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(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     static DICTIONARY *mount_points = NULL;
 134     int do_space, do_inodes;
 135
 136     if(unlikely(!mount_points)) {
 137         mount_points = dictionary_create(DICTIONARY_FLAG_SINGLE_THREADED);
 138     }
 139
 140     if(unlikely(d)) {
 141         // verify we collected the metrics for the right disk.
 142         // if not the mountpoint has changed.
 143
 144         struct stat buff_stat;
 145         if(stat(mount_point, &buff_stat) == -1) {
 146             error("Failed to stat() for '%s' (disk '%s')", mount_point, disk);
 147             return;
 148         }
 149         else if(major(buff_stat.st_dev) != d->major || minor(buff_stat.st_dev) != d->minor) {
 150             error("Disk '%s' (disk '%s') switched major:minor", mount_point, disk);
 151             freez(d->mount_point);
 152             d->mount_point = NULL;
 153             d->mount_point_hash = 0;
 154             return;
 155         }
 156
 157         do_space = d->do_space;
 158         do_inodes = d->do_inodes;
 159     }
 160     else {
 161         struct mount_point_metadata *m = dictionary_get(mount_points, mount_point);
 162         if(!m) {
 163             char var_name[4096 + 1];
 164             snprintfz(var_name, 4096, "plugin:proc:/proc/diskstats:%s", mount_point);
 165
 166             int def_space = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "space usage for all disks", CONFIG_ONDEMAND_ONDEMAND);
 167             int def_inodes = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "inodes usage for all disks", CONFIG_ONDEMAND_ONDEMAND);
 168
 169             if(is_mount_point_excluded(mount_point)) {
 170                 def_space = CONFIG_ONDEMAND_NO;
 171                 def_inodes = CONFIG_ONDEMAND_NO;
 172             }
 173
 174             do_space = config_get_boolean_ondemand(var_name, "space usage", def_space);
 175             do_inodes = config_get_boolean_ondemand(var_name, "inodes usage", def_inodes);
 176
 177             struct mount_point_metadata mp = {
 178                 .do_space = do_space,
 179                 .do_inodes = do_inodes
 180             };
 181
 182             dictionary_set(mount_points, mount_point, &mp, sizeof(struct mount_point_metadata));
 183         }
 184         else {
 185             do_space = m->do_space;
 186             do_inodes = m->do_inodes;
 187         }
 188     }
 189
 190     if(do_space == CONFIG_ONDEMAND_NO && do_inodes == CONFIG_ONDEMAND_NO)
 191         return;
 192
 193     struct statvfs buff_statvfs;
 194     if (statvfs(mount_point, &buff_statvfs) < 0) {
 195         error("Failed statvfs() for '%s' (disk '%s')", mount_point, disk);
 196         return;
 197     }
 198
 199     // taken from get_fs_usage() found in coreutils
 200     unsigned long bsize = (buff_statvfs.f_frsize) ? buff_statvfs.f_frsize : buff_statvfs.f_bsize;
 201
 202     fsblkcnt_t bavail         = buff_statvfs.f_bavail;
 203     fsblkcnt_t btotal         = buff_statvfs.f_blocks;
 204     fsblkcnt_t bavail_root    = buff_statvfs.f_bfree;
 205     fsblkcnt_t breserved_root = bavail_root - bavail;
 206     fsblkcnt_t bused;
 207     if(likely(btotal >= bavail_root))
 208         bused = btotal - bavail_root;
 209     else
 210         bused = bavail_root - btotal;
 211
 212 #ifdef NETDATA_INTERNAL_CHECKS
 213     if(unlikely(btotal != bavail + breserved_root + bused))
 214         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);
 215 #endif
 216
 217     // --------------------------------------------------------------------------
 218
 219     fsfilcnt_t favail         = buff_statvfs.f_favail;
 220     fsfilcnt_t ftotal         = buff_statvfs.f_files;
 221     fsfilcnt_t favail_root    = buff_statvfs.f_ffree;
 222     fsfilcnt_t freserved_root = favail_root - favail;
 223     fsfilcnt_t fused          = ftotal - favail_root;
 224
 225 #ifdef NETDATA_INTERNAL_CHECKS
 226     if(unlikely(btotal != bavail + breserved_root + bused))
 227         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);
 228 #endif
 229
 230     // --------------------------------------------------------------------------
 231
 232     RRDSET *st;
 233
 234     if(do_space == CONFIG_ONDEMAND_YES || (do_space == CONFIG_ONDEMAND_ONDEMAND && (bavail || breserved_root || bused))) {
 235         st = rrdset_find_bytype("disk_space", disk);
 236         if(!st) {
 237             char title[4096 + 1];
 238             snprintfz(title, 4096, "Disk Space Usage for %s [%s]", family, mount_source);
 239             st = rrdset_create("disk_space", disk, NULL, family, "disk.space", title, "GB", 2023, update_every, RRDSET_TYPE_STACKED);
 240
 241             rrddim_add(st, "avail", NULL, bsize, 1024*1024*1024, RRDDIM_ABSOLUTE);
 242             rrddim_add(st, "used" , NULL, bsize, 1024*1024*1024, RRDDIM_ABSOLUTE);
 243             rrddim_add(st, "reserved_for_root", "reserved for root", bsize, 1024*1024*1024, RRDDIM_ABSOLUTE);
 244         }
 245         else rrdset_next_usec(st, dt);
 246
 247         rrddim_set(st, "avail", bavail);
 248         rrddim_set(st, "used", bused);
 249         rrddim_set(st, "reserved_for_root", breserved_root);
 250         rrdset_done(st);
 251     }
 252
 253     // --------------------------------------------------------------------------
 254
 255     if(do_inodes == CONFIG_ONDEMAND_YES || (do_inodes == CONFIG_ONDEMAND_ONDEMAND && (favail || freserved_root || fused))) {
 256         st = rrdset_find_bytype("disk_inodes", disk);
 257         if(!st) {
 258             char title[4096 + 1];
 259             snprintfz(title, 4096, "Disk Files (inodes) Usage for %s [%s]", family, mount_source);
 260             st = rrdset_create("disk_inodes", disk, NULL, family, "disk.inodes", title, "Inodes", 2024, update_every, RRDSET_TYPE_STACKED);
 261
 262             rrddim_add(st, "avail", NULL, 1, 1, RRDDIM_ABSOLUTE);
 263             rrddim_add(st, "used" , NULL, 1, 1, RRDDIM_ABSOLUTE);
 264             rrddim_add(st, "reserved_for_root", "reserved for root", 1, 1, RRDDIM_ABSOLUTE);
 265         }
 266         else rrdset_next_usec(st, dt);
 267
 268         rrddim_set(st, "avail", favail);
 269         rrddim_set(st, "used", fused);
 270         rrddim_set(st, "reserved_for_root", freserved_root);
 271         rrdset_done(st);
 272     }
 273 }
 274
 275 static struct disk *get_disk(unsigned long major, unsigned long minor, char *disk) {
 276     static char path_to_get_hw_sector_size[FILENAME_MAX + 1] = "";
 277     static char path_to_get_hw_sector_size_partitions[FILENAME_MAX + 1] = "";
 278     static char path_find_block_device[FILENAME_MAX + 1] = "";
 279     struct disk *d;
 280
 281     // search for it in our RAM list.
 282     // this is sequential, but since we just walk through
 283     // and the number of disks / partitions in a system
 284     // should not be that many, it should be acceptable
 285     for(d = disk_root; d ; d = d->next)
 286         if(unlikely(d->major == major && d->minor == minor))
 287             break;
 288
 289     // if we found it, return it
 290     if(likely(d))
 291         return d;
 292
 293     // not found
 294     // create a new disk structure
 295     d = (struct disk *)mallocz(sizeof(struct disk));
 296
 297     d->disk = strdupz(disk);
 298     d->major = major;
 299     d->minor = minor;
 300     d->type = DISK_TYPE_PHYSICAL; // Default type. Changed later if not correct.
 301     d->configured = 0;
 302     d->sector_size = 512; // the default, will be changed below
 303     d->next = NULL;
 304
 305     // append it to the list
 306     if(!disk_root)
 307         disk_root = d;
 308     else {
 309         struct disk *last;
 310         for(last = disk_root; last->next ;last = last->next);
 311         last->next = d;
 312     }
 313
 314     // ------------------------------------------------------------------------
 315     // find the type of the device
 316
 317     char buffer[FILENAME_MAX + 1];
 318
 319     // get the default path for finding info about the block device
 320     if(unlikely(!path_find_block_device[0])) {
 321         snprintfz(buffer, FILENAME_MAX, "%s%s", global_host_prefix, "/sys/dev/block/%lu:%lu/%s");
 322         snprintfz(path_find_block_device, FILENAME_MAX, "%s", config_get("plugin:proc:/proc/diskstats", "path to get block device infos", buffer));
 323     }
 324
 325     // find if it is a partition
 326     // by checking if /sys/dev/block/MAJOR:MINOR/partition is readable.
 327     snprintfz(buffer, FILENAME_MAX, path_find_block_device, major, minor, "partition");
 328     if(access(buffer, R_OK) == 0) {
 329         d->type = DISK_TYPE_PARTITION;
 330     } else {
 331         // find if it is a container
 332         // by checking if /sys/dev/block/MAJOR:MINOR/slaves has entries
 333         snprintfz(buffer, FILENAME_MAX, path_find_block_device, major, minor, "slaves/");
 334         DIR *dirp = opendir(buffer);
 335         if (dirp != NULL) {
 336             struct dirent *dp;
 337             while( (dp = readdir(dirp)) ) {
 338                 // . and .. are also files in empty folders.
 339                 if(strcmp(dp->d_name, ".") == 0 || strcmp(dp->d_name, "..") == 0) {
 340                     continue;
 341                 }
 342
 343                 d->type = DISK_TYPE_CONTAINER;
 344
 345                 // Stop the loop after we found one file.
 346                 break;
 347             }
 348             if(closedir(dirp) == -1)
 349                 error("Unable to close dir %s", buffer);
 350         }
 351     }
 352
 353     // ------------------------------------------------------------------------
 354     // check if we can find its mount point
 355
 356     // mountinfo_find() can be called with NULL disk_mountinfo_root
 357     struct mountinfo *mi = mountinfo_find(disk_mountinfo_root, d->major, d->minor);
 358 /*    if(unlikely(!mi)) {
 359         mountinfo_reload(1);
 360
 361         // search again for this disk
 362         mi = mountinfo_find(disk_mountinfo_root, d->major, d->minor);
 363     }
 364 */
 365     if(mi) {
 366         d->mount_point = strdupz(mi->mount_point);
 367         d->mount_point_hash = mi->mount_point_hash;
 368     }
 369     else {
 370         d->mount_point = NULL;
 371         d->mount_point_hash = 0;
 372     }
 373
 374     // ------------------------------------------------------------------------
 375     // find the disk sector size
 376
 377     if(!path_to_get_hw_sector_size[0]) {
 378         snprintfz(buffer, FILENAME_MAX, "%s%s", global_host_prefix, "/sys/block/%s/queue/hw_sector_size");
 379         snprintfz(path_to_get_hw_sector_size, FILENAME_MAX, "%s", config_get("plugin:proc:/proc/diskstats", "path to get h/w sector size", buffer));
 380     }
 381     if(!path_to_get_hw_sector_size_partitions[0]) {
 382         snprintfz(buffer, FILENAME_MAX, "%s%s", global_host_prefix, "/sys/dev/block/%lu:%lu/subsystem/%s/../queue/hw_sector_size");
 383         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));
 384     }
 385
 386     {
 387         char tf[FILENAME_MAX + 1], *t;
 388         strncpyz(tf, d->disk, FILENAME_MAX);
 389
 390         // replace all / with !
 391         for(t = tf; *t ;t++)
 392             if(*t == '/') *t = '!';
 393
 394         if(d->type == DISK_TYPE_PARTITION)
 395             snprintfz(buffer, FILENAME_MAX, path_to_get_hw_sector_size_partitions, d->major, d->minor, tf);
 396         else
 397             snprintfz(buffer, FILENAME_MAX, path_to_get_hw_sector_size, tf);
 398
 399         FILE *fpss = fopen(buffer, "r");
 400         if(fpss) {
 401             char buffer2[1024 + 1];
 402             char *tmp = fgets(buffer2, 1024, fpss);
 403
 404             if(tmp) {
 405                 d->sector_size = atoi(tmp);
 406                 if(d->sector_size <= 0) {
 407                     error("Invalid sector size %d for device %s in %s. Assuming 512.", d->sector_size, d->disk, buffer);
 408                     d->sector_size = 512;
 409                 }
 410             }
 411             else error("Cannot read data for sector size for device %s from %s. Assuming 512.", d->disk, buffer);
 412
 413             fclose(fpss);
 414         }
 415         else error("Cannot read sector size for device %s from %s. Assuming 512.", d->disk, buffer);
 416     }
 417
 418     return d;
 419 }
 420
 421 static inline int select_positive_option(int option1, int option2) {
 422     if(option1 == CONFIG_ONDEMAND_YES || option2 == CONFIG_ONDEMAND_YES)
 423         return CONFIG_ONDEMAND_YES;
 424     else if(option1 == CONFIG_ONDEMAND_ONDEMAND || option2 == CONFIG_ONDEMAND_ONDEMAND)
 425         return CONFIG_ONDEMAND_ONDEMAND;
 426
 427     return CONFIG_ONDEMAND_NO;
 428 }
 429
 430 int do_proc_diskstats(int update_every, unsigned long long dt) {
 431     static procfile *ff = NULL;
 432     static int  global_enable_new_disks_detected_at_runtime = CONFIG_ONDEMAND_YES,
 433                 global_enable_performance_for_physical_disks = CONFIG_ONDEMAND_ONDEMAND,
 434                 global_enable_performance_for_virtual_disks = CONFIG_ONDEMAND_NO,
 435                 global_enable_performance_for_partitions = CONFIG_ONDEMAND_NO,
 436                 global_enable_performance_for_mountpoints = CONFIG_ONDEMAND_NO,
 437                 global_enable_performance_for_virtual_mountpoints = CONFIG_ONDEMAND_ONDEMAND,
 438                 global_do_io = CONFIG_ONDEMAND_ONDEMAND,
 439                 global_do_ops = CONFIG_ONDEMAND_ONDEMAND,
 440                 global_do_mops = CONFIG_ONDEMAND_ONDEMAND,
 441                 global_do_iotime = CONFIG_ONDEMAND_ONDEMAND,
 442                 global_do_qops = CONFIG_ONDEMAND_ONDEMAND,
 443                 global_do_util = CONFIG_ONDEMAND_ONDEMAND,
 444                 global_do_backlog = CONFIG_ONDEMAND_ONDEMAND,
 445                 globals_initialized = 0;
 446
 447     if(unlikely(!globals_initialized)) {
 448         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);
 449
 450         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);
 451         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);
 452         global_enable_performance_for_partitions = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "performance metrics for partitions", global_enable_performance_for_partitions);
 453         global_enable_performance_for_mountpoints = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "performance metrics for mounted filesystems", global_enable_performance_for_mountpoints);
 454         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);
 455
 456         global_do_io      = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "bandwidth for all disks", global_do_io);
 457         global_do_ops     = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "operations for all disks", global_do_ops);
 458         global_do_mops    = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "merged operations for all disks", global_do_mops);
 459         global_do_iotime  = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "i/o time for all disks", global_do_iotime);
 460         global_do_qops    = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "queued operations for all disks", global_do_qops);
 461         global_do_util    = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "utilization percentage for all disks", global_do_util);
 462         global_do_backlog = config_get_boolean_ondemand("plugin:proc:/proc/diskstats", "backlog for all disks", global_do_backlog);
 463
 464         globals_initialized = 1;
 465     }
 466
 467     if(!ff) {
 468         char filename[FILENAME_MAX + 1];
 469         snprintfz(filename, FILENAME_MAX, "%s%s", global_host_prefix, "/proc/diskstats");
 470         ff = procfile_open(config_get("plugin:proc:/proc/diskstats", "filename to monitor", filename), " \t", PROCFILE_FLAG_DEFAULT);
 471     }
 472     if(!ff) return 1;
 473
 474     ff = procfile_readall(ff);
 475     if(!ff) return 0; // we return 0, so that we will retry to open it next time
 476
 477     uint32_t lines = procfile_lines(ff), l;
 478     uint32_t words;
 479
 480     // this is smart enough not to reload it every time
 481     mountinfo_reload(0);
 482
 483     for(l = 0; l < lines ;l++) {
 484         // --------------------------------------------------------------------------
 485         // Read parameters
 486
 487         char *disk;
 488         unsigned long       major = 0, minor = 0;
 489
 490         collected_number    reads = 0,  mreads = 0,  readsectors = 0,  readms = 0,
 491                             writes = 0, mwrites = 0, writesectors = 0, writems = 0,
 492                             queued_ios = 0, busy_ms = 0, backlog_ms = 0;
 493
 494         collected_number    last_reads = 0,  last_readsectors = 0,  last_readms = 0,
 495                             last_writes = 0, last_writesectors = 0, last_writems = 0,
 496                             last_busy_ms = 0;
 497
 498         words = procfile_linewords(ff, l);
 499         if(words < 14) continue;
 500
 501         major           = strtoul(procfile_lineword(ff, l, 0), NULL, 10);
 502         minor           = strtoul(procfile_lineword(ff, l, 1), NULL, 10);
 503         disk            = procfile_lineword(ff, l, 2);
 504
 505         // # of reads completed # of writes completed
 506         // This is the total number of reads or writes completed successfully.
 507         reads           = strtoull(procfile_lineword(ff, l, 3), NULL, 10);  // rd_ios
 508         writes          = strtoull(procfile_lineword(ff, l, 7), NULL, 10);  // wr_ios
 509
 510         // # of reads merged # of writes merged
 511         // Reads and writes which are adjacent to each other may be merged for
 512         // efficiency.  Thus two 4K reads may become one 8K read before it is
 513         // ultimately handed to the disk, and so it will be counted (and queued)
 514         mreads          = strtoull(procfile_lineword(ff, l, 4), NULL, 10);  // rd_merges_or_rd_sec
 515         mwrites         = strtoull(procfile_lineword(ff, l, 8), NULL, 10);  // wr_merges
 516
 517         // # of sectors read # of sectors written
 518         // This is the total number of sectors read or written successfully.
 519         readsectors     = strtoull(procfile_lineword(ff, l, 5), NULL, 10);  // rd_sec_or_wr_ios
 520         writesectors    = strtoull(procfile_lineword(ff, l, 9), NULL, 10);  // wr_sec
 521
 522         // # of milliseconds spent reading # of milliseconds spent writing
 523         // This is the total number of milliseconds spent by all reads or writes (as
 524         // measured from __make_request() to end_that_request_last()).
 525         readms          = strtoull(procfile_lineword(ff, l, 6), NULL, 10);  // rd_ticks_or_wr_sec
 526         writems         = strtoull(procfile_lineword(ff, l, 10), NULL, 10); // wr_ticks
 527
 528         // # of I/Os currently in progress
 529         // The only field that should go to zero. Incremented as requests are
 530         // given to appropriate struct request_queue and decremented as they finish.
 531         queued_ios      = strtoull(procfile_lineword(ff, l, 11), NULL, 10); // ios_pgr
 532
 533         // # of milliseconds spent doing I/Os
 534         // This field increases so long as field queued_ios is nonzero.
 535         busy_ms         = strtoull(procfile_lineword(ff, l, 12), NULL, 10); // tot_ticks
 536
 537         // weighted # of milliseconds spent doing I/Os
 538         // This field is incremented at each I/O start, I/O completion, I/O
 539         // merge, or read of these stats by the number of I/Os in progress
 540         // (field queued_ios) times the number of milliseconds spent doing I/O since the
 541         // last update of this field.  This can provide an easy measure of both
 542         // I/O completion time and the backlog that may be accumulating.
 543         backlog_ms      = strtoull(procfile_lineword(ff, l, 13), NULL, 10); // rq_ticks
 544
 545
 546         // --------------------------------------------------------------------------
 547         // remove slashes from disk names
 548         char *s;
 549         for(s = disk; *s ;s++)
 550             if(*s == '/') *s = '_';
 551
 552         // --------------------------------------------------------------------------
 553         // get a disk structure for the disk
 554
 555         struct disk *d = get_disk(major, minor, disk);
 556
 557
 558         // --------------------------------------------------------------------------
 559         // Set its family based on mount point
 560
 561         char *family = d->mount_point;
 562         if(!family) family = disk;
 563
 564
 565         // --------------------------------------------------------------------------
 566         // Check the configuration for the device
 567
 568         if(unlikely(!d->configured)) {
 569             char var_name[4096 + 1];
 570             snprintfz(var_name, 4096, "plugin:proc:/proc/diskstats:%s", disk);
 571
 572             int def_enable = config_get_boolean_ondemand(var_name, "enable", global_enable_new_disks_detected_at_runtime);
 573             if(def_enable == CONFIG_ONDEMAND_NO) {
 574                 // the user does not want any metrics for this disk
 575                 d->do_io = CONFIG_ONDEMAND_NO;
 576                 d->do_ops = CONFIG_ONDEMAND_NO;
 577                 d->do_mops = CONFIG_ONDEMAND_NO;
 578                 d->do_iotime = CONFIG_ONDEMAND_NO;
 579                 d->do_qops = CONFIG_ONDEMAND_NO;
 580                 d->do_util = CONFIG_ONDEMAND_NO;
 581                 d->do_backlog = CONFIG_ONDEMAND_NO;
 582                 d->do_space = CONFIG_ONDEMAND_NO;
 583                 d->do_inodes = CONFIG_ONDEMAND_NO;
 584             }
 585             else {
 586                 // this disk is enabled
 587                 // check its direct settings
 588
 589                 int def_performance = CONFIG_ONDEMAND_ONDEMAND;
 590                 int def_space = (d->mount_point)?CONFIG_ONDEMAND_ONDEMAND:CONFIG_ONDEMAND_NO;
 591
 592                 // since this is 'on demand' we can figure the performance settings
 593                 // based on the type of disk
 594
 595                 switch(d->type) {
 596                     case DISK_TYPE_PHYSICAL:
 597                         def_performance = global_enable_performance_for_physical_disks;
 598                         break;
 599
 600                     case DISK_TYPE_PARTITION:
 601                         def_performance = global_enable_performance_for_partitions;
 602                         break;
 603
 604                     case DISK_TYPE_CONTAINER:
 605                         def_performance = global_enable_performance_for_virtual_disks;
 606
 607                         if(d->mount_point)
 608                             def_performance = select_positive_option(def_performance, global_enable_performance_for_virtual_mountpoints);
 609                         break;
 610                 }
 611
 612                 if(d->mount_point)
 613                     def_performance = select_positive_option(def_performance, global_enable_performance_for_mountpoints);
 614
 615                 // ------------------------------------------------------------
 616                 // now we have def_performance and def_space
 617                 // to work further
 618
 619                 // def_performance
 620                 // check the user configuration (this will also show our 'on demand' decision)
 621                 def_performance = config_get_boolean_ondemand(var_name, "enable performance metrics", def_performance);
 622
 623                 int ddo_io = CONFIG_ONDEMAND_NO,
 624                     ddo_ops = CONFIG_ONDEMAND_NO,
 625                     ddo_mops = CONFIG_ONDEMAND_NO,
 626                     ddo_iotime = CONFIG_ONDEMAND_NO,
 627                     ddo_qops = CONFIG_ONDEMAND_NO,
 628                     ddo_util = CONFIG_ONDEMAND_NO,
 629                     ddo_backlog = CONFIG_ONDEMAND_NO;
 630
 631                 // we enable individual performance charts only when def_performance is not disabled
 632                 if(def_performance != CONFIG_ONDEMAND_NO) {
 633                     ddo_io = global_do_io,
 634                     ddo_ops = global_do_ops,
 635                     ddo_mops = global_do_mops,
 636                     ddo_iotime = global_do_iotime,
 637                     ddo_qops = global_do_qops,
 638                     ddo_util = global_do_util,
 639                     ddo_backlog = global_do_backlog;
 640                 }
 641
 642                 d->do_io      = config_get_boolean_ondemand(var_name, "bandwidth", ddo_io);
 643                 d->do_ops     = config_get_boolean_ondemand(var_name, "operations", ddo_ops);
 644                 d->do_mops    = config_get_boolean_ondemand(var_name, "merged operations", ddo_mops);
 645                 d->do_iotime  = config_get_boolean_ondemand(var_name, "i/o time", ddo_iotime);
 646                 d->do_qops    = config_get_boolean_ondemand(var_name, "queued operations", ddo_qops);
 647                 d->do_util    = config_get_boolean_ondemand(var_name, "utilization percentage", ddo_util);
 648                 d->do_backlog = config_get_boolean_ondemand(var_name, "backlog", ddo_backlog);
 649
 650                 // def_space
 651                 if(d->mount_point) {
 652                     // check the user configuration (this will also show our 'on demand' decision)
 653                     def_space = config_get_boolean_ondemand(var_name, "enable space metrics", def_space);
 654
 655                     int ddo_space = def_space,
 656                         ddo_inodes = def_space;
 657
 658                     d->do_space = config_get_boolean_ondemand(var_name, "space usage", ddo_space);
 659                     d->do_inodes = config_get_boolean_ondemand(var_name, "inodes usage", ddo_inodes);
 660                 }
 661                 else {
 662                     // don't show settings for this disk
 663                     d->do_space = CONFIG_ONDEMAND_NO;
 664                     d->do_inodes = CONFIG_ONDEMAND_NO;
 665                 }
 666             }
 667
 668             d->configured = 1;
 669         }
 670
 671         RRDSET *st;
 672
 673         // --------------------------------------------------------------------------
 674         // Do performance metrics
 675
 676         if(d->do_io == CONFIG_ONDEMAND_YES || (d->do_io == CONFIG_ONDEMAND_ONDEMAND && (readsectors || writesectors))) {
 677             d->do_io = CONFIG_ONDEMAND_YES;
 678
 679             st = rrdset_find_bytype(RRD_TYPE_DISK, disk);
 680             if(!st) {
 681                 st = rrdset_create(RRD_TYPE_DISK, disk, NULL, family, "disk.io", "Disk I/O Bandwidth", "kilobytes/s", 2000, update_every, RRDSET_TYPE_AREA);
 682
 683                 rrddim_add(st, "reads", NULL, d->sector_size, 1024, RRDDIM_INCREMENTAL);
 684                 rrddim_add(st, "writes", NULL, d->sector_size * -1, 1024, RRDDIM_INCREMENTAL);
 685             }
 686             else rrdset_next_usec(st, dt);
 687
 688             last_readsectors  = rrddim_set(st, "reads", readsectors);
 689             last_writesectors = rrddim_set(st, "writes", writesectors);
 690             rrdset_done(st);
 691         }
 692
 693         // --------------------------------------------------------------------
 694
 695         if(d->do_ops == CONFIG_ONDEMAND_YES || (d->do_ops == CONFIG_ONDEMAND_ONDEMAND && (reads || writes))) {
 696             d->do_ops = CONFIG_ONDEMAND_YES;
 697
 698             st = rrdset_find_bytype("disk_ops", disk);
 699             if(!st) {
 700                 st = rrdset_create("disk_ops", disk, NULL, family, "disk.ops", "Disk Completed I/O Operations", "operations/s", 2001, update_every, RRDSET_TYPE_LINE);
 701                 st->isdetail = 1;
 702
 703                 rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_INCREMENTAL);
 704                 rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_INCREMENTAL);
 705             }
 706             else rrdset_next_usec(st, dt);
 707
 708             last_reads  = rrddim_set(st, "reads", reads);
 709             last_writes = rrddim_set(st, "writes", writes);
 710             rrdset_done(st);
 711         }
 712
 713         // --------------------------------------------------------------------
 714
 715         if(d->do_qops == CONFIG_ONDEMAND_YES || (d->do_qops == CONFIG_ONDEMAND_ONDEMAND && queued_ios)) {
 716             d->do_qops = CONFIG_ONDEMAND_YES;
 717
 718             st = rrdset_find_bytype("disk_qops", disk);
 719             if(!st) {
 720                 st = rrdset_create("disk_qops", disk, NULL, family, "disk.qops", "Disk Current I/O Operations", "operations", 2002, update_every, RRDSET_TYPE_LINE);
 721                 st->isdetail = 1;
 722
 723                 rrddim_add(st, "operations", NULL, 1, 1, RRDDIM_ABSOLUTE);
 724             }
 725             else rrdset_next_usec(st, dt);
 726
 727             rrddim_set(st, "operations", queued_ios);
 728             rrdset_done(st);
 729         }
 730
 731         // --------------------------------------------------------------------
 732
 733         if(d->do_backlog == CONFIG_ONDEMAND_YES || (d->do_backlog == CONFIG_ONDEMAND_ONDEMAND && backlog_ms)) {
 734             d->do_backlog = CONFIG_ONDEMAND_YES;
 735
 736             st = rrdset_find_bytype("disk_backlog", disk);
 737             if(!st) {
 738                 st = rrdset_create("disk_backlog", disk, NULL, family, "disk.backlog", "Disk Backlog", "backlog (ms)", 2003, update_every, RRDSET_TYPE_AREA);
 739                 st->isdetail = 1;
 740
 741                 rrddim_add(st, "backlog", NULL, 1, 10, RRDDIM_INCREMENTAL);
 742             }
 743             else rrdset_next_usec(st, dt);
 744
 745             rrddim_set(st, "backlog", backlog_ms);
 746             rrdset_done(st);
 747         }
 748
 749         // --------------------------------------------------------------------
 750
 751         if(d->do_util == CONFIG_ONDEMAND_YES || (d->do_util == CONFIG_ONDEMAND_ONDEMAND && busy_ms)) {
 752             d->do_util = CONFIG_ONDEMAND_YES;
 753
 754             st = rrdset_find_bytype("disk_util", disk);
 755             if(!st) {
 756                 st = rrdset_create("disk_util", disk, NULL, family, "disk.util", "Disk Utilization Time", "% of time working", 2004, update_every, RRDSET_TYPE_AREA);
 757                 st->isdetail = 1;
 758
 759                 rrddim_add(st, "utilization", NULL, 1, 10, RRDDIM_INCREMENTAL);
 760             }
 761             else rrdset_next_usec(st, dt);
 762
 763             last_busy_ms = rrddim_set(st, "utilization", busy_ms);
 764             rrdset_done(st);
 765         }
 766
 767         // --------------------------------------------------------------------
 768
 769         if(d->do_mops == CONFIG_ONDEMAND_YES || (d->do_mops == CONFIG_ONDEMAND_ONDEMAND && (mreads || mwrites))) {
 770             d->do_mops = CONFIG_ONDEMAND_YES;
 771
 772             st = rrdset_find_bytype("disk_mops", disk);
 773             if(!st) {
 774                 st = rrdset_create("disk_mops", disk, NULL, family, "disk.mops", "Disk Merged Operations", "merged operations/s", 2021, update_every, RRDSET_TYPE_LINE);
 775                 st->isdetail = 1;
 776
 777                 rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_INCREMENTAL);
 778                 rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_INCREMENTAL);
 779             }
 780             else rrdset_next_usec(st, dt);
 781
 782             rrddim_set(st, "reads", mreads);
 783             rrddim_set(st, "writes", mwrites);
 784             rrdset_done(st);
 785         }
 786
 787         // --------------------------------------------------------------------
 788
 789         if(d->do_iotime == CONFIG_ONDEMAND_YES || (d->do_iotime == CONFIG_ONDEMAND_ONDEMAND && (readms || writems))) {
 790             d->do_iotime = CONFIG_ONDEMAND_YES;
 791
 792             st = rrdset_find_bytype("disk_iotime", disk);
 793             if(!st) {
 794                 st = rrdset_create("disk_iotime", disk, NULL, family, "disk.iotime", "Disk Total I/O Time", "milliseconds/s", 2022, update_every, RRDSET_TYPE_LINE);
 795                 st->isdetail = 1;
 796
 797                 rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_INCREMENTAL);
 798                 rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_INCREMENTAL);
 799             }
 800             else rrdset_next_usec(st, dt);
 801
 802             last_readms  = rrddim_set(st, "reads", readms);
 803             last_writems = rrddim_set(st, "writes", writems);
 804             rrdset_done(st);
 805         }
 806
 807         // --------------------------------------------------------------------
 808         // calculate differential charts
 809         // only if this is not the first time we run
 810
 811         if(dt) {
 812             if( (d->do_iotime == CONFIG_ONDEMAND_YES || (d->do_iotime == CONFIG_ONDEMAND_ONDEMAND && (readms || writems))) &&
 813                 (d->do_ops    == CONFIG_ONDEMAND_YES || (d->do_ops    == CONFIG_ONDEMAND_ONDEMAND && (reads || writes)))) {
 814                 st = rrdset_find_bytype("disk_await", disk);
 815                 if(!st) {
 816                     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);
 817                     st->isdetail = 1;
 818
 819                     rrddim_add(st, "reads", NULL, 1, 1, RRDDIM_ABSOLUTE);
 820                     rrddim_add(st, "writes", NULL, -1, 1, RRDDIM_ABSOLUTE);
 821                 }
 822                 else rrdset_next_usec(st, dt);
 823
 824                 rrddim_set(st, "reads", (reads - last_reads) ? (readms - last_readms) / (reads - last_reads) : 0);
 825                 rrddim_set(st, "writes", (writes - last_writes) ? (writems - last_writems) / (writes - last_writes) : 0);
 826                 rrdset_done(st);
 827             }
 828
 829             if( (d->do_io  == CONFIG_ONDEMAND_YES || (d->do_io  == CONFIG_ONDEMAND_ONDEMAND && (readsectors || writesectors))) &&
 830                 (d->do_ops == CONFIG_ONDEMAND_YES || (d->do_ops == CONFIG_ONDEMAND_ONDEMAND && (reads || writes)))) {
 831                 st = rrdset_find_bytype("disk_avgsz", disk);
 832                 if(!st) {
 833                     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);
 834                     st->isdetail = 1;
 835
 836                     rrddim_add(st, "reads", NULL, d->sector_size, 1024, RRDDIM_ABSOLUTE);
 837                     rrddim_add(st, "writes", NULL, d->sector_size * -1, 1024, RRDDIM_ABSOLUTE);
 838                 }
 839                 else rrdset_next_usec(st, dt);
 840
 841                 rrddim_set(st, "reads", (reads - last_reads) ? (readsectors - last_readsectors) / (reads - last_reads) : 0);
 842                 rrddim_set(st, "writes", (writes - last_writes) ? (writesectors - last_writesectors) / (writes - last_writes) : 0);
 843                 rrdset_done(st);
 844             }
 845
 846             if( (d->do_util == CONFIG_ONDEMAND_YES || (d->do_util == CONFIG_ONDEMAND_ONDEMAND && busy_ms)) &&
 847                 (d->do_ops  == CONFIG_ONDEMAND_YES || (d->do_ops  == CONFIG_ONDEMAND_ONDEMAND && (reads || writes)))) {
 848                 st = rrdset_find_bytype("disk_svctm", disk);
 849                 if(!st) {
 850                     st = rrdset_create("disk_svctm", disk, NULL, family, "disk.svctm", "Average Service Time", "ms per operation", 2007, update_every, RRDSET_TYPE_LINE);
 851                     st->isdetail = 1;
 852
 853                     rrddim_add(st, "svctm", NULL, 1, 1, RRDDIM_ABSOLUTE);
 854                 }
 855                 else rrdset_next_usec(st, dt);
 856
 857                 rrddim_set(st, "svctm", ((reads - last_reads) + (writes - last_writes)) ? (busy_ms - last_busy_ms) / ((reads - last_reads) + (writes - last_writes)) : 0);
 858                 rrdset_done(st);
 859             }
 860         }
 861
 862 /*
 863         // --------------------------------------------------------------------------
 864         // space metrics
 865
 866         if(d->mount_point && (d->do_space || d->do_inodes) ) {
 867             do_disk_space_stats(d, d->mount_point, disk, disk, family, update_every, dt);
 868         }
 869 */
 870     }
 871
 872     // --------------------------------------------------------------------------
 873     // space metrics for non-block devices
 874
 875     struct mountinfo *mi;
 876     for(mi = disk_mountinfo_root; mi ;mi = mi->next) {
 877         if(unlikely(mi->flags & (MOUNTINFO_IS_DUMMY|MOUNTINFO_IS_BIND|MOUNTINFO_IS_SAME_DEV|MOUNTINFO_NO_STAT|MOUNTINFO_NO_SIZE)))
 878             continue;
 879
 880 /*
 881         // skip the ones with block devices
 882         int skip = 0;
 883         struct disk *d;
 884         for(d = disk_root; d ;d = d->next) {
 885             if(unlikely(d->mount_point && mi->mount_point_hash == d->mount_point_hash && strcmp(mi->mount_point, d->mount_point))) {
 886                 skip = 1;
 887                 break;
 888             }
 889         }
 890
 891         if(unlikely(skip))
 892             continue;
 893
 894         // fprintf(stderr, "Will process mount point '%s', source '%s', filesystem '%s'\n", mi->mount_point, mi->mount_source, mi->filesystem);
 895 */
 896
 897         do_disk_space_stats(NULL, mi->mount_point, mi->mount_source, mi->persistent_id, mi->mount_point , update_every, dt);
 898     }
 899
 900     return 0;
 901 }