src/disk.c

   1 /**
   2  * collectd - src/disk.c
   3  * Copyright (C) 2005-2012  Florian octo Forster
   4  * Copyright (C) 2009       Manuel Sanmartin
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the
   8  * Free Software Foundation; only version 2 of the License is applicable.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along
  16  * with this program; if not, write to the Free Software Foundation, Inc.,
  17  * 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
  18  *
  19  * Authors:
  20  *   Florian octo Forster <octo at collectd.org>
  21  *   Manuel Sanmartin
  22  **/
  23
  24 #include "collectd.h"
  25 #include "common.h"
  26 #include "plugin.h"
  27 #include "utils_ignorelist.h"
  28
  29 #if HAVE_MACH_MACH_TYPES_H
  30 #  include <mach/mach_types.h>
  31 #endif
  32 #if HAVE_MACH_MACH_INIT_H
  33 #  include <mach/mach_init.h>
  34 #endif
  35 #if HAVE_MACH_MACH_ERROR_H
  36 #  include <mach/mach_error.h>
  37 #endif
  38 #if HAVE_MACH_MACH_PORT_H
  39 #  include <mach/mach_port.h>
  40 #endif
  41 #if HAVE_COREFOUNDATION_COREFOUNDATION_H
  42 #  include <CoreFoundation/CoreFoundation.h>
  43 #endif
  44 #if HAVE_IOKIT_IOKITLIB_H
  45 #  include <IOKit/IOKitLib.h>
  46 #endif
  47 #if HAVE_IOKIT_IOTYPES_H
  48 #  include <IOKit/IOTypes.h>
  49 #endif
  50 #if HAVE_IOKIT_STORAGE_IOBLOCKSTORAGEDRIVER_H
  51 #  include <IOKit/storage/IOBlockStorageDriver.h>
  52 #endif
  53 #if HAVE_IOKIT_IOBSD_H
  54 #  include <IOKit/IOBSD.h>
  55 #endif
  56 #if KERNEL_FREEBSD
  57 #include <devstat.h>
  58 #include <libgeom.h>
  59 #endif
  60
  61 #if HAVE_LIMITS_H
  62 # include <limits.h>
  63 #endif
  64 #ifndef UINT_MAX
  65 #  define UINT_MAX 4294967295U
  66 #endif
  67
  68 #if HAVE_STATGRAB_H
  69 # include <statgrab.h>
  70 #endif
  71
  72 #if HAVE_PERFSTAT
  73 # ifndef _AIXVERSION_610
  74 # include <sys/systemcfg.h>
  75 # endif
  76 # include <sys/protosw.h>
  77 # include <libperfstat.h>
  78 #endif
  79
  80 #if HAVE_IOKIT_IOKITLIB_H
  81 static mach_port_t io_master_port = MACH_PORT_NULL;
  82 /* This defaults to false for backwards compatibility. Please fix in the next
  83  * major version. */
  84 static _Bool use_bsd_name = 0;
  85 /* #endif HAVE_IOKIT_IOKITLIB_H */
  86
  87 #elif KERNEL_LINUX
  88 typedef struct diskstats
  89 {
  90         char *name;
  91
  92         /* This overflows in roughly 1361 years */
  93         unsigned int poll_count;
  94
  95         derive_t read_sectors;
  96         derive_t write_sectors;
  97
  98         derive_t read_bytes;
  99         derive_t write_bytes;
 100
 101         derive_t read_ops;
 102         derive_t write_ops;
 103         derive_t read_time;
 104         derive_t write_time;
 105
 106         derive_t avg_read_time;
 107         derive_t avg_write_time;
 108
 109         _Bool has_merged;
 110         _Bool has_in_progress;
 111         _Bool has_io_time;
 112
 113         struct diskstats *next;
 114 } diskstats_t;
 115
 116 static diskstats_t *disklist;
 117 /* #endif KERNEL_LINUX */
 118 #elif KERNEL_FREEBSD
 119 static struct gmesh geom_tree;
 120 /* #endif KERNEL_FREEBSD */
 121
 122 #elif HAVE_LIBKSTAT
 123 #define MAX_NUMDISK 1024
 124 extern kstat_ctl_t *kc;
 125 static kstat_t *ksp[MAX_NUMDISK];
 126 static int numdisk = 0;
 127 /* #endif HAVE_LIBKSTAT */
 128
 129 #elif defined(HAVE_LIBSTATGRAB)
 130 /* #endif HAVE_LIBKSTATGRAB */
 131
 132 #elif HAVE_PERFSTAT
 133 static perfstat_disk_t * stat_disk;
 134 static int numdisk;
 135 static int pnumdisk;
 136 /* #endif HAVE_PERFSTAT */
 137
 138 #else
 139 # error "No applicable input method."
 140 #endif
 141
 142 #if HAVE_LIBUDEV
 143 #include <libudev.h>
 144
 145 static char *conf_udev_name_attr = NULL;
 146 static struct udev *handle_udev;
 147 #endif
 148
 149 static const char *config_keys[] =
 150 {
 151         "Disk",
 152         "UseBSDName",
 153         "IgnoreSelected",
 154         "UdevNameAttr"
 155 };
 156 static int config_keys_num = STATIC_ARRAY_SIZE (config_keys);
 157
 158 static ignorelist_t *ignorelist = NULL;
 159
 160 static int disk_config (const char *key, const char *value)
 161 {
 162   if (ignorelist == NULL)
 163     ignorelist = ignorelist_create (/* invert = */ 1);
 164   if (ignorelist == NULL)
 165     return (1);
 166
 167   if (strcasecmp ("Disk", key) == 0)
 168   {
 169     ignorelist_add (ignorelist, value);
 170   }
 171   else if (strcasecmp ("IgnoreSelected", key) == 0)
 172   {
 173     int invert = 1;
 174     if (IS_TRUE (value))
 175       invert = 0;
 176     ignorelist_set_invert (ignorelist, invert);
 177   }
 178   else if (strcasecmp ("UseBSDName", key) == 0)
 179   {
 180 #if HAVE_IOKIT_IOKITLIB_H
 181     use_bsd_name = IS_TRUE (value) ? 1 : 0;
 182 #else
 183     WARNING ("disk plugin: The \"UseBSDName\" option is only supported "
 184         "on Mach / Mac OS X and will be ignored.");
 185 #endif
 186   }
 187   else if (strcasecmp ("UdevNameAttr", key) == 0)
 188   {
 189 #if HAVE_LIBUDEV
 190     if (conf_udev_name_attr != NULL)
 191     {
 192       free (conf_udev_name_attr);
 193       conf_udev_name_attr = NULL;
 194     }
 195     if ((conf_udev_name_attr = strdup (value)) == NULL)
 196       return (1);
 197 #else
 198     WARNING ("disk plugin: The \"UdevNameAttr\" option is only supported "
 199         "if collectd is built with libudev support");
 200 #endif
 201   }
 202   else
 203   {
 204     return (-1);
 205   }
 206
 207   return (0);
 208 } /* int disk_config */
 209
 210 static int disk_init (void)
 211 {
 212 #if HAVE_IOKIT_IOKITLIB_H
 213         kern_return_t status;
 214
 215         if (io_master_port != MACH_PORT_NULL)
 216         {
 217                 mach_port_deallocate (mach_task_self (),
 218                                 io_master_port);
 219                 io_master_port = MACH_PORT_NULL;
 220         }
 221
 222         status = IOMasterPort (MACH_PORT_NULL, &io_master_port);
 223         if (status != kIOReturnSuccess)
 224         {
 225                 ERROR ("IOMasterPort failed: %s",
 226                                 mach_error_string (status));
 227                 io_master_port = MACH_PORT_NULL;
 228                 return (-1);
 229         }
 230 /* #endif HAVE_IOKIT_IOKITLIB_H */
 231
 232 #elif KERNEL_LINUX
 233         /* do nothing */
 234 /* #endif KERNEL_LINUX */
 235
 236 #elif KERNEL_FREEBSD
 237         int rv;
 238
 239         rv = geom_gettree(&geom_tree);
 240         if (rv != 0) {
 241                 ERROR ("geom_gettree() failed, returned %d", rv);
 242                 return (-1);
 243         }
 244         rv = geom_stats_open();
 245         if (rv != 0) {
 246                 ERROR ("geom_stats_open() failed, returned %d", rv);
 247                 return (-1);
 248         }
 249 /* #endif KERNEL_FREEBSD */
 250
 251 #elif HAVE_LIBKSTAT
 252         kstat_t *ksp_chain;
 253
 254         numdisk = 0;
 255
 256         if (kc == NULL)
 257                 return (-1);
 258
 259         for (numdisk = 0, ksp_chain = kc->kc_chain;
 260                         (numdisk < MAX_NUMDISK) && (ksp_chain != NULL);
 261                         ksp_chain = ksp_chain->ks_next)
 262         {
 263                 if (strncmp (ksp_chain->ks_class, "disk", 4)
 264                                 && strncmp (ksp_chain->ks_class, "partition", 9))
 265                         continue;
 266                 if (ksp_chain->ks_type != KSTAT_TYPE_IO)
 267                         continue;
 268                 ksp[numdisk++] = ksp_chain;
 269         }
 270 #endif /* HAVE_LIBKSTAT */
 271
 272         return (0);
 273 } /* int disk_init */
 274
 275 static void disk_submit (const char *plugin_instance,
 276                 const char *type,
 277                 derive_t read, derive_t write)
 278 {
 279         value_t values[2];
 280         value_list_t vl = VALUE_LIST_INIT;
 281
 282         /* Both `ignorelist' and `plugin_instance' may be NULL. */
 283         if (ignorelist_match (ignorelist, plugin_instance) != 0)
 284           return;
 285
 286         values[0].derive = read;
 287         values[1].derive = write;
 288
 289         vl.values = values;
 290         vl.values_len = 2;
 291         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
 292         sstrncpy (vl.plugin, "disk", sizeof (vl.plugin));
 293         sstrncpy (vl.plugin_instance, plugin_instance,
 294                         sizeof (vl.plugin_instance));
 295         sstrncpy (vl.type, type, sizeof (vl.type));
 296
 297         plugin_dispatch_values (&vl);
 298 } /* void disk_submit */
 299
 300 #if KERNEL_FREEBSD || KERNEL_LINUX
 301 static void submit_io_time (char const *plugin_instance, derive_t io_time, derive_t weighted_time)
 302 {
 303         value_t values[2];
 304         value_list_t vl = VALUE_LIST_INIT;
 305
 306         if (ignorelist_match (ignorelist, plugin_instance) != 0)
 307           return;
 308
 309         values[0].derive = io_time;
 310         values[1].derive = weighted_time;
 311
 312         vl.values = values;
 313         vl.values_len = 2;
 314         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
 315         sstrncpy (vl.plugin, "disk", sizeof (vl.plugin));
 316         sstrncpy (vl.plugin_instance, plugin_instance, sizeof (vl.plugin_instance));
 317         sstrncpy (vl.type, "disk_io_time", sizeof (vl.type));
 318
 319         plugin_dispatch_values (&vl);
 320 } /* void submit_io_time */
 321 #endif /* KERNEL_FREEBSD || KERNEL_LINUX */
 322
 323 #if KERNEL_LINUX
 324 static void submit_in_progress (char const *disk_name, gauge_t in_progress)
 325 {
 326         value_t v;
 327         value_list_t vl = VALUE_LIST_INIT;
 328
 329         if (ignorelist_match (ignorelist, disk_name) != 0)
 330           return;
 331
 332         v.gauge = in_progress;
 333
 334         vl.values = &v;
 335         vl.values_len = 1;
 336         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
 337         sstrncpy (vl.plugin, "disk", sizeof (vl.plugin));
 338         sstrncpy (vl.plugin_instance, disk_name, sizeof (vl.plugin_instance));
 339         sstrncpy (vl.type, "pending_operations", sizeof (vl.type));
 340
 341         plugin_dispatch_values (&vl);
 342 }
 343
 344
 345 static counter_t disk_calc_time_incr (counter_t delta_time, counter_t delta_ops)
 346 {
 347         double interval = CDTIME_T_TO_DOUBLE (plugin_get_interval ());
 348         double avg_time = ((double) delta_time) / ((double) delta_ops);
 349         double avg_time_incr = interval * avg_time;
 350
 351         return ((counter_t) (avg_time_incr + .5));
 352 }
 353 #endif
 354
 355 #if HAVE_LIBUDEV
 356 /**
 357  * Attempt to provide an rename disk instance from an assigned udev attribute.
 358  *
 359  * On success, it returns a strduped char* to the desired attribute value.
 360  * Otherwise it returns NULL.
 361  */
 362
 363 static char *disk_udev_attr_name (struct udev *udev, char *disk_name, const char *attr)
 364 {
 365         struct udev_device *dev;
 366         const char *prop;
 367         char *output = NULL;
 368
 369         dev = udev_device_new_from_subsystem_sysname (udev, "block", disk_name);
 370         if (dev != NULL)
 371         {
 372                 prop = udev_device_get_property_value (dev, attr);
 373                 if (prop) {
 374                         output = strdup (prop);
 375                         DEBUG ("disk plugin: renaming %s => %s", disk_name, output);
 376                 }
 377                 udev_device_unref (dev);
 378         }
 379         return output;
 380 }
 381 #endif
 382
 383 #if HAVE_IOKIT_IOKITLIB_H
 384 static signed long long dict_get_value (CFDictionaryRef dict, const char *key)
 385 {
 386         signed long long val_int;
 387         CFNumberRef      val_obj;
 388         CFStringRef      key_obj;
 389
 390         /* `key_obj' needs to be released. */
 391         key_obj = CFStringCreateWithCString (kCFAllocatorDefault, key,
 392                         kCFStringEncodingASCII);
 393         if (key_obj == NULL)
 394         {
 395                 DEBUG ("CFStringCreateWithCString (%s) failed.", key);
 396                 return (-1LL);
 397         }
 398
 399         /* get => we don't need to release (== free) the object */
 400         val_obj = (CFNumberRef) CFDictionaryGetValue (dict, key_obj);
 401
 402         CFRelease (key_obj);
 403
 404         if (val_obj == NULL)
 405         {
 406                 DEBUG ("CFDictionaryGetValue (%s) failed.", key);
 407                 return (-1LL);
 408         }
 409
 410         if (!CFNumberGetValue (val_obj, kCFNumberSInt64Type, &val_int))
 411         {
 412                 DEBUG ("CFNumberGetValue (%s) failed.", key);
 413                 return (-1LL);
 414         }
 415
 416         return (val_int);
 417 }
 418 #endif /* HAVE_IOKIT_IOKITLIB_H */
 419
 420 static int disk_read (void)
 421 {
 422 #if HAVE_IOKIT_IOKITLIB_H
 423         io_registry_entry_t     disk;
 424         io_registry_entry_t     disk_child;
 425         io_iterator_t           disk_list;
 426         CFMutableDictionaryRef  props_dict, child_dict;
 427         CFDictionaryRef         stats_dict;
 428         CFStringRef             tmp_cf_string_ref;
 429         kern_return_t           status;
 430
 431         signed long long read_ops, read_byt, read_tme;
 432         signed long long write_ops, write_byt, write_tme;
 433
 434         int  disk_major, disk_minor;
 435         char disk_name[DATA_MAX_NAME_LEN];
 436         char child_disk_name_bsd[DATA_MAX_NAME_LEN], props_disk_name_bsd[DATA_MAX_NAME_LEN];
 437
 438         /* Get the list of all disk objects. */
 439         if (IOServiceGetMatchingServices (io_master_port, IOServiceMatching (kIOBlockStorageDriverClass), &disk_list) != kIOReturnSuccess) {
 440                 ERROR ("disk plugin: IOServiceGetMatchingServices failed.");
 441                 return (-1);
 442         }
 443
 444         while ((disk = IOIteratorNext (disk_list)) != 0) {
 445                 props_dict = NULL;
 446                 stats_dict = NULL;
 447                 child_dict = NULL;
 448
 449                 /* get child of disk entry and corresponding property dictionary */
 450                 if ((status = IORegistryEntryGetChildEntry (disk, kIOServicePlane, &disk_child)) != kIOReturnSuccess) {
 451                         /* This fails for example for DVD/CD drives, which we want to ignore anyway */
 452                         DEBUG ("IORegistryEntryGetChildEntry (disk) failed: 0x%08x", status);
 453                         IOObjectRelease (disk);
 454                         continue;
 455                 }
 456                 if (IORegistryEntryCreateCFProperties (disk_child, (CFMutableDictionaryRef *) &child_dict, kCFAllocatorDefault, kNilOptions) != kIOReturnSuccess || child_dict == NULL) {
 457                         ERROR ("disk plugin: IORegistryEntryCreateCFProperties (disk_child) failed.");
 458                         IOObjectRelease (disk_child);
 459                         IOObjectRelease (disk);
 460                         continue;
 461                 }
 462
 463                 /* extract name and major/minor numbers */
 464                 memset (child_disk_name_bsd, 0, sizeof (child_disk_name_bsd));
 465                 tmp_cf_string_ref = (CFStringRef) CFDictionaryGetValue (child_dict, CFSTR(kIOBSDNameKey));
 466                 if (tmp_cf_string_ref) {
 467                         assert (CFGetTypeID (tmp_cf_string_ref) == CFStringGetTypeID ());
 468                         CFStringGetCString (tmp_cf_string_ref, child_disk_name_bsd, sizeof (child_disk_name_bsd), kCFStringEncodingUTF8);
 469                 }
 470                 disk_major = (int) dict_get_value (child_dict, kIOBSDMajorKey);
 471                 disk_minor = (int) dict_get_value (child_dict, kIOBSDMinorKey);
 472                 DEBUG ("disk plugin: child_disk_name_bsd=\"%s\" major=%d minor=%d", child_disk_name_bsd, disk_major, disk_minor);
 473                 CFRelease (child_dict);
 474                 IOObjectRelease (disk_child);
 475
 476                 /* get property dictionary of the disk entry itself */
 477                 if (IORegistryEntryCreateCFProperties (disk, (CFMutableDictionaryRef *) &props_dict, kCFAllocatorDefault, kNilOptions) != kIOReturnSuccess || props_dict == NULL) {
 478                         ERROR ("disk-plugin: IORegistryEntryCreateCFProperties failed.");
 479                         IOObjectRelease (disk);
 480                         continue;
 481                 }
 482
 483                 /* extract name and stats dictionary */
 484                 memset (props_disk_name_bsd, 0, sizeof (props_disk_name_bsd));
 485                 tmp_cf_string_ref = (CFStringRef) CFDictionaryGetValue (props_dict, CFSTR(kIOBSDNameKey));
 486                 if (tmp_cf_string_ref) {
 487                         assert (CFGetTypeID (tmp_cf_string_ref) == CFStringGetTypeID ());
 488                         CFStringGetCString (tmp_cf_string_ref, props_disk_name_bsd, sizeof (props_disk_name_bsd), kCFStringEncodingUTF8);
 489                 }
 490                 stats_dict = (CFDictionaryRef) CFDictionaryGetValue (props_dict, CFSTR (kIOBlockStorageDriverStatisticsKey));
 491                 if (stats_dict == NULL) {
 492                         ERROR ("disk plugin: CFDictionaryGetValue (%s) failed.", kIOBlockStorageDriverStatisticsKey);
 493                         CFRelease (props_dict);
 494                         IOObjectRelease (disk);
 495                         continue;
 496                 }
 497                 DEBUG ("disk plugin: props_disk_name_bsd=\"%s\"", props_disk_name_bsd);
 498
 499                 /* choose name */
 500                 if (use_bsd_name) {
 501                         if (child_disk_name_bsd[0] != 0)
 502                                 sstrncpy (disk_name, child_disk_name_bsd, sizeof (disk_name));
 503                         else if (props_disk_name_bsd[0] != 0)
 504                                 sstrncpy (disk_name, props_disk_name_bsd, sizeof (disk_name));
 505                         else {
 506                                 ERROR ("disk plugin: can't find bsd disk name.");
 507                                 ssnprintf (disk_name, sizeof (disk_name), "%i-%i", disk_major, disk_minor);
 508                         }
 509                 }
 510                 else
 511                         ssnprintf (disk_name, sizeof (disk_name), "%i-%i", disk_major, disk_minor);
 512
 513                 /* extract the stats */
 514                 read_ops  = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsReadsKey);
 515                 read_byt  = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsBytesReadKey);
 516                 read_tme  = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsTotalReadTimeKey);
 517                 write_ops = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsWritesKey);
 518                 write_byt = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsBytesWrittenKey);
 519                 write_tme = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsTotalWriteTimeKey);
 520                 CFRelease (props_dict);
 521                 IOObjectRelease (disk);
 522
 523                 /* and submit */
 524                 DEBUG ("disk plugin: disk_name = \"%s\"", disk_name);
 525                 if ((read_byt != -1LL) || (write_byt != -1LL))
 526                         disk_submit (disk_name, "disk_octets", read_byt, write_byt);
 527                 if ((read_ops != -1LL) || (write_ops != -1LL))
 528                         disk_submit (disk_name, "disk_ops", read_ops, write_ops);
 529                 if ((read_tme != -1LL) || (write_tme != -1LL))
 530                         disk_submit (disk_name, "disk_time", read_tme / 1000, write_tme / 1000);
 531
 532         }
 533         IOObjectRelease (disk_list);
 534 /* #endif HAVE_IOKIT_IOKITLIB_H */
 535
 536 #elif KERNEL_FREEBSD
 537         int retry, dirty;
 538
 539         void *snap = NULL;
 540         struct devstat *snap_iter;
 541
 542         struct gident *geom_id;
 543
 544         const char *disk_name;
 545         long double read_time, write_time, busy_time, total_duration;
 546
 547         for (retry = 0, dirty = 1; retry < 5 && dirty == 1; retry++) {
 548                 if (snap != NULL)
 549                         geom_stats_snapshot_free(snap);
 550
 551                 /* Get a fresh copy of stats snapshot */
 552                 snap = geom_stats_snapshot_get();
 553                 if (snap == NULL) {
 554                         ERROR("disk plugin: geom_stats_snapshot_get() failed.");
 555                         return (-1);
 556                 }
 557
 558                 /* Check if we have dirty read from this snapshot */
 559                 dirty = 0;
 560                 geom_stats_snapshot_reset(snap);
 561                 while ((snap_iter = geom_stats_snapshot_next(snap)) != NULL) {
 562                         if (snap_iter->id == NULL)
 563                                 continue;
 564                         geom_id = geom_lookupid(&geom_tree, snap_iter->id);
 565
 566                         /* New device? refresh GEOM tree */
 567                         if (geom_id == NULL) {
 568                                 geom_deletetree(&geom_tree);
 569                                 if (geom_gettree(&geom_tree) != 0) {
 570                                         ERROR("disk plugin: geom_gettree() failed");
 571                                         geom_stats_snapshot_free(snap);
 572                                         return (-1);
 573                                 }
 574                                 geom_id = geom_lookupid(&geom_tree, snap_iter->id);
 575                         }
 576                         /*
 577                          * This should be rare: the device come right before we take the
 578                          * snapshot and went away right after it.  We will handle this
 579                          * case later, so don't mark dirty but silently ignore it.
 580                          */
 581                         if (geom_id == NULL)
 582                                 continue;
 583
 584                         /* Only collect PROVIDER data */
 585                         if (geom_id->lg_what != ISPROVIDER)
 586                                 continue;
 587
 588                         /* Only collect data when rank is 1 (physical devices) */
 589                         if (((struct gprovider *)(geom_id->lg_ptr))->lg_geom->lg_rank != 1)
 590                                 continue;
 591
 592                         /* Check if this is a dirty read quit for another try */
 593                         if (snap_iter->sequence0 != snap_iter->sequence1) {
 594                                 dirty = 1;
 595                                 break;
 596                         }
 597                 }
 598         }
 599
 600         /* Reset iterator */
 601         geom_stats_snapshot_reset(snap);
 602         for (;;) {
 603                 snap_iter = geom_stats_snapshot_next(snap);
 604                 if (snap_iter == NULL)
 605                         break;
 606
 607                 if (snap_iter->id == NULL)
 608                         continue;
 609                 geom_id = geom_lookupid(&geom_tree, snap_iter->id);
 610                 if (geom_id == NULL)
 611                         continue;
 612                 if (geom_id->lg_what != ISPROVIDER)
 613                         continue;
 614                 if (((struct gprovider *)(geom_id->lg_ptr))->lg_geom->lg_rank != 1)
 615                         continue;
 616                 /* Skip dirty reads, if present */
 617                 if (dirty && (snap_iter->sequence0 != snap_iter->sequence1))
 618                         continue;
 619
 620                 disk_name = ((struct gprovider *)geom_id->lg_ptr)->lg_name;
 621
 622                 if ((snap_iter->bytes[DEVSTAT_READ] != 0) || (snap_iter->bytes[DEVSTAT_WRITE] != 0)) {
 623                         disk_submit(disk_name, "disk_octets",
 624                                         (derive_t)snap_iter->bytes[DEVSTAT_READ],
 625                                         (derive_t)snap_iter->bytes[DEVSTAT_WRITE]);
 626                 }
 627
 628                 if ((snap_iter->operations[DEVSTAT_READ] != 0) || (snap_iter->operations[DEVSTAT_WRITE] != 0)) {
 629                         disk_submit(disk_name, "disk_ops",
 630                                         (derive_t)snap_iter->operations[DEVSTAT_READ],
 631                                         (derive_t)snap_iter->operations[DEVSTAT_WRITE]);
 632                 }
 633
 634                 read_time = devstat_compute_etime(&snap_iter->duration[DEVSTAT_READ], NULL);
 635                 write_time = devstat_compute_etime(&snap_iter->duration[DEVSTAT_WRITE], NULL);
 636                 if ((read_time != 0) || (write_time != 0)) {
 637                         disk_submit (disk_name, "disk_time",
 638                                         (derive_t)(read_time*1000), (derive_t)(write_time*1000));
 639                 }
 640                 if (devstat_compute_statistics(snap_iter, NULL, 1.0,
 641                     DSM_TOTAL_BUSY_TIME, &busy_time,
 642                     DSM_TOTAL_DURATION, &total_duration,
 643                     DSM_NONE) != 0) {
 644                         WARNING("%s", devstat_errbuf);
 645                 }
 646                 else
 647                 {
 648                         submit_io_time(disk_name, busy_time, total_duration);
 649                 }
 650         }
 651         geom_stats_snapshot_free(snap);
 652
 653 #elif KERNEL_LINUX
 654         FILE *fh;
 655         char buffer[1024];
 656
 657         char *fields[32];
 658         int numfields;
 659         int fieldshift = 0;
 660
 661         int minor = 0;
 662
 663         derive_t read_sectors  = 0;
 664         derive_t write_sectors = 0;
 665
 666         derive_t read_ops      = 0;
 667         derive_t read_merged   = 0;
 668         derive_t read_time     = 0;
 669         derive_t write_ops     = 0;
 670         derive_t write_merged  = 0;
 671         derive_t write_time    = 0;
 672         gauge_t in_progress    = NAN;
 673         derive_t io_time       = 0;
 674         derive_t weighted_time = 0;
 675         int is_disk = 0;
 676
 677         diskstats_t *ds, *pre_ds;
 678
 679         if ((fh = fopen ("/proc/diskstats", "r")) == NULL)
 680         {
 681                 fh = fopen ("/proc/partitions", "r");
 682                 if (fh == NULL)
 683                 {
 684                         ERROR ("disk plugin: fopen (/proc/{diskstats,partitions}) failed.");
 685                         return (-1);
 686                 }
 687
 688                 /* Kernel is 2.4.* */
 689                 fieldshift = 1;
 690         }
 691
 692 #if HAVE_LIBUDEV
 693         handle_udev = udev_new();
 694 #endif
 695
 696         while (fgets (buffer, sizeof (buffer), fh) != NULL)
 697         {
 698                 char *disk_name;
 699                 char *output_name;
 700
 701                 numfields = strsplit (buffer, fields, 32);
 702
 703                 if ((numfields != (14 + fieldshift)) && (numfields != 7))
 704                         continue;
 705
 706                 minor = atoll (fields[1]);
 707
 708                 disk_name = fields[2 + fieldshift];
 709
 710                 for (ds = disklist, pre_ds = disklist; ds != NULL; pre_ds = ds, ds = ds->next)
 711                         if (strcmp (disk_name, ds->name) == 0)
 712                                 break;
 713
 714                 if (ds == NULL)
 715                 {
 716                         if ((ds = (diskstats_t *) calloc (1, sizeof (diskstats_t))) == NULL)
 717                                 continue;
 718
 719                         if ((ds->name = strdup (disk_name)) == NULL)
 720                         {
 721                                 free (ds);
 722                                 continue;
 723                         }
 724
 725                         if (pre_ds == NULL)
 726                                 disklist = ds;
 727                         else
 728                                 pre_ds->next = ds;
 729                 }
 730
 731                 is_disk = 0;
 732                 if (numfields == 7)
 733                 {
 734                         /* Kernel 2.6, Partition */
 735                         read_ops      = atoll (fields[3]);
 736                         read_sectors  = atoll (fields[4]);
 737                         write_ops     = atoll (fields[5]);
 738                         write_sectors = atoll (fields[6]);
 739                 }
 740                 else if (numfields == (14 + fieldshift))
 741                 {
 742                         read_ops  =  atoll (fields[3 + fieldshift]);
 743                         write_ops =  atoll (fields[7 + fieldshift]);
 744
 745                         read_sectors  = atoll (fields[5 + fieldshift]);
 746                         write_sectors = atoll (fields[9 + fieldshift]);
 747
 748                         if ((fieldshift == 0) || (minor == 0))
 749                         {
 750                                 is_disk = 1;
 751                                 read_merged  = atoll (fields[4 + fieldshift]);
 752                                 read_time    = atoll (fields[6 + fieldshift]);
 753                                 write_merged = atoll (fields[8 + fieldshift]);
 754                                 write_time   = atoll (fields[10+ fieldshift]);
 755
 756                                 in_progress = atof (fields[11 + fieldshift]);
 757
 758                                 io_time       = atof (fields[12 + fieldshift]);
 759                                 weighted_time = atof (fields[13 + fieldshift]);
 760                         }
 761                 }
 762                 else
 763                 {
 764                         DEBUG ("numfields = %i; => unknown file format.", numfields);
 765                         continue;
 766                 }
 767
 768                 {
 769                         derive_t diff_read_sectors;
 770                         derive_t diff_write_sectors;
 771
 772                 /* If the counter wraps around, it's only 32 bits.. */
 773                         if (read_sectors < ds->read_sectors)
 774                                 diff_read_sectors = 1 + read_sectors
 775                                         + (UINT_MAX - ds->read_sectors);
 776                         else
 777                                 diff_read_sectors = read_sectors - ds->read_sectors;
 778                         if (write_sectors < ds->write_sectors)
 779                                 diff_write_sectors = 1 + write_sectors
 780                                         + (UINT_MAX - ds->write_sectors);
 781                         else
 782                                 diff_write_sectors = write_sectors - ds->write_sectors;
 783
 784                         ds->read_bytes += 512 * diff_read_sectors;
 785                         ds->write_bytes += 512 * diff_write_sectors;
 786                         ds->read_sectors = read_sectors;
 787                         ds->write_sectors = write_sectors;
 788                 }
 789
 790                 /* Calculate the average time an io-op needs to complete */
 791                 if (is_disk)
 792                 {
 793                         derive_t diff_read_ops;
 794                         derive_t diff_write_ops;
 795                         derive_t diff_read_time;
 796                         derive_t diff_write_time;
 797
 798                         if (read_ops < ds->read_ops)
 799                                 diff_read_ops = 1 + read_ops
 800                                         + (UINT_MAX - ds->read_ops);
 801                         else
 802                                 diff_read_ops = read_ops - ds->read_ops;
 803                         DEBUG ("disk plugin: disk_name = %s; read_ops = %"PRIi64"; "
 804                                         "ds->read_ops = %"PRIi64"; diff_read_ops = %"PRIi64";",
 805                                         disk_name,
 806                                         read_ops, ds->read_ops, diff_read_ops);
 807
 808                         if (write_ops < ds->write_ops)
 809                                 diff_write_ops = 1 + write_ops
 810                                         + (UINT_MAX - ds->write_ops);
 811                         else
 812                                 diff_write_ops = write_ops - ds->write_ops;
 813
 814                         if (read_time < ds->read_time)
 815                                 diff_read_time = 1 + read_time
 816                                         + (UINT_MAX - ds->read_time);
 817                         else
 818                                 diff_read_time = read_time - ds->read_time;
 819
 820                         if (write_time < ds->write_time)
 821                                 diff_write_time = 1 + write_time
 822                                         + (UINT_MAX - ds->write_time);
 823                         else
 824                                 diff_write_time = write_time - ds->write_time;
 825
 826                         if (diff_read_ops != 0)
 827                                 ds->avg_read_time += disk_calc_time_incr (
 828                                                 diff_read_time, diff_read_ops);
 829                         if (diff_write_ops != 0)
 830                                 ds->avg_write_time += disk_calc_time_incr (
 831                                                 diff_write_time, diff_write_ops);
 832
 833                         ds->read_ops = read_ops;
 834                         ds->read_time = read_time;
 835                         ds->write_ops = write_ops;
 836                         ds->write_time = write_time;
 837
 838                         if (read_merged || write_merged)
 839                                 ds->has_merged = 1;
 840
 841                         if (in_progress)
 842                                 ds->has_in_progress = 1;
 843
 844                         if (io_time)
 845                                 ds->has_io_time = 1;
 846
 847                 } /* if (is_disk) */
 848
 849                 /* Don't write to the RRDs if we've just started.. */
 850                 ds->poll_count++;
 851                 if (ds->poll_count <= 2)
 852                 {
 853                         DEBUG ("disk plugin: (ds->poll_count = %i) <= "
 854                                         "(min_poll_count = 2); => Not writing.",
 855                                         ds->poll_count);
 856                         continue;
 857                 }
 858
 859                 if ((read_ops == 0) && (write_ops == 0))
 860                 {
 861                         DEBUG ("disk plugin: ((read_ops == 0) && "
 862                                         "(write_ops == 0)); => Not writing.");
 863                         continue;
 864                 }
 865
 866                 output_name = disk_name;
 867
 868 #if HAVE_LIBUDEV
 869                 char *alt_name = disk_udev_attr_name (handle_udev, disk_name, conf_udev_name_attr);
 870                 if (alt_name != NULL)
 871                         output_name = alt_name;
 872 #endif
 873
 874                 if ((ds->read_bytes != 0) || (ds->write_bytes != 0))
 875                         disk_submit (output_name, "disk_octets",
 876                                         ds->read_bytes, ds->write_bytes);
 877
 878                 if ((ds->read_ops != 0) || (ds->write_ops != 0))
 879                         disk_submit (output_name, "disk_ops",
 880                                         read_ops, write_ops);
 881
 882                 if ((ds->avg_read_time != 0) || (ds->avg_write_time != 0))
 883                         disk_submit (output_name, "disk_time",
 884                                         ds->avg_read_time, ds->avg_write_time);
 885
 886                 if (is_disk)
 887                 {
 888                         if (ds->has_merged)
 889                                 disk_submit (output_name, "disk_merged",
 890                                         read_merged, write_merged);
 891                         if (ds->has_in_progress)
 892                                 submit_in_progress (output_name, in_progress);
 893                         if (ds->has_io_time)
 894                                 submit_io_time (output_name, io_time, weighted_time);
 895                 } /* if (is_disk) */
 896
 897 #if HAVE_LIBUDEV
 898                 /* release udev-based alternate name, if allocated */
 899                 sfree (alt_name);
 900 #endif
 901         } /* while (fgets (buffer, sizeof (buffer), fh) != NULL) */
 902
 903 #if HAVE_LIBUDEV
 904         udev_unref(handle_udev);
 905 #endif
 906
 907         fclose (fh);
 908 /* #endif defined(KERNEL_LINUX) */
 909
 910 #elif HAVE_LIBKSTAT
 911 # if HAVE_KSTAT_IO_T_WRITES && HAVE_KSTAT_IO_T_NWRITES && HAVE_KSTAT_IO_T_WTIME
 912 #  define KIO_ROCTETS reads
 913 #  define KIO_WOCTETS writes
 914 #  define KIO_ROPS    nreads
 915 #  define KIO_WOPS    nwrites
 916 #  define KIO_RTIME   rtime
 917 #  define KIO_WTIME   wtime
 918 # elif HAVE_KSTAT_IO_T_NWRITTEN && HAVE_KSTAT_IO_T_WRITES && HAVE_KSTAT_IO_T_WTIME
 919 #  define KIO_ROCTETS nread
 920 #  define KIO_WOCTETS nwritten
 921 #  define KIO_ROPS    reads
 922 #  define KIO_WOPS    writes
 923 #  define KIO_RTIME   rtime
 924 #  define KIO_WTIME   wtime
 925 # else
 926 #  error "kstat_io_t does not have the required members"
 927 # endif
 928         static kstat_io_t kio;
 929         int i;
 930
 931         if (kc == NULL)
 932                 return (-1);
 933
 934         for (i = 0; i < numdisk; i++)
 935         {
 936                 if (kstat_read (kc, ksp[i], &kio) == -1)
 937                         continue;
 938
 939                 if (strncmp (ksp[i]->ks_class, "disk", 4) == 0)
 940                 {
 941                         disk_submit (ksp[i]->ks_name, "disk_octets",
 942                                         kio.KIO_ROCTETS, kio.KIO_WOCTETS);
 943                         disk_submit (ksp[i]->ks_name, "disk_ops",
 944                                         kio.KIO_ROPS, kio.KIO_WOPS);
 945                         /* FIXME: Convert this to microseconds if necessary */
 946                         disk_submit (ksp[i]->ks_name, "disk_time",
 947                                         kio.KIO_RTIME, kio.KIO_WTIME);
 948                 }
 949                 else if (strncmp (ksp[i]->ks_class, "partition", 9) == 0)
 950                 {
 951                         disk_submit (ksp[i]->ks_name, "disk_octets",
 952                                         kio.KIO_ROCTETS, kio.KIO_WOCTETS);
 953                         disk_submit (ksp[i]->ks_name, "disk_ops",
 954                                         kio.KIO_ROPS, kio.KIO_WOPS);
 955                 }
 956         }
 957 /* #endif defined(HAVE_LIBKSTAT) */
 958
 959 #elif defined(HAVE_LIBSTATGRAB)
 960         sg_disk_io_stats *ds;
 961 # if HAVE_LIBSTATGRAB_0_90
 962         size_t disks;
 963 # else
 964         int disks;
 965 #endif
 966         int counter;
 967         char name[DATA_MAX_NAME_LEN];
 968
 969         if ((ds = sg_get_disk_io_stats(&disks)) == NULL)
 970                 return (0);
 971
 972         for (counter=0; counter < disks; counter++) {
 973                 strncpy(name, ds->disk_name, sizeof(name));
 974                 name[sizeof(name)-1] = '\0'; /* strncpy doesn't terminate longer strings */
 975                 disk_submit (name, "disk_octets", ds->read_bytes, ds->write_bytes);
 976                 ds++;
 977         }
 978 /* #endif defined(HAVE_LIBSTATGRAB) */
 979
 980 #elif defined(HAVE_PERFSTAT)
 981         derive_t read_sectors;
 982         derive_t write_sectors;
 983         derive_t read_time;
 984         derive_t write_time;
 985         derive_t read_ops;
 986         derive_t write_ops;
 987         perfstat_id_t firstpath;
 988         int rnumdisk;
 989         int i;
 990
 991         if ((numdisk = perfstat_disk(NULL, NULL, sizeof(perfstat_disk_t), 0)) < 0)
 992         {
 993                 char errbuf[1024];
 994                 WARNING ("disk plugin: perfstat_disk: %s",
 995                                 sstrerror (errno, errbuf, sizeof (errbuf)));
 996                 return (-1);
 997         }
 998
 999         if (numdisk != pnumdisk || stat_disk==NULL) {
1000                 if (stat_disk!=NULL)
1001                         free(stat_disk);
1002                 stat_disk = (perfstat_disk_t *)calloc(numdisk, sizeof(perfstat_disk_t));
1003         }
1004         pnumdisk = numdisk;
1005
1006         firstpath.name[0]='\0';
1007         if ((rnumdisk = perfstat_disk(&firstpath, stat_disk, sizeof(perfstat_disk_t), numdisk)) < 0)
1008         {
1009                 char errbuf[1024];
1010                 WARNING ("disk plugin: perfstat_disk : %s",
1011                                 sstrerror (errno, errbuf, sizeof (errbuf)));
1012                 return (-1);
1013         }
1014
1015         for (i = 0; i < rnumdisk; i++)
1016         {
1017                 read_sectors = stat_disk[i].rblks*stat_disk[i].bsize;
1018                 write_sectors = stat_disk[i].wblks*stat_disk[i].bsize;
1019                 disk_submit (stat_disk[i].name, "disk_octets", read_sectors, write_sectors);
1020
1021                 read_ops = stat_disk[i].xrate;
1022                 write_ops = stat_disk[i].xfers - stat_disk[i].xrate;
1023                 disk_submit (stat_disk[i].name, "disk_ops", read_ops, write_ops);
1024
1025                 read_time = stat_disk[i].rserv;
1026                 read_time *= ((double)(_system_configuration.Xint)/(double)(_system_configuration.Xfrac)) / 1000000.0;
1027                 write_time = stat_disk[i].wserv;
1028                 write_time *= ((double)(_system_configuration.Xint)/(double)(_system_configuration.Xfrac)) / 1000000.0;
1029                 disk_submit (stat_disk[i].name, "disk_time", read_time, write_time);
1030         }
1031 #endif /* defined(HAVE_PERFSTAT) */
1032
1033         return (0);
1034 } /* int disk_read */
1035
1036 void module_register (void)
1037 {
1038   plugin_register_config ("disk", disk_config,
1039       config_keys, config_keys_num);
1040   plugin_register_init ("disk", disk_init);
1041   plugin_register_read ("disk", disk_read);
1042 } /* void module_register */