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         values[0].derive = read;
 283         values[1].derive = write;
 284
 285         vl.values = values;
 286         vl.values_len = 2;
 287         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
 288         sstrncpy (vl.plugin, "disk", sizeof (vl.plugin));
 289         sstrncpy (vl.plugin_instance, plugin_instance,
 290                         sizeof (vl.plugin_instance));
 291         sstrncpy (vl.type, type, sizeof (vl.type));
 292
 293         plugin_dispatch_values (&vl);
 294 } /* void disk_submit */
 295
 296 #if KERNEL_FREEBSD || KERNEL_LINUX
 297 static void submit_io_time (char const *plugin_instance, derive_t io_time, derive_t weighted_time)
 298 {
 299         value_t values[2];
 300         value_list_t vl = VALUE_LIST_INIT;
 301
 302         values[0].derive = io_time;
 303         values[1].derive = weighted_time;
 304
 305         vl.values = values;
 306         vl.values_len = 2;
 307         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
 308         sstrncpy (vl.plugin, "disk", sizeof (vl.plugin));
 309         sstrncpy (vl.plugin_instance, plugin_instance, sizeof (vl.plugin_instance));
 310         sstrncpy (vl.type, "disk_io_time", sizeof (vl.type));
 311
 312         plugin_dispatch_values (&vl);
 313 } /* void submit_io_time */
 314 #endif /* KERNEL_FREEBSD || KERNEL_LINUX */
 315
 316 #if KERNEL_LINUX
 317 static void submit_in_progress (char const *disk_name, gauge_t in_progress)
 318 {
 319         value_t v;
 320         value_list_t vl = VALUE_LIST_INIT;
 321
 322         v.gauge = in_progress;
 323
 324         vl.values = &v;
 325         vl.values_len = 1;
 326         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
 327         sstrncpy (vl.plugin, "disk", sizeof (vl.plugin));
 328         sstrncpy (vl.plugin_instance, disk_name, sizeof (vl.plugin_instance));
 329         sstrncpy (vl.type, "pending_operations", sizeof (vl.type));
 330
 331         plugin_dispatch_values (&vl);
 332 }
 333
 334
 335 static counter_t disk_calc_time_incr (counter_t delta_time, counter_t delta_ops)
 336 {
 337         double interval = CDTIME_T_TO_DOUBLE (plugin_get_interval ());
 338         double avg_time = ((double) delta_time) / ((double) delta_ops);
 339         double avg_time_incr = interval * avg_time;
 340
 341         return ((counter_t) (avg_time_incr + .5));
 342 }
 343 #endif
 344
 345 #if HAVE_LIBUDEV
 346 /**
 347  * Attempt to provide an rename disk instance from an assigned udev attribute.
 348  *
 349  * On success, it returns a strduped char* to the desired attribute value.
 350  * Otherwise it returns NULL.
 351  */
 352
 353 static char *disk_udev_attr_name (struct udev *udev, char *disk_name, const char *attr)
 354 {
 355         struct udev_device *dev;
 356         const char *prop;
 357         char *output = NULL;
 358
 359         dev = udev_device_new_from_subsystem_sysname (udev, "block", disk_name);
 360         if (dev != NULL)
 361         {
 362                 prop = udev_device_get_property_value (dev, attr);
 363                 if (prop) {
 364                         output = strdup (prop);
 365                         DEBUG ("disk plugin: renaming %s => %s", disk_name, output);
 366                 }
 367                 udev_device_unref (dev);
 368         }
 369         return output;
 370 }
 371 #endif
 372
 373 #if HAVE_IOKIT_IOKITLIB_H
 374 static signed long long dict_get_value (CFDictionaryRef dict, const char *key)
 375 {
 376         signed long long val_int;
 377         CFNumberRef      val_obj;
 378         CFStringRef      key_obj;
 379
 380         /* `key_obj' needs to be released. */
 381         key_obj = CFStringCreateWithCString (kCFAllocatorDefault, key,
 382                         kCFStringEncodingASCII);
 383         if (key_obj == NULL)
 384         {
 385                 DEBUG ("CFStringCreateWithCString (%s) failed.", key);
 386                 return (-1LL);
 387         }
 388
 389         /* get => we don't need to release (== free) the object */
 390         val_obj = (CFNumberRef) CFDictionaryGetValue (dict, key_obj);
 391
 392         CFRelease (key_obj);
 393
 394         if (val_obj == NULL)
 395         {
 396                 DEBUG ("CFDictionaryGetValue (%s) failed.", key);
 397                 return (-1LL);
 398         }
 399
 400         if (!CFNumberGetValue (val_obj, kCFNumberSInt64Type, &val_int))
 401         {
 402                 DEBUG ("CFNumberGetValue (%s) failed.", key);
 403                 return (-1LL);
 404         }
 405
 406         return (val_int);
 407 }
 408 #endif /* HAVE_IOKIT_IOKITLIB_H */
 409
 410 static int disk_read (void)
 411 {
 412 #if HAVE_IOKIT_IOKITLIB_H
 413         io_registry_entry_t     disk;
 414         io_registry_entry_t     disk_child;
 415         io_iterator_t           disk_list;
 416         CFMutableDictionaryRef  props_dict, child_dict;
 417         CFDictionaryRef         stats_dict;
 418         CFStringRef             tmp_cf_string_ref;
 419         kern_return_t           status;
 420
 421         signed long long read_ops, read_byt, read_tme;
 422         signed long long write_ops, write_byt, write_tme;
 423
 424         int  disk_major, disk_minor;
 425         char disk_name[DATA_MAX_NAME_LEN];
 426         char child_disk_name_bsd[DATA_MAX_NAME_LEN], props_disk_name_bsd[DATA_MAX_NAME_LEN];
 427
 428         /* Get the list of all disk objects. */
 429         if (IOServiceGetMatchingServices (io_master_port, IOServiceMatching (kIOBlockStorageDriverClass), &disk_list) != kIOReturnSuccess) {
 430                 ERROR ("disk plugin: IOServiceGetMatchingServices failed.");
 431                 return (-1);
 432         }
 433
 434         while ((disk = IOIteratorNext (disk_list)) != 0) {
 435                 props_dict = NULL;
 436                 stats_dict = NULL;
 437                 child_dict = NULL;
 438
 439                 /* get child of disk entry and corresponding property dictionary */
 440                 if ((status = IORegistryEntryGetChildEntry (disk, kIOServicePlane, &disk_child)) != kIOReturnSuccess) {
 441                         /* This fails for example for DVD/CD drives, which we want to ignore anyway */
 442                         DEBUG ("IORegistryEntryGetChildEntry (disk) failed: 0x%08x", status);
 443                         IOObjectRelease (disk);
 444                         continue;
 445                 }
 446                 if (IORegistryEntryCreateCFProperties (disk_child, (CFMutableDictionaryRef *) &child_dict, kCFAllocatorDefault, kNilOptions) != kIOReturnSuccess || child_dict == NULL) {
 447                         ERROR ("disk plugin: IORegistryEntryCreateCFProperties (disk_child) failed.");
 448                         IOObjectRelease (disk_child);
 449                         IOObjectRelease (disk);
 450                         continue;
 451                 }
 452
 453                 /* extract name and major/minor numbers */
 454                 memset (child_disk_name_bsd, 0, sizeof (child_disk_name_bsd));
 455                 tmp_cf_string_ref = (CFStringRef) CFDictionaryGetValue (child_dict, CFSTR(kIOBSDNameKey));
 456                 if (tmp_cf_string_ref) {
 457                         assert (CFGetTypeID (tmp_cf_string_ref) == CFStringGetTypeID ());
 458                         CFStringGetCString (tmp_cf_string_ref, child_disk_name_bsd, sizeof (child_disk_name_bsd), kCFStringEncodingUTF8);
 459                 }
 460                 disk_major = (int) dict_get_value (child_dict, kIOBSDMajorKey);
 461                 disk_minor = (int) dict_get_value (child_dict, kIOBSDMinorKey);
 462                 DEBUG ("disk plugin: child_disk_name_bsd=\"%s\" major=%d minor=%d", child_disk_name_bsd, disk_major, disk_minor);
 463                 CFRelease (child_dict);
 464                 IOObjectRelease (disk_child);
 465
 466                 /* get property dictionary of the disk entry itself */
 467                 if (IORegistryEntryCreateCFProperties (disk, (CFMutableDictionaryRef *) &props_dict, kCFAllocatorDefault, kNilOptions) != kIOReturnSuccess || props_dict == NULL) {
 468                         ERROR ("disk-plugin: IORegistryEntryCreateCFProperties failed.");
 469                         IOObjectRelease (disk);
 470                         continue;
 471                 }
 472
 473                 /* extract name and stats dictionary */
 474                 memset (props_disk_name_bsd, 0, sizeof (props_disk_name_bsd));
 475                 tmp_cf_string_ref = (CFStringRef) CFDictionaryGetValue (props_dict, CFSTR(kIOBSDNameKey));
 476                 if (tmp_cf_string_ref) {
 477                         assert (CFGetTypeID (tmp_cf_string_ref) == CFStringGetTypeID ());
 478                         CFStringGetCString (tmp_cf_string_ref, props_disk_name_bsd, sizeof (props_disk_name_bsd), kCFStringEncodingUTF8);
 479                 }
 480                 stats_dict = (CFDictionaryRef) CFDictionaryGetValue (props_dict, CFSTR (kIOBlockStorageDriverStatisticsKey));
 481                 if (stats_dict == NULL) {
 482                         ERROR ("disk plugin: CFDictionaryGetValue (%s) failed.", kIOBlockStorageDriverStatisticsKey);
 483                         CFRelease (props_dict);
 484                         IOObjectRelease (disk);
 485                         continue;
 486                 }
 487                 DEBUG ("disk plugin: props_disk_name_bsd=\"%s\"", props_disk_name_bsd);
 488
 489                 /* choose name */
 490                 if (use_bsd_name) {
 491                         if (child_disk_name_bsd[0] != 0)
 492                                 sstrncpy (disk_name, child_disk_name_bsd, sizeof (disk_name));
 493                         else if (props_disk_name_bsd[0] != 0)
 494                                 sstrncpy (disk_name, props_disk_name_bsd, sizeof (disk_name));
 495                         else {
 496                                 ERROR ("disk plugin: can't find bsd disk name.");
 497                                 ssnprintf (disk_name, sizeof (disk_name), "%i-%i", disk_major, disk_minor);
 498                         }
 499                 }
 500                 else
 501                         ssnprintf (disk_name, sizeof (disk_name), "%i-%i", disk_major, disk_minor);
 502
 503                 DEBUG ("disk plugin: disk_name = \"%s\"", disk_name);
 504
 505                 /* check the name against ignore list */
 506                 if (ignorelist_match (ignorelist, disk_name) != 0) {
 507                         CFRelease (props_dict);
 508                         IOObjectRelease (disk);
 509                         continue;
 510                 }
 511
 512                 /* extract the stats */
 513                 read_ops  = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsReadsKey);
 514                 read_byt  = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsBytesReadKey);
 515                 read_tme  = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsTotalReadTimeKey);
 516                 write_ops = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsWritesKey);
 517                 write_byt = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsBytesWrittenKey);
 518                 write_tme = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsTotalWriteTimeKey);
 519                 CFRelease (props_dict);
 520                 IOObjectRelease (disk);
 521
 522                 /* and submit */
 523                 if ((read_byt != -1LL) || (write_byt != -1LL))
 524                         disk_submit (disk_name, "disk_octets", read_byt, write_byt);
 525                 if ((read_ops != -1LL) || (write_ops != -1LL))
 526                         disk_submit (disk_name, "disk_ops", read_ops, write_ops);
 527                 if ((read_tme != -1LL) || (write_tme != -1LL))
 528                         disk_submit (disk_name, "disk_time", read_tme / 1000, write_tme / 1000);
 529
 530         }
 531         IOObjectRelease (disk_list);
 532 /* #endif HAVE_IOKIT_IOKITLIB_H */
 533
 534 #elif KERNEL_FREEBSD
 535         int retry, dirty;
 536
 537         void *snap = NULL;
 538         struct devstat *snap_iter;
 539
 540         struct gident *geom_id;
 541
 542         const char *disk_name;
 543         long double read_time, write_time, busy_time, total_duration;
 544
 545         for (retry = 0, dirty = 1; retry < 5 && dirty == 1; retry++) {
 546                 if (snap != NULL)
 547                         geom_stats_snapshot_free(snap);
 548
 549                 /* Get a fresh copy of stats snapshot */
 550                 snap = geom_stats_snapshot_get();
 551                 if (snap == NULL) {
 552                         ERROR("disk plugin: geom_stats_snapshot_get() failed.");
 553                         return (-1);
 554                 }
 555
 556                 /* Check if we have dirty read from this snapshot */
 557                 dirty = 0;
 558                 geom_stats_snapshot_reset(snap);
 559                 while ((snap_iter = geom_stats_snapshot_next(snap)) != NULL) {
 560                         if (snap_iter->id == NULL)
 561                                 continue;
 562                         geom_id = geom_lookupid(&geom_tree, snap_iter->id);
 563
 564                         /* New device? refresh GEOM tree */
 565                         if (geom_id == NULL) {
 566                                 geom_deletetree(&geom_tree);
 567                                 if (geom_gettree(&geom_tree) != 0) {
 568                                         ERROR("disk plugin: geom_gettree() failed");
 569                                         geom_stats_snapshot_free(snap);
 570                                         return (-1);
 571                                 }
 572                                 geom_id = geom_lookupid(&geom_tree, snap_iter->id);
 573                         }
 574                         /*
 575                          * This should be rare: the device come right before we take the
 576                          * snapshot and went away right after it.  We will handle this
 577                          * case later, so don't mark dirty but silently ignore it.
 578                          */
 579                         if (geom_id == NULL)
 580                                 continue;
 581
 582                         /* Only collect PROVIDER data */
 583                         if (geom_id->lg_what != ISPROVIDER)
 584                                 continue;
 585
 586                         /* Only collect data when rank is 1 (physical devices) */
 587                         if (((struct gprovider *)(geom_id->lg_ptr))->lg_geom->lg_rank != 1)
 588                                 continue;
 589
 590                         /* Check if this is a dirty read quit for another try */
 591                         if (snap_iter->sequence0 != snap_iter->sequence1) {
 592                                 dirty = 1;
 593                                 break;
 594                         }
 595                 }
 596         }
 597
 598         /* Reset iterator */
 599         geom_stats_snapshot_reset(snap);
 600         for (;;) {
 601                 snap_iter = geom_stats_snapshot_next(snap);
 602                 if (snap_iter == NULL)
 603                         break;
 604
 605                 if (snap_iter->id == NULL)
 606                         continue;
 607                 geom_id = geom_lookupid(&geom_tree, snap_iter->id);
 608                 if (geom_id == NULL)
 609                         continue;
 610                 if (geom_id->lg_what != ISPROVIDER)
 611                         continue;
 612                 if (((struct gprovider *)(geom_id->lg_ptr))->lg_geom->lg_rank != 1)
 613                         continue;
 614                 /* Skip dirty reads, if present */
 615                 if (dirty && (snap_iter->sequence0 != snap_iter->sequence1))
 616                         continue;
 617
 618                 disk_name = ((struct gprovider *)geom_id->lg_ptr)->lg_name;
 619
 620                 if (ignorelist_match (ignorelist, disk_name) != 0)
 621                         continue;
 622
 623                 if ((snap_iter->bytes[DEVSTAT_READ] != 0) || (snap_iter->bytes[DEVSTAT_WRITE] != 0)) {
 624                         disk_submit(disk_name, "disk_octets",
 625                                         (derive_t)snap_iter->bytes[DEVSTAT_READ],
 626                                         (derive_t)snap_iter->bytes[DEVSTAT_WRITE]);
 627                 }
 628
 629                 if ((snap_iter->operations[DEVSTAT_READ] != 0) || (snap_iter->operations[DEVSTAT_WRITE] != 0)) {
 630                         disk_submit(disk_name, "disk_ops",
 631                                         (derive_t)snap_iter->operations[DEVSTAT_READ],
 632                                         (derive_t)snap_iter->operations[DEVSTAT_WRITE]);
 633                 }
 634
 635                 read_time = devstat_compute_etime(&snap_iter->duration[DEVSTAT_READ], NULL);
 636                 write_time = devstat_compute_etime(&snap_iter->duration[DEVSTAT_WRITE], NULL);
 637                 if ((read_time != 0) || (write_time != 0)) {
 638                         disk_submit (disk_name, "disk_time",
 639                                         (derive_t)(read_time*1000), (derive_t)(write_time*1000));
 640                 }
 641                 if (devstat_compute_statistics(snap_iter, NULL, 1.0,
 642                     DSM_TOTAL_BUSY_TIME, &busy_time,
 643                     DSM_TOTAL_DURATION, &total_duration,
 644                     DSM_NONE) != 0) {
 645                         WARNING("%s", devstat_errbuf);
 646                 }
 647                 else
 648                 {
 649                         submit_io_time(disk_name, busy_time, total_duration);
 650                 }
 651         }
 652         geom_stats_snapshot_free(snap);
 653
 654 #elif KERNEL_LINUX
 655         FILE *fh;
 656         char buffer[1024];
 657
 658         char *fields[32];
 659         int numfields;
 660         int fieldshift = 0;
 661
 662         int minor = 0;
 663
 664         derive_t read_sectors  = 0;
 665         derive_t write_sectors = 0;
 666
 667         derive_t read_ops      = 0;
 668         derive_t read_merged   = 0;
 669         derive_t read_time     = 0;
 670         derive_t write_ops     = 0;
 671         derive_t write_merged  = 0;
 672         derive_t write_time    = 0;
 673         gauge_t in_progress    = NAN;
 674         derive_t io_time       = 0;
 675         derive_t weighted_time = 0;
 676         int is_disk = 0;
 677
 678         diskstats_t *ds, *pre_ds;
 679
 680         if ((fh = fopen ("/proc/diskstats", "r")) == NULL)
 681         {
 682                 fh = fopen ("/proc/partitions", "r");
 683                 if (fh == NULL)
 684                 {
 685                         ERROR ("disk plugin: fopen (/proc/{diskstats,partitions}) failed.");
 686                         return (-1);
 687                 }
 688
 689                 /* Kernel is 2.4.* */
 690                 fieldshift = 1;
 691         }
 692
 693 #if HAVE_LIBUDEV
 694         handle_udev = udev_new();
 695 #endif
 696
 697         while (fgets (buffer, sizeof (buffer), fh) != NULL)
 698         {
 699                 char *disk_name;
 700                 char *output_name;
 701
 702                 numfields = strsplit (buffer, fields, 32);
 703
 704                 if ((numfields != (14 + fieldshift)) && (numfields != 7))
 705                         continue;
 706
 707                 minor = atoll (fields[1]);
 708
 709                 disk_name = fields[2 + fieldshift];
 710
 711                 for (ds = disklist, pre_ds = disklist; ds != NULL; pre_ds = ds, ds = ds->next)
 712                         if (strcmp (disk_name, ds->name) == 0)
 713                                 break;
 714
 715                 if (ds == NULL)
 716                 {
 717                         if ((ds = (diskstats_t *) calloc (1, sizeof (diskstats_t))) == NULL)
 718                                 continue;
 719
 720                         if ((ds->name = strdup (disk_name)) == NULL)
 721                         {
 722                                 free (ds);
 723                                 continue;
 724                         }
 725
 726                         if (pre_ds == NULL)
 727                                 disklist = ds;
 728                         else
 729                                 pre_ds->next = ds;
 730                 }
 731
 732                 is_disk = 0;
 733                 if (numfields == 7)
 734                 {
 735                         /* Kernel 2.6, Partition */
 736                         read_ops      = atoll (fields[3]);
 737                         read_sectors  = atoll (fields[4]);
 738                         write_ops     = atoll (fields[5]);
 739                         write_sectors = atoll (fields[6]);
 740                 }
 741                 else if (numfields == (14 + fieldshift))
 742                 {
 743                         read_ops  =  atoll (fields[3 + fieldshift]);
 744                         write_ops =  atoll (fields[7 + fieldshift]);
 745
 746                         read_sectors  = atoll (fields[5 + fieldshift]);
 747                         write_sectors = atoll (fields[9 + fieldshift]);
 748
 749                         if ((fieldshift == 0) || (minor == 0))
 750                         {
 751                                 is_disk = 1;
 752                                 read_merged  = atoll (fields[4 + fieldshift]);
 753                                 read_time    = atoll (fields[6 + fieldshift]);
 754                                 write_merged = atoll (fields[8 + fieldshift]);
 755                                 write_time   = atoll (fields[10+ fieldshift]);
 756
 757                                 in_progress = atof (fields[11 + fieldshift]);
 758
 759                                 io_time       = atof (fields[12 + fieldshift]);
 760                                 weighted_time = atof (fields[13 + fieldshift]);
 761                         }
 762                 }
 763                 else
 764                 {
 765                         DEBUG ("numfields = %i; => unknown file format.", numfields);
 766                         continue;
 767                 }
 768
 769                 {
 770                         derive_t diff_read_sectors;
 771                         derive_t diff_write_sectors;
 772
 773                 /* If the counter wraps around, it's only 32 bits.. */
 774                         if (read_sectors < ds->read_sectors)
 775                                 diff_read_sectors = 1 + read_sectors
 776                                         + (UINT_MAX - ds->read_sectors);
 777                         else
 778                                 diff_read_sectors = read_sectors - ds->read_sectors;
 779                         if (write_sectors < ds->write_sectors)
 780                                 diff_write_sectors = 1 + write_sectors
 781                                         + (UINT_MAX - ds->write_sectors);
 782                         else
 783                                 diff_write_sectors = write_sectors - ds->write_sectors;
 784
 785                         ds->read_bytes += 512 * diff_read_sectors;
 786                         ds->write_bytes += 512 * diff_write_sectors;
 787                         ds->read_sectors = read_sectors;
 788                         ds->write_sectors = write_sectors;
 789                 }
 790
 791                 /* Calculate the average time an io-op needs to complete */
 792                 if (is_disk)
 793                 {
 794                         derive_t diff_read_ops;
 795                         derive_t diff_write_ops;
 796                         derive_t diff_read_time;
 797                         derive_t diff_write_time;
 798
 799                         if (read_ops < ds->read_ops)
 800                                 diff_read_ops = 1 + read_ops
 801                                         + (UINT_MAX - ds->read_ops);
 802                         else
 803                                 diff_read_ops = read_ops - ds->read_ops;
 804                         DEBUG ("disk plugin: disk_name = %s; read_ops = %"PRIi64"; "
 805                                         "ds->read_ops = %"PRIi64"; diff_read_ops = %"PRIi64";",
 806                                         disk_name,
 807                                         read_ops, ds->read_ops, diff_read_ops);
 808
 809                         if (write_ops < ds->write_ops)
 810                                 diff_write_ops = 1 + write_ops
 811                                         + (UINT_MAX - ds->write_ops);
 812                         else
 813                                 diff_write_ops = write_ops - ds->write_ops;
 814
 815                         if (read_time < ds->read_time)
 816                                 diff_read_time = 1 + read_time
 817                                         + (UINT_MAX - ds->read_time);
 818                         else
 819                                 diff_read_time = read_time - ds->read_time;
 820
 821                         if (write_time < ds->write_time)
 822                                 diff_write_time = 1 + write_time
 823                                         + (UINT_MAX - ds->write_time);
 824                         else
 825                                 diff_write_time = write_time - ds->write_time;
 826
 827                         if (diff_read_ops != 0)
 828                                 ds->avg_read_time += disk_calc_time_incr (
 829                                                 diff_read_time, diff_read_ops);
 830                         if (diff_write_ops != 0)
 831                                 ds->avg_write_time += disk_calc_time_incr (
 832                                                 diff_write_time, diff_write_ops);
 833
 834                         ds->read_ops = read_ops;
 835                         ds->read_time = read_time;
 836                         ds->write_ops = write_ops;
 837                         ds->write_time = write_time;
 838
 839                         if (read_merged || write_merged)
 840                                 ds->has_merged = 1;
 841
 842                         if (in_progress)
 843                                 ds->has_in_progress = 1;
 844
 845                         if (io_time)
 846                                 ds->has_io_time = 1;
 847
 848                 } /* if (is_disk) */
 849
 850                 /* Don't write to the RRDs if we've just started.. */
 851                 ds->poll_count++;
 852                 if (ds->poll_count <= 2)
 853                 {
 854                         DEBUG ("disk plugin: (ds->poll_count = %i) <= "
 855                                         "(min_poll_count = 2); => Not writing.",
 856                                         ds->poll_count);
 857                         continue;
 858                 }
 859
 860                 if ((read_ops == 0) && (write_ops == 0))
 861                 {
 862                         DEBUG ("disk plugin: ((read_ops == 0) && "
 863                                         "(write_ops == 0)); => Not writing.");
 864                         continue;
 865                 }
 866
 867                 output_name = disk_name;
 868
 869 #if HAVE_LIBUDEV
 870                 char *alt_name = disk_udev_attr_name (handle_udev, disk_name, conf_udev_name_attr);
 871                 if (alt_name != NULL)
 872                         output_name = alt_name;
 873 #endif
 874
 875                 if (ignorelist_match (ignorelist, output_name) != 0)
 876                         continue;
 877
 878                 if ((ds->read_bytes != 0) || (ds->write_bytes != 0))
 879                         disk_submit (output_name, "disk_octets",
 880                                         ds->read_bytes, ds->write_bytes);
 881
 882                 if ((ds->read_ops != 0) || (ds->write_ops != 0))
 883                         disk_submit (output_name, "disk_ops",
 884                                         read_ops, write_ops);
 885
 886                 if ((ds->avg_read_time != 0) || (ds->avg_write_time != 0))
 887                         disk_submit (output_name, "disk_time",
 888                                         ds->avg_read_time, ds->avg_write_time);
 889
 890                 if (is_disk)
 891                 {
 892                         if (ds->has_merged)
 893                                 disk_submit (output_name, "disk_merged",
 894                                         read_merged, write_merged);
 895                         if (ds->has_in_progress)
 896                                 submit_in_progress (output_name, in_progress);
 897                         if (ds->has_io_time)
 898                                 submit_io_time (output_name, io_time, weighted_time);
 899                 } /* if (is_disk) */
 900
 901 #if HAVE_LIBUDEV
 902                 /* release udev-based alternate name, if allocated */
 903                 sfree (alt_name);
 904 #endif
 905         } /* while (fgets (buffer, sizeof (buffer), fh) != NULL) */
 906
 907 #if HAVE_LIBUDEV
 908         udev_unref(handle_udev);
 909 #endif
 910
 911         fclose (fh);
 912 /* #endif defined(KERNEL_LINUX) */
 913
 914 #elif HAVE_LIBKSTAT
 915 # if HAVE_KSTAT_IO_T_WRITES && HAVE_KSTAT_IO_T_NWRITES && HAVE_KSTAT_IO_T_WTIME
 916 #  define KIO_ROCTETS reads
 917 #  define KIO_WOCTETS writes
 918 #  define KIO_ROPS    nreads
 919 #  define KIO_WOPS    nwrites
 920 #  define KIO_RTIME   rtime
 921 #  define KIO_WTIME   wtime
 922 # elif HAVE_KSTAT_IO_T_NWRITTEN && HAVE_KSTAT_IO_T_WRITES && HAVE_KSTAT_IO_T_WTIME
 923 #  define KIO_ROCTETS nread
 924 #  define KIO_WOCTETS nwritten
 925 #  define KIO_ROPS    reads
 926 #  define KIO_WOPS    writes
 927 #  define KIO_RTIME   rtime
 928 #  define KIO_WTIME   wtime
 929 # else
 930 #  error "kstat_io_t does not have the required members"
 931 # endif
 932         static kstat_io_t kio;
 933         int i;
 934
 935         if (kc == NULL)
 936                 return (-1);
 937
 938         for (i = 0; i < numdisk; i++)
 939         {
 940                 if (kstat_read (kc, ksp[i], &kio) == -1)
 941                         continue;
 942
 943                 if (strncmp (ksp[i]->ks_class, "disk", 4) == 0)
 944                 {
 945                         if (ignorelist_match (ignorelist, ksp[i]->ks_name) != 0)
 946                                 continue;
 947
 948                         disk_submit (ksp[i]->ks_name, "disk_octets",
 949                                         kio.KIO_ROCTETS, kio.KIO_WOCTETS);
 950                         disk_submit (ksp[i]->ks_name, "disk_ops",
 951                                         kio.KIO_ROPS, kio.KIO_WOPS);
 952                         /* FIXME: Convert this to microseconds if necessary */
 953                         disk_submit (ksp[i]->ks_name, "disk_time",
 954                                         kio.KIO_RTIME, kio.KIO_WTIME);
 955                 }
 956                 else if (strncmp (ksp[i]->ks_class, "partition", 9) == 0)
 957                 {
 958                         if (ignorelist_match (ignorelist, ksp[i]->ks_name) != 0)
 959                                 continue;
 960
 961                         disk_submit (ksp[i]->ks_name, "disk_octets",
 962                                         kio.KIO_ROCTETS, kio.KIO_WOCTETS);
 963                         disk_submit (ksp[i]->ks_name, "disk_ops",
 964                                         kio.KIO_ROPS, kio.KIO_WOPS);
 965                 }
 966         }
 967 /* #endif defined(HAVE_LIBKSTAT) */
 968
 969 #elif defined(HAVE_LIBSTATGRAB)
 970         sg_disk_io_stats *ds;
 971 # if HAVE_LIBSTATGRAB_0_90
 972         size_t disks;
 973 # else
 974         int disks;
 975 #endif
 976         int counter;
 977         char name[DATA_MAX_NAME_LEN];
 978
 979         if ((ds = sg_get_disk_io_stats(&disks)) == NULL)
 980                 return (0);
 981
 982         for (counter=0; counter < disks; counter++) {
 983                 strncpy(name, ds->disk_name, sizeof(name));
 984                 name[sizeof(name)-1] = '\0'; /* strncpy doesn't terminate longer strings */
 985
 986                 if (ignorelist_match (ignorelist, name) != 0) {
 987                         ds++;
 988                         continue;
 989                 }
 990
 991                 disk_submit (name, "disk_octets", ds->read_bytes, ds->write_bytes);
 992                 ds++;
 993         }
 994 /* #endif defined(HAVE_LIBSTATGRAB) */
 995
 996 #elif defined(HAVE_PERFSTAT)
 997         derive_t read_sectors;
 998         derive_t write_sectors;
 999         derive_t read_time;
1000         derive_t write_time;
1001         derive_t read_ops;
1002         derive_t write_ops;
1003         perfstat_id_t firstpath;
1004         int rnumdisk;
1005         int i;
1006
1007         if ((numdisk = perfstat_disk(NULL, NULL, sizeof(perfstat_disk_t), 0)) < 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         if (numdisk != pnumdisk || stat_disk==NULL) {
1016                 if (stat_disk!=NULL)
1017                         free(stat_disk);
1018                 stat_disk = (perfstat_disk_t *)calloc(numdisk, sizeof(perfstat_disk_t));
1019         }
1020         pnumdisk = numdisk;
1021
1022         firstpath.name[0]='\0';
1023         if ((rnumdisk = perfstat_disk(&firstpath, stat_disk, sizeof(perfstat_disk_t), numdisk)) < 0)
1024         {
1025                 char errbuf[1024];
1026                 WARNING ("disk plugin: perfstat_disk : %s",
1027                                 sstrerror (errno, errbuf, sizeof (errbuf)));
1028                 return (-1);
1029         }
1030
1031         for (i = 0; i < rnumdisk; i++)
1032         {
1033                 if (ignorelist_match (ignorelist, stat_disk[i].name) != 0)
1034                         continue;
1035
1036                 read_sectors = stat_disk[i].rblks*stat_disk[i].bsize;
1037                 write_sectors = stat_disk[i].wblks*stat_disk[i].bsize;
1038                 disk_submit (stat_disk[i].name, "disk_octets", read_sectors, write_sectors);
1039
1040                 read_ops = stat_disk[i].xrate;
1041                 write_ops = stat_disk[i].xfers - stat_disk[i].xrate;
1042                 disk_submit (stat_disk[i].name, "disk_ops", read_ops, write_ops);
1043
1044                 read_time = stat_disk[i].rserv;
1045                 read_time *= ((double)(_system_configuration.Xint)/(double)(_system_configuration.Xfrac)) / 1000000.0;
1046                 write_time = stat_disk[i].wserv;
1047                 write_time *= ((double)(_system_configuration.Xint)/(double)(_system_configuration.Xfrac)) / 1000000.0;
1048                 disk_submit (stat_disk[i].name, "disk_time", read_time, write_time);
1049         }
1050 #endif /* defined(HAVE_PERFSTAT) */
1051
1052         return (0);
1053 } /* int disk_read */
1054
1055 void module_register (void)
1056 {
1057   plugin_register_config ("disk", disk_config,
1058       config_keys, config_keys_num);
1059   plugin_register_init ("disk", disk_init);
1060   plugin_register_read ("disk", disk_read);
1061 } /* void module_register */