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         struct diskstats *next;
 110 } diskstats_t;
 111
 112 static diskstats_t *disklist;
 113 /* #endif KERNEL_LINUX */
 114 #elif KERNEL_FREEBSD
 115 static struct gmesh geom_tree;
 116 /* #endif KERNEL_FREEBSD */
 117
 118 #elif HAVE_LIBKSTAT
 119 #define MAX_NUMDISK 1024
 120 extern kstat_ctl_t *kc;
 121 static kstat_t *ksp[MAX_NUMDISK];
 122 static int numdisk = 0;
 123 /* #endif HAVE_LIBKSTAT */
 124
 125 #elif defined(HAVE_LIBSTATGRAB)
 126 /* #endif HAVE_LIBKSTATGRAB */
 127
 128 #elif HAVE_PERFSTAT
 129 static perfstat_disk_t * stat_disk;
 130 static int numdisk;
 131 static int pnumdisk;
 132 /* #endif HAVE_PERFSTAT */
 133
 134 #else
 135 # error "No applicable input method."
 136 #endif
 137
 138 #if HAVE_LIBUDEV
 139 #include <libudev.h>
 140
 141 static char *conf_udev_name_attr = NULL;
 142 static struct udev *handle_udev;
 143 #endif
 144
 145 static const char *config_keys[] =
 146 {
 147         "Disk",
 148         "UseBSDName",
 149         "IgnoreSelected",
 150         "UdevNameAttr"
 151 };
 152 static int config_keys_num = STATIC_ARRAY_SIZE (config_keys);
 153
 154 static ignorelist_t *ignorelist = NULL;
 155
 156 static int disk_config (const char *key, const char *value)
 157 {
 158   if (ignorelist == NULL)
 159     ignorelist = ignorelist_create (/* invert = */ 1);
 160   if (ignorelist == NULL)
 161     return (1);
 162
 163   if (strcasecmp ("Disk", key) == 0)
 164   {
 165     ignorelist_add (ignorelist, value);
 166   }
 167   else if (strcasecmp ("IgnoreSelected", key) == 0)
 168   {
 169     int invert = 1;
 170     if (IS_TRUE (value))
 171       invert = 0;
 172     ignorelist_set_invert (ignorelist, invert);
 173   }
 174   else if (strcasecmp ("UseBSDName", key) == 0)
 175   {
 176 #if HAVE_IOKIT_IOKITLIB_H
 177     use_bsd_name = IS_TRUE (value) ? 1 : 0;
 178 #else
 179     WARNING ("disk plugin: The \"UseBSDName\" option is only supported "
 180         "on Mach / Mac OS X and will be ignored.");
 181 #endif
 182   }
 183   else if (strcasecmp ("UdevNameAttr", key) == 0)
 184   {
 185 #if HAVE_LIBUDEV
 186     if (conf_udev_name_attr != NULL)
 187     {
 188       free (conf_udev_name_attr);
 189       conf_udev_name_attr = NULL;
 190     }
 191     if ((conf_udev_name_attr = strdup (value)) == NULL)
 192       return (1);
 193 #else
 194     WARNING ("disk plugin: The \"UdevNameAttr\" option is only supported "
 195         "if collectd is built with libudev support");
 196 #endif
 197   }
 198   else
 199   {
 200     return (-1);
 201   }
 202
 203   return (0);
 204 } /* int disk_config */
 205
 206 static int disk_init (void)
 207 {
 208 #if HAVE_IOKIT_IOKITLIB_H
 209         kern_return_t status;
 210
 211         if (io_master_port != MACH_PORT_NULL)
 212         {
 213                 mach_port_deallocate (mach_task_self (),
 214                                 io_master_port);
 215                 io_master_port = MACH_PORT_NULL;
 216         }
 217
 218         status = IOMasterPort (MACH_PORT_NULL, &io_master_port);
 219         if (status != kIOReturnSuccess)
 220         {
 221                 ERROR ("IOMasterPort failed: %s",
 222                                 mach_error_string (status));
 223                 io_master_port = MACH_PORT_NULL;
 224                 return (-1);
 225         }
 226 /* #endif HAVE_IOKIT_IOKITLIB_H */
 227
 228 #elif KERNEL_LINUX
 229         /* do nothing */
 230 /* #endif KERNEL_LINUX */
 231
 232 #elif KERNEL_FREEBSD
 233         int rv;
 234
 235         rv = geom_gettree(&geom_tree);
 236         if (rv != 0) {
 237                 ERROR ("geom_gettree() failed, returned %d", rv);
 238                 return (-1);
 239         }
 240         rv = geom_stats_open();
 241         if (rv != 0) {
 242                 ERROR ("geom_stats_open() failed, returned %d", rv);
 243                 return (-1);
 244         }
 245 /* #endif KERNEL_FREEBSD */
 246
 247 #elif HAVE_LIBKSTAT
 248         kstat_t *ksp_chain;
 249
 250         numdisk = 0;
 251
 252         if (kc == NULL)
 253                 return (-1);
 254
 255         for (numdisk = 0, ksp_chain = kc->kc_chain;
 256                         (numdisk < MAX_NUMDISK) && (ksp_chain != NULL);
 257                         ksp_chain = ksp_chain->ks_next)
 258         {
 259                 if (strncmp (ksp_chain->ks_class, "disk", 4)
 260                                 && strncmp (ksp_chain->ks_class, "partition", 9))
 261                         continue;
 262                 if (ksp_chain->ks_type != KSTAT_TYPE_IO)
 263                         continue;
 264                 ksp[numdisk++] = ksp_chain;
 265         }
 266 #endif /* HAVE_LIBKSTAT */
 267
 268         return (0);
 269 } /* int disk_init */
 270
 271 static void disk_submit (const char *plugin_instance,
 272                 const char *type,
 273                 derive_t read, derive_t write)
 274 {
 275         value_t values[2];
 276         value_list_t vl = VALUE_LIST_INIT;
 277
 278         /* Both `ignorelist' and `plugin_instance' may be NULL. */
 279         if (ignorelist_match (ignorelist, plugin_instance) != 0)
 280           return;
 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_LINUX
 297 static void submit_in_progress (char const *disk_name, gauge_t in_progress)
 298 {
 299         value_t v;
 300         value_list_t vl = VALUE_LIST_INIT;
 301
 302         if (ignorelist_match (ignorelist, disk_name) != 0)
 303           return;
 304
 305         v.gauge = in_progress;
 306
 307         vl.values = &v;
 308         vl.values_len = 1;
 309         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
 310         sstrncpy (vl.plugin, "disk", sizeof (vl.plugin));
 311         sstrncpy (vl.plugin_instance, disk_name, sizeof (vl.plugin_instance));
 312         sstrncpy (vl.type, "pending_operations", sizeof (vl.type));
 313
 314         plugin_dispatch_values (&vl);
 315 }
 316
 317 static void submit_io_time (char const *plugin_instance, derive_t io_time, derive_t weighted_time)
 318 {
 319         value_t values[2];
 320         value_list_t vl = VALUE_LIST_INIT;
 321
 322         if (ignorelist_match (ignorelist, plugin_instance) != 0)
 323           return;
 324
 325         values[0].derive = io_time;
 326         values[1].derive = weighted_time;
 327
 328         vl.values = values;
 329         vl.values_len = 2;
 330         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
 331         sstrncpy (vl.plugin, "disk", sizeof (vl.plugin));
 332         sstrncpy (vl.plugin_instance, plugin_instance, sizeof (vl.plugin_instance));
 333         sstrncpy (vl.type, "disk_io_time", sizeof (vl.type));
 334
 335         plugin_dispatch_values (&vl);
 336 }
 337
 338
 339 static counter_t disk_calc_time_incr (counter_t delta_time, counter_t delta_ops)
 340 {
 341         double interval = CDTIME_T_TO_DOUBLE (plugin_get_interval ());
 342         double avg_time = ((double) delta_time) / ((double) delta_ops);
 343         double avg_time_incr = interval * avg_time;
 344
 345         return ((counter_t) (avg_time_incr + .5));
 346 }
 347 #endif
 348
 349 #if HAVE_LIBUDEV
 350 /**
 351  * Attempt to provide an rename disk instance from an assigned udev attribute.
 352  *
 353  * On success, it returns a strduped char* to the desired attribute value.
 354  * Otherwise it returns NULL.
 355  */
 356
 357 static char *disk_udev_attr_name (struct udev *udev, char *disk_name, const char *attr)
 358 {
 359         struct udev_device *dev;
 360         const char *prop;
 361         char *output = NULL;
 362
 363         dev = udev_device_new_from_subsystem_sysname (udev, "block", disk_name);
 364         if (dev != NULL)
 365         {
 366                 prop = udev_device_get_property_value (dev, attr);
 367                 if (prop) {
 368                         output = strdup (prop);
 369                         DEBUG ("disk plugin: renaming %s => %s", disk_name, output);
 370                 }
 371                 udev_device_unref (dev);
 372         }
 373         return output;
 374 }
 375 #endif
 376
 377 #if HAVE_IOKIT_IOKITLIB_H
 378 static signed long long dict_get_value (CFDictionaryRef dict, const char *key)
 379 {
 380         signed long long val_int;
 381         CFNumberRef      val_obj;
 382         CFStringRef      key_obj;
 383
 384         /* `key_obj' needs to be released. */
 385         key_obj = CFStringCreateWithCString (kCFAllocatorDefault, key,
 386                         kCFStringEncodingASCII);
 387         if (key_obj == NULL)
 388         {
 389                 DEBUG ("CFStringCreateWithCString (%s) failed.", key);
 390                 return (-1LL);
 391         }
 392
 393         /* get => we don't need to release (== free) the object */
 394         val_obj = (CFNumberRef) CFDictionaryGetValue (dict, key_obj);
 395
 396         CFRelease (key_obj);
 397
 398         if (val_obj == NULL)
 399         {
 400                 DEBUG ("CFDictionaryGetValue (%s) failed.", key);
 401                 return (-1LL);
 402         }
 403
 404         if (!CFNumberGetValue (val_obj, kCFNumberSInt64Type, &val_int))
 405         {
 406                 DEBUG ("CFNumberGetValue (%s) failed.", key);
 407                 return (-1LL);
 408         }
 409
 410         return (val_int);
 411 }
 412 #endif /* HAVE_IOKIT_IOKITLIB_H */
 413
 414 static int disk_read (void)
 415 {
 416 #if HAVE_IOKIT_IOKITLIB_H
 417         io_registry_entry_t     disk;
 418         io_registry_entry_t     disk_child;
 419         io_iterator_t           disk_list;
 420         CFMutableDictionaryRef  props_dict, child_dict;
 421         CFDictionaryRef         stats_dict;
 422         CFStringRef             tmp_cf_string_ref;
 423         kern_return_t           status;
 424
 425         signed long long read_ops, read_byt, read_tme;
 426         signed long long write_ops, write_byt, write_tme;
 427
 428         int  disk_major, disk_minor;
 429         char disk_name[DATA_MAX_NAME_LEN];
 430         char child_disk_name_bsd[DATA_MAX_NAME_LEN], props_disk_name_bsd[DATA_MAX_NAME_LEN];
 431
 432         /* Get the list of all disk objects. */
 433         if (IOServiceGetMatchingServices (io_master_port, IOServiceMatching (kIOBlockStorageDriverClass), &disk_list) != kIOReturnSuccess) {
 434                 ERROR ("disk plugin: IOServiceGetMatchingServices failed.");
 435                 return (-1);
 436         }
 437
 438         while ((disk = IOIteratorNext (disk_list)) != 0) {
 439                 props_dict = NULL;
 440                 stats_dict = NULL;
 441                 child_dict = NULL;
 442
 443                 /* get child of disk entry and corresponding property dictionary */
 444                 if ((status = IORegistryEntryGetChildEntry (disk, kIOServicePlane, &disk_child)) != kIOReturnSuccess) {
 445                         /* This fails for example for DVD/CD drives, which we want to ignore anyway */
 446                         DEBUG ("IORegistryEntryGetChildEntry (disk) failed: 0x%08x", status);
 447                         IOObjectRelease (disk);
 448                         continue;
 449                 }
 450                 if (IORegistryEntryCreateCFProperties (disk_child, (CFMutableDictionaryRef *) &child_dict, kCFAllocatorDefault, kNilOptions) != kIOReturnSuccess || child_dict == NULL) {
 451                         ERROR ("disk plugin: IORegistryEntryCreateCFProperties (disk_child) failed.");
 452                         IOObjectRelease (disk_child);
 453                         IOObjectRelease (disk);
 454                         continue;
 455                 }
 456
 457                 /* extract name and major/minor numbers */
 458                 memset (child_disk_name_bsd, 0, sizeof (child_disk_name_bsd));
 459                 tmp_cf_string_ref = (CFStringRef) CFDictionaryGetValue (child_dict, CFSTR(kIOBSDNameKey));
 460                 if (tmp_cf_string_ref) {
 461                         assert (CFGetTypeID (tmp_cf_string_ref) == CFStringGetTypeID ());
 462                         CFStringGetCString (tmp_cf_string_ref, child_disk_name_bsd, sizeof (child_disk_name_bsd), kCFStringEncodingUTF8);
 463                 }
 464                 disk_major = (int) dict_get_value (child_dict, kIOBSDMajorKey);
 465                 disk_minor = (int) dict_get_value (child_dict, kIOBSDMinorKey);
 466                 DEBUG ("disk plugin: child_disk_name_bsd=\"%s\" major=%d minor=%d", child_disk_name_bsd, disk_major, disk_minor);
 467                 CFRelease (child_dict);
 468                 IOObjectRelease (disk_child);
 469
 470                 /* get property dictionary of the disk entry itself */
 471                 if (IORegistryEntryCreateCFProperties (disk, (CFMutableDictionaryRef *) &props_dict, kCFAllocatorDefault, kNilOptions) != kIOReturnSuccess || props_dict == NULL) {
 472                         ERROR ("disk-plugin: IORegistryEntryCreateCFProperties failed.");
 473                         IOObjectRelease (disk);
 474                         continue;
 475                 }
 476
 477                 /* extract name and stats dictionary */
 478                 memset (props_disk_name_bsd, 0, sizeof (props_disk_name_bsd));
 479                 tmp_cf_string_ref = (CFStringRef) CFDictionaryGetValue (props_dict, CFSTR(kIOBSDNameKey));
 480                 if (tmp_cf_string_ref) {
 481                         assert (CFGetTypeID (tmp_cf_string_ref) == CFStringGetTypeID ());
 482                         CFStringGetCString (tmp_cf_string_ref, props_disk_name_bsd, sizeof (props_disk_name_bsd), kCFStringEncodingUTF8);
 483                 }
 484                 stats_dict = (CFDictionaryRef) CFDictionaryGetValue (props_dict, CFSTR (kIOBlockStorageDriverStatisticsKey));
 485                 if (stats_dict == NULL) {
 486                         ERROR ("disk plugin: CFDictionaryGetValue (%s) failed.", kIOBlockStorageDriverStatisticsKey);
 487                         CFRelease (props_dict);
 488                         IOObjectRelease (disk);
 489                         continue;
 490                 }
 491                 DEBUG ("disk plugin: props_disk_name_bsd=\"%s\"", props_disk_name_bsd);
 492
 493                 /* choose name */
 494                 if (use_bsd_name) {
 495                         if (child_disk_name_bsd[0] != 0)
 496                                 sstrncpy (disk_name, child_disk_name_bsd, sizeof (disk_name));
 497                         else if (props_disk_name_bsd[0] != 0)
 498                                 sstrncpy (disk_name, props_disk_name_bsd, sizeof (disk_name));
 499                         else {
 500                                 ERROR ("disk plugin: can't find bsd disk name.");
 501                                 ssnprintf (disk_name, sizeof (disk_name), "%i-%i", disk_major, disk_minor);
 502                         }
 503                 }
 504                 else
 505                         ssnprintf (disk_name, sizeof (disk_name), "%i-%i", disk_major, disk_minor);
 506
 507                 /* extract the stats */
 508                 read_ops  = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsReadsKey);
 509                 read_byt  = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsBytesReadKey);
 510                 read_tme  = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsTotalReadTimeKey);
 511                 write_ops = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsWritesKey);
 512                 write_byt = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsBytesWrittenKey);
 513                 write_tme = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsTotalWriteTimeKey);
 514                 CFRelease (props_dict);
 515                 IOObjectRelease (disk);
 516
 517                 /* and submit */
 518                 DEBUG ("disk plugin: disk_name = \"%s\"", disk_name);
 519                 if ((read_byt != -1LL) || (write_byt != -1LL))
 520                         disk_submit (disk_name, "disk_octets", read_byt, write_byt);
 521                 if ((read_ops != -1LL) || (write_ops != -1LL))
 522                         disk_submit (disk_name, "disk_ops", read_ops, write_ops);
 523                 if ((read_tme != -1LL) || (write_tme != -1LL))
 524                         disk_submit (disk_name, "disk_time", read_tme / 1000, write_tme / 1000);
 525
 526         }
 527         IOObjectRelease (disk_list);
 528 /* #endif HAVE_IOKIT_IOKITLIB_H */
 529
 530 #elif KERNEL_FREEBSD
 531         int retry, dirty;
 532
 533         void *snap_present = NULL;
 534         struct devstat *snap_present_iter;
 535
 536         struct gident *geom_id;
 537
 538         const char *disk_name;
 539         uint64_t read_bytes, write_bytes;
 540         uint64_t read_ops, write_ops;
 541         long double read_time, write_time;
 542
 543         for (retry = 0, dirty = 1; retry < 5 && dirty == 1; retry++) {
 544                 if (snap_present != NULL)
 545                         geom_stats_snapshot_free(snap_present);
 546
 547                 /* Get a fresh copy of stats snapshot */
 548                 snap_present = geom_stats_snapshot_get();
 549                 if (snap_present == NULL) {
 550                         ERROR("disk plugin: geom_stats_snapshot_get() failed.");
 551                         return (-1);
 552                 }
 553
 554                 /* Check if we have dirty read from this snapshot */
 555                 dirty = 0;
 556                 geom_stats_snapshot_reset(snap_present);
 557                 while ((snap_present_iter = geom_stats_snapshot_next(snap_present)) != NULL) {
 558                         if (snap_present_iter->id == NULL)
 559                                 continue;
 560                         geom_id = geom_lookupid(&geom_tree, snap_present_iter->id);
 561
 562                         /* New device? refresh GEOM tree */
 563                         if (geom_id == NULL) {
 564                                 geom_deletetree(&geom_tree);
 565                                 if (geom_gettree(&geom_tree) != 0) {
 566                                         ERROR("disk plugin: geom_gettree() failed");
 567                                         geom_stats_snapshot_free(snap_present);
 568                                         return (-1);
 569                                 }
 570                                 geom_id = geom_lookupid(&geom_tree, snap_present_iter->id);
 571                         }
 572                         /*
 573                          * This should be rare: the device come right before we take the
 574                          * snapshot and went away right after it.  We will handle this
 575                          * case later, so don't mark dirty but silently ignore it.
 576                          */
 577                         if (geom_id == NULL)
 578                                 continue;
 579
 580                         /* Only collect PROVIDER data */
 581                         if (geom_id->lg_what != ISPROVIDER)
 582                                 continue;
 583
 584                         /* Only collect data when rank is 1 (physical devices) */
 585                         if (((struct gprovider *)(geom_id->lg_ptr))->lg_geom->lg_rank != 1)
 586                                 continue;
 587
 588                         /* Check if this is a dirty read quit for another try */
 589                         if (snap_present_iter->sequence0 != snap_present_iter->sequence1) {
 590                                 dirty = 1;
 591                                 break;
 592                         }
 593                 }
 594         }
 595
 596         /* Reset iterators for both snapshots */
 597         geom_stats_snapshot_reset(snap_present);
 598         for (;;) {
 599                 snap_present_iter = geom_stats_snapshot_next(snap_present);
 600                 if (snap_present_iter == NULL)
 601                         break;
 602
 603                 if (snap_present_iter->id == NULL)
 604                         continue;
 605                 geom_id = geom_lookupid(&geom_tree, snap_present_iter->id);
 606                 if (geom_id == NULL)
 607                         continue;
 608                 if (geom_id->lg_what != ISPROVIDER)
 609                         continue;
 610                 if (((struct gprovider *)(geom_id->lg_ptr))->lg_geom->lg_rank != 1)
 611                         continue;
 612                 /* Skip dirty reads, if present */
 613                 if (dirty && (snap_present_iter->sequence0 != snap_present_iter->sequence1))
 614                         continue;
 615
 616                 devstat_compute_statistics(snap_present_iter, NULL, 10.0,
 617                     DSM_TOTAL_TRANSFERS_READ, &read_ops,
 618                     DSM_TOTAL_BYTES_READ, &read_bytes,
 619                     DSM_TOTAL_DURATION_READ, &read_time,
 620
 621                     DSM_TOTAL_TRANSFERS_WRITE, &write_ops,
 622                     DSM_TOTAL_BYTES_WRITE, &write_bytes,
 623                     DSM_TOTAL_DURATION_WRITE, &write_time,
 624
 625                     DSM_NONE);
 626
 627                 disk_name = ((struct gprovider *)geom_id->lg_ptr)->lg_name;
 628
 629                 if ((read_bytes != 0) || (write_bytes != 0))
 630                         disk_submit(disk_name, "disk_octets",
 631                                         (derive_t)read_bytes, (derive_t)write_bytes);
 632
 633                 if ((read_ops != 0) || (write_ops != 0))
 634                         disk_submit(disk_name, "disk_ops",
 635                                         (derive_t)read_ops, (derive_t)write_ops);
 636
 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         }
 642         geom_stats_snapshot_free(snap_present);
 643
 644 #elif KERNEL_LINUX
 645         FILE *fh;
 646         char buffer[1024];
 647
 648         char *fields[32];
 649         int numfields;
 650         int fieldshift = 0;
 651
 652         int minor = 0;
 653
 654         derive_t read_sectors  = 0;
 655         derive_t write_sectors = 0;
 656
 657         derive_t read_ops      = 0;
 658         derive_t read_merged   = 0;
 659         derive_t read_time     = 0;
 660         derive_t write_ops     = 0;
 661         derive_t write_merged  = 0;
 662         derive_t write_time    = 0;
 663         gauge_t in_progress    = NAN;
 664         derive_t io_time       = 0;
 665         derive_t weighted_time = 0;
 666         int is_disk = 0;
 667
 668         diskstats_t *ds, *pre_ds;
 669
 670         if ((fh = fopen ("/proc/diskstats", "r")) == NULL)
 671         {
 672                 fh = fopen ("/proc/partitions", "r");
 673                 if (fh == NULL)
 674                 {
 675                         ERROR ("disk plugin: fopen (/proc/{diskstats,partitions}) failed.");
 676                         return (-1);
 677                 }
 678
 679                 /* Kernel is 2.4.* */
 680                 fieldshift = 1;
 681         }
 682
 683 #if HAVE_LIBUDEV
 684         handle_udev = udev_new();
 685 #endif
 686
 687         while (fgets (buffer, sizeof (buffer), fh) != NULL)
 688         {
 689                 char *disk_name;
 690                 char *output_name;
 691                 char *alt_name;
 692
 693                 numfields = strsplit (buffer, fields, 32);
 694
 695                 if ((numfields != (14 + fieldshift)) && (numfields != 7))
 696                         continue;
 697
 698                 minor = atoll (fields[1]);
 699
 700                 disk_name = fields[2 + fieldshift];
 701
 702                 for (ds = disklist, pre_ds = disklist; ds != NULL; pre_ds = ds, ds = ds->next)
 703                         if (strcmp (disk_name, ds->name) == 0)
 704                                 break;
 705
 706                 if (ds == NULL)
 707                 {
 708                         if ((ds = (diskstats_t *) calloc (1, sizeof (diskstats_t))) == NULL)
 709                                 continue;
 710
 711                         if ((ds->name = strdup (disk_name)) == NULL)
 712                         {
 713                                 free (ds);
 714                                 continue;
 715                         }
 716
 717                         if (pre_ds == NULL)
 718                                 disklist = ds;
 719                         else
 720                                 pre_ds->next = ds;
 721                 }
 722
 723                 is_disk = 0;
 724                 if (numfields == 7)
 725                 {
 726                         /* Kernel 2.6, Partition */
 727                         read_ops      = atoll (fields[3]);
 728                         read_sectors  = atoll (fields[4]);
 729                         write_ops     = atoll (fields[5]);
 730                         write_sectors = atoll (fields[6]);
 731                 }
 732                 else if (numfields == (14 + fieldshift))
 733                 {
 734                         read_ops  =  atoll (fields[3 + fieldshift]);
 735                         write_ops =  atoll (fields[7 + fieldshift]);
 736
 737                         read_sectors  = atoll (fields[5 + fieldshift]);
 738                         write_sectors = atoll (fields[9 + fieldshift]);
 739
 740                         if ((fieldshift == 0) || (minor == 0))
 741                         {
 742                                 is_disk = 1;
 743                                 read_merged  = atoll (fields[4 + fieldshift]);
 744                                 read_time    = atoll (fields[6 + fieldshift]);
 745                                 write_merged = atoll (fields[8 + fieldshift]);
 746                                 write_time   = atoll (fields[10+ fieldshift]);
 747
 748                                 in_progress = atof (fields[11 + fieldshift]);
 749
 750                                 io_time       = atof (fields[12 + fieldshift]);
 751                                 weighted_time = atof (fields[13 + fieldshift]);
 752                         }
 753                 }
 754                 else
 755                 {
 756                         DEBUG ("numfields = %i; => unknown file format.", numfields);
 757                         continue;
 758                 }
 759
 760                 {
 761                         derive_t diff_read_sectors;
 762                         derive_t diff_write_sectors;
 763
 764                 /* If the counter wraps around, it's only 32 bits.. */
 765                         if (read_sectors < ds->read_sectors)
 766                                 diff_read_sectors = 1 + read_sectors
 767                                         + (UINT_MAX - ds->read_sectors);
 768                         else
 769                                 diff_read_sectors = read_sectors - ds->read_sectors;
 770                         if (write_sectors < ds->write_sectors)
 771                                 diff_write_sectors = 1 + write_sectors
 772                                         + (UINT_MAX - ds->write_sectors);
 773                         else
 774                                 diff_write_sectors = write_sectors - ds->write_sectors;
 775
 776                         ds->read_bytes += 512 * diff_read_sectors;
 777                         ds->write_bytes += 512 * diff_write_sectors;
 778                         ds->read_sectors = read_sectors;
 779                         ds->write_sectors = write_sectors;
 780                 }
 781
 782                 /* Calculate the average time an io-op needs to complete */
 783                 if (is_disk)
 784                 {
 785                         derive_t diff_read_ops;
 786                         derive_t diff_write_ops;
 787                         derive_t diff_read_time;
 788                         derive_t diff_write_time;
 789
 790                         if (read_ops < ds->read_ops)
 791                                 diff_read_ops = 1 + read_ops
 792                                         + (UINT_MAX - ds->read_ops);
 793                         else
 794                                 diff_read_ops = read_ops - ds->read_ops;
 795                         DEBUG ("disk plugin: disk_name = %s; read_ops = %"PRIi64"; "
 796                                         "ds->read_ops = %"PRIi64"; diff_read_ops = %"PRIi64";",
 797                                         disk_name,
 798                                         read_ops, ds->read_ops, diff_read_ops);
 799
 800                         if (write_ops < ds->write_ops)
 801                                 diff_write_ops = 1 + write_ops
 802                                         + (UINT_MAX - ds->write_ops);
 803                         else
 804                                 diff_write_ops = write_ops - ds->write_ops;
 805
 806                         if (read_time < ds->read_time)
 807                                 diff_read_time = 1 + read_time
 808                                         + (UINT_MAX - ds->read_time);
 809                         else
 810                                 diff_read_time = read_time - ds->read_time;
 811
 812                         if (write_time < ds->write_time)
 813                                 diff_write_time = 1 + write_time
 814                                         + (UINT_MAX - ds->write_time);
 815                         else
 816                                 diff_write_time = write_time - ds->write_time;
 817
 818                         if (diff_read_ops != 0)
 819                                 ds->avg_read_time += disk_calc_time_incr (
 820                                                 diff_read_time, diff_read_ops);
 821                         if (diff_write_ops != 0)
 822                                 ds->avg_write_time += disk_calc_time_incr (
 823                                                 diff_write_time, diff_write_ops);
 824
 825                         ds->read_ops = read_ops;
 826                         ds->read_time = read_time;
 827                         ds->write_ops = write_ops;
 828                         ds->write_time = write_time;
 829                 } /* if (is_disk) */
 830
 831                 /* Don't write to the RRDs if we've just started.. */
 832                 ds->poll_count++;
 833                 if (ds->poll_count <= 2)
 834                 {
 835                         DEBUG ("disk plugin: (ds->poll_count = %i) <= "
 836                                         "(min_poll_count = 2); => Not writing.",
 837                                         ds->poll_count);
 838                         continue;
 839                 }
 840
 841                 if ((read_ops == 0) && (write_ops == 0))
 842                 {
 843                         DEBUG ("disk plugin: ((read_ops == 0) && "
 844                                         "(write_ops == 0)); => Not writing.");
 845                         continue;
 846                 }
 847
 848                 output_name = disk_name;
 849
 850 #if HAVE_LIBUDEV
 851                 alt_name = disk_udev_attr_name (handle_udev, disk_name,
 852                                 conf_udev_name_attr);
 853 #else
 854                 alt_name = NULL;
 855 #endif
 856                 if (alt_name != NULL)
 857                         output_name = alt_name;
 858
 859                 if ((ds->read_bytes != 0) || (ds->write_bytes != 0))
 860                         disk_submit (output_name, "disk_octets",
 861                                         ds->read_bytes, ds->write_bytes);
 862
 863                 if ((ds->read_ops != 0) || (ds->write_ops != 0))
 864                         disk_submit (output_name, "disk_ops",
 865                                         read_ops, write_ops);
 866
 867                 if ((ds->avg_read_time != 0) || (ds->avg_write_time != 0))
 868                         disk_submit (output_name, "disk_time",
 869                                         ds->avg_read_time, ds->avg_write_time);
 870
 871                 if (is_disk)
 872                 {
 873                         disk_submit (output_name, "disk_merged",
 874                                         read_merged, write_merged);
 875                         submit_in_progress (output_name, in_progress);
 876                         submit_io_time (output_name, io_time, weighted_time);
 877                 } /* if (is_disk) */
 878
 879                 /* release udev-based alternate name, if allocated */
 880                 free(alt_name);
 881         } /* while (fgets (buffer, sizeof (buffer), fh) != NULL) */
 882
 883 #if HAVE_LIBUDEV
 884         udev_unref(handle_udev);
 885 #endif
 886
 887         fclose (fh);
 888 /* #endif defined(KERNEL_LINUX) */
 889
 890 #elif HAVE_LIBKSTAT
 891 # if HAVE_KSTAT_IO_T_WRITES && HAVE_KSTAT_IO_T_NWRITES && HAVE_KSTAT_IO_T_WTIME
 892 #  define KIO_ROCTETS reads
 893 #  define KIO_WOCTETS writes
 894 #  define KIO_ROPS    nreads
 895 #  define KIO_WOPS    nwrites
 896 #  define KIO_RTIME   rtime
 897 #  define KIO_WTIME   wtime
 898 # elif HAVE_KSTAT_IO_T_NWRITTEN && HAVE_KSTAT_IO_T_WRITES && HAVE_KSTAT_IO_T_WTIME
 899 #  define KIO_ROCTETS nread
 900 #  define KIO_WOCTETS nwritten
 901 #  define KIO_ROPS    reads
 902 #  define KIO_WOPS    writes
 903 #  define KIO_RTIME   rtime
 904 #  define KIO_WTIME   wtime
 905 # else
 906 #  error "kstat_io_t does not have the required members"
 907 # endif
 908         static kstat_io_t kio;
 909         int i;
 910
 911         if (kc == NULL)
 912                 return (-1);
 913
 914         for (i = 0; i < numdisk; i++)
 915         {
 916                 if (kstat_read (kc, ksp[i], &kio) == -1)
 917                         continue;
 918
 919                 if (strncmp (ksp[i]->ks_class, "disk", 4) == 0)
 920                 {
 921                         disk_submit (ksp[i]->ks_name, "disk_octets",
 922                                         kio.KIO_ROCTETS, kio.KIO_WOCTETS);
 923                         disk_submit (ksp[i]->ks_name, "disk_ops",
 924                                         kio.KIO_ROPS, kio.KIO_WOPS);
 925                         /* FIXME: Convert this to microseconds if necessary */
 926                         disk_submit (ksp[i]->ks_name, "disk_time",
 927                                         kio.KIO_RTIME, kio.KIO_WTIME);
 928                 }
 929                 else if (strncmp (ksp[i]->ks_class, "partition", 9) == 0)
 930                 {
 931                         disk_submit (ksp[i]->ks_name, "disk_octets",
 932                                         kio.KIO_ROCTETS, kio.KIO_WOCTETS);
 933                         disk_submit (ksp[i]->ks_name, "disk_ops",
 934                                         kio.KIO_ROPS, kio.KIO_WOPS);
 935                 }
 936         }
 937 /* #endif defined(HAVE_LIBKSTAT) */
 938
 939 #elif defined(HAVE_LIBSTATGRAB)
 940         sg_disk_io_stats *ds;
 941 # if HAVE_LIBSTATGRAB_0_90
 942         size_t disks;
 943 # else
 944         int disks;
 945 #endif
 946         int counter;
 947         char name[DATA_MAX_NAME_LEN];
 948
 949         if ((ds = sg_get_disk_io_stats(&disks)) == NULL)
 950                 return (0);
 951
 952         for (counter=0; counter < disks; counter++) {
 953                 strncpy(name, ds->disk_name, sizeof(name));
 954                 name[sizeof(name)-1] = '\0'; /* strncpy doesn't terminate longer strings */
 955                 disk_submit (name, "disk_octets", ds->read_bytes, ds->write_bytes);
 956                 ds++;
 957         }
 958 /* #endif defined(HAVE_LIBSTATGRAB) */
 959
 960 #elif defined(HAVE_PERFSTAT)
 961         derive_t read_sectors;
 962         derive_t write_sectors;
 963         derive_t read_time;
 964         derive_t write_time;
 965         derive_t read_ops;
 966         derive_t write_ops;
 967         perfstat_id_t firstpath;
 968         int rnumdisk;
 969         int i;
 970
 971         if ((numdisk = perfstat_disk(NULL, NULL, sizeof(perfstat_disk_t), 0)) < 0)
 972         {
 973                 char errbuf[1024];
 974                 WARNING ("disk plugin: perfstat_disk: %s",
 975                                 sstrerror (errno, errbuf, sizeof (errbuf)));
 976                 return (-1);
 977         }
 978
 979         if (numdisk != pnumdisk || stat_disk==NULL) {
 980                 if (stat_disk!=NULL)
 981                         free(stat_disk);
 982                 stat_disk = (perfstat_disk_t *)calloc(numdisk, sizeof(perfstat_disk_t));
 983         }
 984         pnumdisk = numdisk;
 985
 986         firstpath.name[0]='\0';
 987         if ((rnumdisk = perfstat_disk(&firstpath, stat_disk, sizeof(perfstat_disk_t), numdisk)) < 0)
 988         {
 989                 char errbuf[1024];
 990                 WARNING ("disk plugin: perfstat_disk : %s",
 991                                 sstrerror (errno, errbuf, sizeof (errbuf)));
 992                 return (-1);
 993         }
 994
 995         for (i = 0; i < rnumdisk; i++)
 996         {
 997                 read_sectors = stat_disk[i].rblks*stat_disk[i].bsize;
 998                 write_sectors = stat_disk[i].wblks*stat_disk[i].bsize;
 999                 disk_submit (stat_disk[i].name, "disk_octets", read_sectors, write_sectors);
1000
1001                 read_ops = stat_disk[i].xrate;
1002                 write_ops = stat_disk[i].xfers - stat_disk[i].xrate;
1003                 disk_submit (stat_disk[i].name, "disk_ops", read_ops, write_ops);
1004
1005                 read_time = stat_disk[i].rserv;
1006                 read_time *= ((double)(_system_configuration.Xint)/(double)(_system_configuration.Xfrac)) / 1000000.0;
1007                 write_time = stat_disk[i].wserv;
1008                 write_time *= ((double)(_system_configuration.Xint)/(double)(_system_configuration.Xfrac)) / 1000000.0;
1009                 disk_submit (stat_disk[i].name, "disk_time", read_time, write_time);
1010         }
1011 #endif /* defined(HAVE_PERFSTAT) */
1012
1013         return (0);
1014 } /* int disk_read */
1015
1016 void module_register (void)
1017 {
1018   plugin_register_config ("disk", disk_config,
1019       config_keys, config_keys_num);
1020   plugin_register_init ("disk", disk_init);
1021   plugin_register_read ("disk", disk_read);
1022 } /* void module_register */