // check the user configuration (this will also show our 'on demand' decision)
 626                 def_performance = config_get_boolean_ondemand(var_name, "enable performance metrics", def_performance);
 627
 628                 int ddo_io = CONFIG_ONDEMAND_NO,
 629                     ddo_ops = CONFIG_ONDEMAND_NO,
 630                     ddo_mops = CONFIG_ONDEMAND_NO,
 631                     ddo_iotime = CONFIG_ONDEMAND_NO,
 632                     ddo_qops = CONFIG_ONDEMAND_NO,
 633                     ddo_util = CONFIG_ONDEMAND_NO,
 634                     ddo_backlog = CONFIG_ONDEMAND_NO;
 635
 636                 // we enable individual performance charts only when def_performance is not disabled
 637                 if(unlikely(def_performance != CONFIG_ONDEMAND_NO)) {
 638                     ddo_io = global_do_io,
 639                     ddo_ops = global_do_ops,
 640                     ddo_mops = global_do_mops,
 641                     ddo_iotime = global_do_iotime,
 642                     ddo_qops = global_do_qops,
 643                     ddo_util = global_do_util,
 644                     ddo_backlog = global_do_backlog;
 645                 }
 646
 647                 d->do_io      = config_get_boolean_ondemand(var_name, "bandwidth", ddo_io);
 648                 d->do_ops     = config_get_boolean_ondemand(var_name, "operations", ddo_ops);
 649                 d->do_mops    = config_get_boolean_ondemand(var_name, "merged operations", ddo_mops);
 650                 d->do_iotime  = config_get_boolean_ondemand(var_name, "i/o time", ddo_iotime);
 651                 d->do_qops    = config_get_boolean_ondemand(var_name, "queued operations", ddo_qops);
 652                 d->do_util    = config_get_boolean_ondemand(var_name, "utilization percentage", ddo_util);
 653                 d->do_backlog = config_get_boolean_ondemand(var_name, "backlog", ddo_backlog);
 654
 655                 // def_space
 656                 if(unlikely(d->mount_point)) {
 657                     // check the user configuration (this will also show our 'on demand' decision)
 658                     def_space = config_get_boolean_ondemand(var_name, "enable space metrics", def_space);
 659
 660                     int ddo_space = def_space,
 661                         ddo_inodes = def_space;
 662
 663                     d->do_space = config_get_boolean_ondemand(var_name, "space usage", ddo_space);
 664                     d->do_inodes = config_get_boolean_ondemand(var_name, "inodes usage", ddo_inodes);
 665                 }
 666                 else {
 667                     // don't show settings for this disk
 668                     d->do_space = CONFIG_ONDEMAND_NO;
 669                     d->do_inodes = CONFIG_ONDEMAND_NO;
 670                 }
 671             }
 672
 673             d->configured = 1;
 674         }
 675
 676         RRDSET *st;
 677
 678         // --------------------------------------------------------------------------
 679         // Do performance metrics
 680
 681         if(d->do_io == CONFIG_ONDEMAND_YES || (d->do_io == CONFIG_ONDEMAND_ONDEMAND && (readsectors || writesectors))) {
 682             d->do_io = CONFIG_ONDEMAND_YES;
 683
 684             st = rrdset_find_bytype(RRD_TYPE_DISK, disk);
 685             if(unlikely(!st)) {
 686                 st = rrdset_create(RRD_TYPE_DISK, disk, NULL, family, "disk.io", "Disk I/O Bandwidth", "kilobytes/s", 2000, update_every, RRDSET_TYPE_AREA);
 687
 688                 rrddim_add(st, "reads", NULL, d->sector_size, 1024, RRDDIM_INCREMENTAL);
 689                 rrddim_add(st, "writes", NULL, d->sector_size * -1, 1024, RRDDIM_INCREMENTAL);
 690             }
 691             else rrdset_next(st);
 692
 693             last_readsectors  = rrddim_set(st, "reads", readsectors);
 694             last_writesectors = rrddim_set(st, "writes", writesectors);
 695             rrdset_done(st);
 696         }
 697
 698         // --------------------------------------------------------------------
 699
 700         if(d->do_ops == CONFIG_ONDEMAND_YES || (d->do_ops == CONFIG_ONDEMAND_ONDEMAND && (reads || writes))) {
 701             d->do_ops = CONFIG_ONDEMAND_YES;
 702
 703             st = rrdset_find_bytype("disk_ops", disk);
 704             if(unlikely(!st)) {
 705                 st = rrdset_create("disk_ops", disk, NULL, family, "disk.ops", "Disk Completed I/O Operations", "operations/s", 2001, update_every, RRDSET_TYPE_LINE);
 706                 st->isdetail = 1;
 707
 708                 rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_INCREMENTAL);
 709                 rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_INCREMENTAL);
 710             }
 711             else rrdset_next(st);
 712
 713             last_reads  = rrddim_set(st, "reads", reads);
 714             last_writes = rrddim_set(st, "writes", writes);
 715             rrdset_done(st);
 716         }
 717
 718         // --------------------------------------------------------------------
 719
 720         if(d->do_qops == CONFIG_ONDEMAND_YES || (d->do_qops == CONFIG_ONDEMAND_ONDEMAND && queued_ios)) {
 721             d->do_qops = CONFIG_ONDEMAND_YES;
 722
 723             st = rrdset_find_bytype("disk_qops", disk);
 724             if(unlikely(!st)) {
 725                 st = rrdset_create("disk_qops", disk, NULL, family, "disk.qops", "Disk Current I/O Operations", "operations", 2002, update_every, RRDSET_TYPE_LINE);
 726                 st->isdetail = 1;
 727
 728                 rrddim_add(st, "operations", NULL, 1, 1, RRDDIM_ABSOLUTE);
 729             }
 730             else rrdset_next(st);
 731
 732             rrddim_set(st, "operations", queued_ios);
 733             rrdset_done(st);
 734         }
 735
 736         // --------------------------------------------------------------------
 737
 738         if(d->do_backlog == CONFIG_ONDEMAND_YES || (d->do_backlog == CONFIG_ONDEMAND_ONDEMAND && backlog_ms)) {
 739             d->do_backlog = CONFIG_ONDEMAND_YES;
 740
 741             st = rrdset_find_bytype("disk_backlog", disk);
 742             if(unlikely(!st)) {
 743                 st = rrdset_create("disk_backlog", disk, NULL, family, "disk.backlog", "Disk Backlog", "backlog (ms)", 2003, update_every, RRDSET_TYPE_AREA);
 744                 st->isdetail = 1;
 745
 746                 rrddim_add(st, "backlog", NULL, 1, 10, RRDDIM_INCREMENTAL);
 747             }
 748             else rrdset_next(st);
 749
 750             rrddim_set(st, "backlog", backlog_ms);
 751             rrdset_done(st);
 752         }
 753
 754         // --------------------------------------------------------------------
 755
 756         if(d->do_util == CONFIG_ONDEMAND_YES || (d->do_util == CONFIG_ONDEMAND_ONDEMAND && busy_ms)) {
 757             d->do_util = CONFIG_ONDEMAND_YES;
 758
 759             st = rrdset_find_bytype("disk_util", disk);
 760             if(unlikely(!st)) {
 761                 st = rrdset_create("disk_util", disk, NULL, family, "disk.util", "Disk Utilization Time", "% of time working", 2004, update_every, RRDSET_TYPE_AREA);
 762                 st->isdetail = 1;
 763
 764                 rrddim_add(st, "utilization", NULL, 1, 10, RRDDIM_INCREMENTAL);
 765             }
 766             else rrdset_next(st);
 767
 768             last_busy_ms = rrddim_set(st, "utilization", busy_ms);
 769             rrdset_done(st);
 770         }
 771
 772         // --------------------------------------------------------------------
 773
 774         if(d->do_mops == CONFIG_ONDEMAND_YES || (d->do_mops == CONFIG_ONDEMAND_ONDEMAND && (mreads || mwrites))) {
 775             d->do_mops = CONFIG_ONDEMAND_YES;
 776
 777             st = rrdset_find_bytype("disk_mops", disk);
 778             if(unlikely(!st)) {
 779                 st = rrdset_create("disk_mops", disk, NULL, family, "disk.mops", "Disk Merged Operations", "merged operations/s", 2021, update_every, RRDSET_TYPE_LINE);
 780                 st->isdetail = 1;
 781
 782                 rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_INCREMENTAL);
 783                 rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_INCREMENTAL);
 784             }
 785             else rrdset_next(st);
 786
 787             rrddim_set(st, "reads", mreads);
 788             rrddim_set(st, "writes", mwrites);
 789             rrdset_done(st);
 790         }
 791
 792         // --------------------------------------------------------------------
 793
 794         if(d->do_iotime == CONFIG_ONDEMAND_YES || (d->do_iotime == CONFIG_ONDEMAND_ONDEMAND && (readms || writems))) {
 795             d->do_iotime = CONFIG_ONDEMAND_YES;
 796
 797             st = rrdset_find_bytype("disk_iotime", disk);
 798             if(unlikely(!st)) {
 799                 st = rrdset_create("disk_iotime", disk, NULL, family, "disk.iotime", "Disk Total I/O Time", "milliseconds/s", 2022, update_every, RRDSET_TYPE_LINE);
 800                 st->isdetail = 1;
 801
 802                 rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_INCREMENTAL);
 803                 rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_INCREMENTAL);
 804             }
 805             else rrdset_next(st);
 806
 807             last_readms  = rrddim_set(st, "reads", readms);
 808             last_writems = rrddim_set(st, "writes", writems);
 809             rrdset_done(st);
 810         }
 811
 812         // --------------------------------------------------------------------
 813         // calculate differential charts
 814         // only if this is not the first time we run
 815
 816         if(likely(dt)) {
 817             if( (d->do_iotime == CONFIG_ONDEMAND_YES || (d->do_iotime == CONFIG_ONDEMAND_ONDEMAND && (readms || writems))) &&
 818                 (d->do_ops    == CONFIG_ONDEMAND_YES || (d->do_ops    == CONFIG_ONDEMAND_ONDEMAND && (reads || writes)))) {
 819                 st = rrdset_find_bytype("disk_await", disk);
 820                 if(unlikely(!st)) {
 821                     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);
 822                     st->isdetail = 1;
 823
 824                     rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_ABSOLUTE);
 825                     rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_ABSOLUTE);
 826                 }
 827                 else rrdset_next(st);
 828
 829                 rrddim_set(st, "reads", (reads - last_reads) ? (readms - last_readms) / (reads - last_reads) : 0);
 830                 rrddim_set(st, "writes", (writes - last_writes) ? (writems - last_writems) / (writes - last_writes) : 0);
 831                 rrdset_done(st);
 832             }
 833
 834             if( (d->do_io  == CONFIG_ONDEMAND_YES || (d->do_io  == CONFIG_ONDEMAND_ONDEMAND && (readsectors || writesectors))) &&
 835                 (d->do_ops == CONFIG_ONDEMAND_YES || (d->do_ops == CONFIG_ONDEMAND_ONDEMAND && (reads || writes)))) {
 836                 st = rrdset_find_bytype("disk_avgsz", disk);
 837                 if(unlikely(!st)) {
 838                     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);
 839                     st->isdetail = 1;
 840
 841                     rrddim_add(st, "reads", NULL, d->sector_size, 1024, RRDDIM_ABSOLUTE);
 842                     rrddim_add(st, "writes", NULL, d->sector_size * -1, 1024, RRDDIM_ABSOLUTE);
 843                 }
 844                 else rrdset_next(st);
 845
 846                 rrddim_set(st, "reads", (reads - last_reads) ? (readsectors - last_readsectors) / (reads - last_reads) : 0);
 847                 rrddim_set(st, "writes", (writes - last_writes) ? (writesectors - last_writesectors) / (writes - last_writes) : 0);
 848                 rrdset_done(st);
 849             }
 850
 851             if( (d->do_util == CONFIG_ONDEMAND_YES || (d->do_util == CONFIG_ONDEMAND_ONDEMAND && busy_ms)) &&
 852                 (d->do_ops  == CONFIG_ONDEMAND_YES || (d->do_ops  == CONFIG_ONDEMAND_ONDEMAND && (reads || writes)))) {
 853                 st = rrdset_find_bytype("disk_svctm", disk);
 854                 if(unlikely(!st)) {
 855                     st = rrdset_create("disk_svctm", disk, NULL, family, "disk.svctm", "Average Service Time", "ms per operation", 2007, update_every, RRDSET_TYPE_LINE);
 856                     st->isdetail = 1;
 857
 858                     rrddim_add(st, "svctm", NULL, 1, 1, RRDDIM_ABSOLUTE);
 859                 }
 860                 else rrdset_next(st);
 861
 862                 rrddim_set(st, "svctm", ((reads - last_reads) + (writes - last_writes)) ? (busy_ms - last_busy_ms) / ((reads - last_reads) + (writes - last_writes)) : 0);
 863                 rrdset_done(st);
 864             }
 865         }
 866
 867 /*
 868         // --------------------------------------------------------------------------
 869         // space metrics
 870
 871         if(unlikely( d->mount_point && (d->do_space || d->do_inodes) )) {
 872             do_disk_space_stats(d, d->mount_point, disk, disk, family, update_every, dt);
 873         }
 874 */
 875     }
 876
 877     // --------------------------------------------------------------------------
 878     // space metrics for non-block devices
 879
 880     struct mountinfo *mi;
 881     for(mi = disk_mountinfo_root; mi ;mi = mi->next) {
 882         if(unlikely(mi->flags & (MOUNTINFO_IS_DUMMY|MOUNTINFO_IS_BIND|MOUNTINFO_IS_SAME_DEV|MOUNTINFO_NO_STAT|MOUNTINFO_NO_SIZE)))
 883             continue;
 884
 885 /*
 886         // skip the ones with block devices
 887         int skip = 0;
 888         struct disk *d;
 889         for(d = disk_root; d ;d = d->next) {
 890             if(unlikely(d->mount_point && mi->mount_point_hash == d->mount_point_hash && strcmp(mi->mount_point, d->mount_point))) {
 891                 skip = 1;
 892                 break;
 893             }
 894         }
 895
 896         if(unlikely(skip))
 897             continue;
 898
 899         // fprintf(stderr, "Will process mount point '%s', source '%s', filesystem '%s'\n", mi->mount_point, mi->mount_source, mi->filesystem);
 900 */
 901
 902         do_disk_space_stats(NULL, mi->mount_point, mi->mount_source, mi->persistent_id, mi->mount_point , update_every, dt);
 903     }
 904
 905     return 0;
 906 }