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_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         if (ignorelist_match (ignorelist, plugin_instance) != 0)
 303           return;
 304
 305         values[0].derive = io_time;
 306         values[1].derive = weighted_time;
 307
 308         vl.values = values;
 309         vl.values_len = 2;
 310         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
 311         sstrncpy (vl.plugin, "disk", sizeof (vl.plugin));
 312         sstrncpy (vl.plugin_instance, plugin_instance, sizeof (vl.plugin_instance));
 313         sstrncpy (vl.type, "disk_io_time", sizeof (vl.type));
 314
 315         plugin_dispatch_values (&vl);
 316 } /* void submit_io_time */
 317 #endif /* KERNEL_FREEBSD || KERNEL_LINUX */
 318
 319 #if KERNEL_LINUX
 320 static void submit_in_progress (char const *disk_name, gauge_t in_progress)
 321 {
 322         value_t v;
 323         value_list_t vl = VALUE_LIST_INIT;
 324
 325         if (ignorelist_match (ignorelist, disk_name) != 0)
 326           return;
 327
 328         v.gauge = in_progress;
 329
 330         vl.values = &v;
 331         vl.values_len = 1;
 332         sstrncpy (vl.host, hostname_g, sizeof (vl.host));
 333         sstrncpy (vl.plugin, "disk", sizeof (vl.plugin));
 334         sstrncpy (vl.plugin_instance, disk_name, sizeof (vl.plugin_instance));
 335         sstrncpy (vl.type, "pending_operations", sizeof (vl.type));
 336
 337         plugin_dispatch_values (&vl);
 338 }
 339
 340
 341 static counter_t disk_calc_time_incr (counter_t delta_time, counter_t delta_ops)
 342 {
 343         double interval = CDTIME_T_TO_DOUBLE (plugin_get_interval ());
 344         double avg_time = ((double) delta_time) / ((double) delta_ops);
 345         double avg_time_incr = interval * avg_time;
 346
 347         return ((counter_t) (avg_time_incr + .5));
 348 }
 349 #endif
 350
 351 #if HAVE_LIBUDEV
 352 /**
 353  * Attempt to provide an rename disk instance from an assigned udev attribute.
 354  *
 355  * On success, it returns a strduped char* to the desired attribute value.
 356  * Otherwise it returns NULL.
 357  */
 358
 359 static char *disk_udev_attr_name (struct udev *udev, char *disk_name, const char *attr)
 360 {
 361         struct udev_device *dev;
 362         const char *prop;
 363         char *output = NULL;
 364
 365         dev = udev_device_new_from_subsystem_sysname (udev, "block", disk_name);
 366         if (dev != NULL)
 367         {
 368                 prop = udev_device_get_property_value (dev, attr);
 369                 if (prop) {
 370                         output = strdup (prop);
 371                         DEBUG ("disk plugin: renaming %s => %s", disk_name, output);
 372                 }
 373                 udev_device_unref (dev);
 374         }
 375         return output;
 376 }
 377 #endif
 378
 379 #if HAVE_IOKIT_IOKITLIB_H
 380 static signed long long dict_get_value (CFDictionaryRef dict, const char *key)
 381 {
 382         signed long long val_int;
 383         CFNumberRef      val_obj;
 384         CFStringRef      key_obj;
 385
 386         /* `key_obj' needs to be released. */
 387         key_obj = CFStringCreateWithCString (kCFAllocatorDefault, key,
 388                         kCFStringEncodingASCII);
 389         if (key_obj == NULL)
 390         {
 391                 DEBUG ("CFStringCreateWithCString (%s) failed.", key);
 392                 return (-1LL);
 393         }
 394
 395         /* get => we don't need to release (== free) the object */
 396         val_obj = (CFNumberRef) CFDictionaryGetValue (dict, key_obj);
 397
 398         CFRelease (key_obj);
 399
 400         if (val_obj == NULL)
 401         {
 402                 DEBUG ("CFDictionaryGetValue (%s) failed.", key);
 403                 return (-1LL);
 404         }
 405
 406         if (!CFNumberGetValue (val_obj, kCFNumberSInt64Type, &val_int))
 407         {
 408                 DEBUG ("CFNumberGetValue (%s) failed.", key);
 409                 return (-1LL);
 410         }
 411
 412         return (val_int);
 413 }
 414 #endif /* HAVE_IOKIT_IOKITLIB_H */
 415
 416 static int disk_read (void)
 417 {
 418 #if HAVE_IOKIT_IOKITLIB_H
 419         io_registry_entry_t     disk;
 420         io_registry_entry_t     disk_child;
 421         io_iterator_t           disk_list;
 422         CFMutableDictionaryRef  props_dict, child_dict;
 423         CFDictionaryRef         stats_dict;
 424         CFStringRef             tmp_cf_string_ref;
 425         kern_return_t           status;
 426
 427         signed long long read_ops, read_byt, read_tme;
 428         signed long long write_ops, write_byt, write_tme;
 429
 430         int  disk_major, disk_minor;
 431         char disk_name[DATA_MAX_NAME_LEN];
 432         char child_disk_name_bsd[DATA_MAX_NAME_LEN], props_disk_name_bsd[DATA_MAX_NAME_LEN];
 433
 434         /* Get the list of all disk objects. */
 435         if (IOServiceGetMatchingServices (io_master_port, IOServiceMatching (kIOBlockStorageDriverClass), &disk_list) != kIOReturnSuccess) {
 436                 ERROR ("disk plugin: IOServiceGetMatchingServices failed.");
 437                 return (-1);
 438         }
 439
 440         while ((disk = IOIteratorNext (disk_list)) != 0) {
 441                 props_dict = NULL;
 442                 stats_dict = NULL;
 443                 child_dict = NULL;
 444
 445                 /* get child of disk entry and corresponding property dictionary */
 446                 if ((status = IORegistryEntryGetChildEntry (disk, kIOServicePlane, &disk_child)) != kIOReturnSuccess) {
 447                         /* This fails for example for DVD/CD drives, which we want to ignore anyway */
 448                         DEBUG ("IORegistryEntryGetChildEntry (disk) failed: 0x%08x", status);
 449                         IOObjectRelease (disk);
 450                         continue;
 451                 }
 452                 if (IORegistryEntryCreateCFProperties (disk_child, (CFMutableDictionaryRef *) &child_dict, kCFAllocatorDefault, kNilOptions) != kIOReturnSuccess || child_dict == NULL) {
 453                         ERROR ("disk plugin: IORegistryEntryCreateCFProperties (disk_child) failed.");
 454                         IOObjectRelease (disk_child);
 455                         IOObjectRelease (disk);
 456                         continue;
 457                 }
 458
 459                 /* extract name and major/minor numbers */
 460                 memset (child_disk_name_bsd, 0, sizeof (child_disk_name_bsd));
 461                 tmp_cf_string_ref = (CFStringRef) CFDictionaryGetValue (child_dict, CFSTR(kIOBSDNameKey));
 462                 if (tmp_cf_string_ref) {
 463                         assert (CFGetTypeID (tmp_cf_string_ref) == CFStringGetTypeID ());
 464                         CFStringGetCString (tmp_cf_string_ref, child_disk_name_bsd, sizeof (child_disk_name_bsd), kCFStringEncodingUTF8);
 465                 }
 466                 disk_major = (int) dict_get_value (child_dict, kIOBSDMajorKey);
 467                 disk_minor = (int) dict_get_value (child_dict, kIOBSDMinorKey);
 468                 DEBUG ("disk plugin: child_disk_name_bsd=\"%s\" major=%d minor=%d", child_disk_name_bsd, disk_major, disk_minor);
 469                 CFRelease (child_dict);
 470                 IOObjectRelease (disk_child);
 471
 472                 /* get property dictionary of the disk entry itself */
 473                 if (IORegistryEntryCreateCFProperties (disk, (CFMutableDictionaryRef *) &props_dict, kCFAllocatorDefault, kNilOptions) != kIOReturnSuccess || props_dict == NULL) {
 474                         ERROR ("disk-plugin: IORegistryEntryCreateCFProperties failed.");
 475                         IOObjectRelease (disk);
 476                         continue;
 477                 }
 478
 479                 /* extract name and stats dictionary */
 480                 memset (props_disk_name_bsd, 0, sizeof (props_disk_name_bsd));
 481                 tmp_cf_string_ref = (CFStringRef) CFDictionaryGetValue (props_dict, CFSTR(kIOBSDNameKey));
 482                 if (tmp_cf_string_ref) {
 483                         assert (CFGetTypeID (tmp_cf_string_ref) == CFStringGetTypeID ());
 484                         CFStringGetCString (tmp_cf_string_ref, props_disk_name_bsd, sizeof (props_disk_name_bsd), kCFStringEncodingUTF8);
 485                 }
 486                 stats_dict = (CFDictionaryRef) CFDictionaryGetValue (props_dict, CFSTR (kIOBlockStorageDriverStatisticsKey));
 487                 if (stats_dict == NULL) {
 488                         ERROR ("disk plugin: CFDictionaryGetValue (%s) failed.", kIOBlockStorageDriverStatisticsKey);
 489                         CFRelease (props_dict);
 490                         IOObjectRelease (disk);
 491                         continue;
 492                 }
 493                 DEBUG ("disk plugin: props_disk_name_bsd=\"%s\"", props_disk_name_bsd);
 494
 495                 /* choose name */
 496                 if (use_bsd_name) {
 497                         if (child_disk_name_bsd[0] != 0)
 498                                 sstrncpy (disk_name, child_disk_name_bsd, sizeof (disk_name));
 499                         else if (props_disk_name_bsd[0] != 0)
 500                                 sstrncpy (disk_name, props_disk_name_bsd, sizeof (disk_name));
 501                         else {
 502                                 ERROR ("disk plugin: can't find bsd disk name.");
 503                                 ssnprintf (disk_name, sizeof (disk_name), "%i-%i", disk_major, disk_minor);
 504                         }
 505                 }
 506                 else
 507                         ssnprintf (disk_name, sizeof (disk_name), "%i-%i", disk_major, disk_minor);
 508
 509                 /* extract the stats */
 510                 read_ops  = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsReadsKey);
 511                 read_byt  = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsBytesReadKey);
 512                 read_tme  = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsTotalReadTimeKey);
 513                 write_ops = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsWritesKey);
 514                 write_byt = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsBytesWrittenKey);
 515                 write_tme = dict_get_value (stats_dict, kIOBlockStorageDriverStatisticsTotalWriteTimeKey);
 516                 CFRelease (props_dict);
 517                 IOObjectRelease (disk);
 518
 519                 /* and submit */
 520                 DEBUG ("disk plugin: disk_name = \"%s\"", disk_name);
 521                 if ((read_byt != -1LL) || (write_byt != -1LL))
 522                         disk_submit (disk_name, "disk_octets", read_byt, write_byt);
 523                 if ((read_ops != -1LL) || (write_ops != -1LL))
 524                         disk_submit (disk_name, "disk_ops", read_ops, write_ops);
 525                 if ((read_tme != -1LL) || (write_tme != -1LL))
 526                         disk_submit (disk_name, "disk_time", read_tme / 1000, write_tme / 1000);
 527
 528         }
 529         IOObjectRelease (disk_list);
 530 /* #endif HAVE_IOKIT_IOKITLIB_H */
 531
 532 #elif KERNEL_FREEBSD
 533         int retry, dirty;
 534
 535         void *snap = NULL;
 536         struct devstat *snap_iter;
 537
 538         struct gident *geom_id;
 539
 540         const char *disk_name;
 541         long double read_time, write_time, busy_time, total_duration;
 542
 543         for (retry = 0, dirty = 1; retry < 5 && dirty == 1; retry++) {
 544                 if (snap != NULL)
 545                         geom_stats_snapshot_free(snap);
 546
 547                 /* Get a fresh copy of stats snapshot */
 548                 snap = geom_stats_snapshot_get();
 549                 if (snap == 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);
 557                 while ((snap_iter = geom_stats_snapshot_next(snap)) != NULL) {
 558                         if (snap_iter->id == NULL)
 559                                 continue;
 560                         geom_id = geom_lookupid(&geom_tree, snap_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);
 568                                         return (-1);
 569                                 }
 570                                 geom_id = geom_lookupid(&geom_tree, snap_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_iter->sequence0 != snap_iter->sequence1) {
 590                                 dirty = 1;
 591                                 break;
 592                         }
 593                 }
 594         }
 595
 596         /* Reset iterator */
 597         geom_stats_snapshot_reset(snap);
 598         for (;;) {
 599                 snap_iter = geom_stats_snapshot_next(snap);
 600                 if (snap_iter == NULL)
 601                         break;
 602
 603                 if (snap_iter->id == NULL)
 604                         continue;
 605                 geom_id = geom_lookupid(&geom_tree, snap_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_iter->sequence0 != snap_iter->sequence1))
 614                         continue;
 615
 616                 disk_name = ((struct gprovider *)geom_id->lg_ptr)->lg_name;
 617
 618                 if ((snap_iter->bytes[DEVSTAT_READ] != 0) || (snap_iter->bytes[DEVSTAT_WRITE] != 0)) {
 619                         disk_submit(disk_name, "disk_octets",
 620                                         (derive_t)snap_iter->bytes[DEVSTAT_READ],
 621                                         (derive_t)snap_iter->bytes[DEVSTAT_WRITE]);
 622                 }
 623
 624                 if ((snap_iter->operations[DEVSTAT_READ] != 0) || (snap_iter->operations[DEVSTAT_WRITE] != 0)) {
 625                         disk_submit(disk_name, "disk_ops",
 626                                         (derive_t)snap_iter->operations[DEVSTAT_READ],
 627                                         (derive_t)snap_iter->operations[DEVSTAT_WRITE]);
 628                 }
 629
 630                 read_time = devstat_compute_etime(&snap_iter->duration[DEVSTAT_READ], NULL);
 631                 write_time = devstat_compute_etime(&snap_iter->duration[DEVSTAT_WRITE], NULL);
 632                 if ((read_time != 0) || (write_time != 0)) {
 633                         disk_submit (disk_name, "disk_time",
 634                                         (derive_t)(read_time*1000), (derive_t)(write_time*1000));
 635                 }
 636                 if (devstat_compute_statistics(snap_iter, NULL, 1.0,
 637                     DSM_TOTAL_BUSY_TIME, &busy_time,
 638                     DSM_TOTAL_DURATION, &total_duration,
 639                     DSM_NONE) != 0) {
 640                         WARNING("%s", devstat_errbuf);
 641                 }
 642                 else
 643                 {
 644                         submit_io_time(disk_name, busy_time, total_duration);
 645                 }
 646         }
 647         geom_stats_snapshot_free(snap);
 648
 649 #elif KERNEL_LINUX
 650         FILE *fh;
 651         char buffer[1024];
 652
 653         char *fields[32];
 654         int numfields;
 655         int fieldshift = 0;
 656
 657         int minor = 0;
 658
 659         derive_t read_sectors  = 0;
 660         derive_t write_sectors = 0;
 661
 662         derive_t read_ops      = 0;
 663         derive_t read_merged   = 0;
 664         derive_t read_time     = 0;
 665         derive_t write_ops     = 0;
 666         derive_t write_merged  = 0;
 667         derive_t write_time    = 0;
 668         gauge_t in_progress    = NAN;
 669         derive_t io_time       = 0;
 670         derive_t weighted_time = 0;
 671         int is_disk = 0;
 672
 673         diskstats_t *ds, *pre_ds;
 674
 675         if ((fh = fopen ("/proc/diskstats", "r")) == NULL)
 676         {
 677                 fh = fopen ("/proc/partitions", "r");
 678                 if (fh == NULL)
 679                 {
 680                         ERROR ("disk plugin: fopen (/proc/{diskstats,partitions}) failed.");
 681                         return (-1);
 682                 }
 683
 684                 /* Kernel is 2.4.* */
 685                 fieldshift = 1;
 686         }
 687
 688 #if HAVE_LIBUDEV
 689         handle_udev = udev_new();
 690 #endif
 691
 692         while (fgets (buffer, sizeof (buffer), fh) != NULL)
 693         {
 694                 char *disk_name;
 695                 char *output_name;
 696
 697                 numfields = strsplit (buffer, fields, 32);
 698
 699                 if ((numfields != (14 + fieldshift)) && (numfields != 7))
 700                         continue;
 701
 702                 minor = atoll (fields[1]);
 703
 704                 disk_name = fields[2 + fieldshift];
 705
 706                 for (ds = disklist, pre_ds = disklist; ds != NULL; pre_ds = ds, ds = ds->next)
 707                         if (strcmp (disk_name, ds->name) == 0)
 708                                 break;
 709
 710                 if (ds == NULL)
 711                 {
 712                         if ((ds = (diskstats_t *) calloc (1, sizeof (diskstats_t))) == NULL)
 713                                 continue;
 714
 715                         if ((ds->name = strdup (disk_name)) == NULL)
 716                         {
 717                                 free (ds);
 718                                 continue;
 719                         }
 720
 721                         if (pre_ds == NULL)
 722                                 disklist = ds;
 723                         else
 724                                 pre_ds->next = ds;
 725                 }
 726
 727                 is_disk = 0;
 728                 if (numfields == 7)
 729                 {
 730                         /* Kernel 2.6, Partition */
 731                         read_ops      = atoll (fields[3]);
 732                         read_sectors  = atoll (fields[4]);
 733                         write_ops     = atoll (fields[5]);
 734                         write_sectors = atoll (fields[6]);
 735                 }
 736                 else if (numfields == (14 + fieldshift))
 737                 {
 738                         read_ops  =  atoll (fields[3 + fieldshift]);
 739                         write_ops =  atoll (fields[7 + fieldshift]);
 740
 741                         read_sectors  = atoll (fields[5 + fieldshift]);
 742                         write_sectors = atoll (fields[9 + fieldshift]);
 743
 744                         if ((fieldshift == 0) || (minor == 0))
 745                         {
 746                                 is_disk = 1;
 747                                 read_merged  = atoll (fields[4 + fieldshift]);
 748                                 read_time    = atoll (fields[6 + fieldshift]);
 749                                 write_merged = atoll (fields[8 + fieldshift]);
 750                                 write_time   = atoll (fields[10+ fieldshift]);
 751
 752                                 in_progress = atof (fields[11 + fieldshift]);
 753
 754                                 io_time       = atof (fields[12 + fieldshift]);
 755                                 weighted_time = atof (fields[13 + fieldshift]);
 756                         }
 757                 }
 758                 else
 759                 {
 760                         DEBUG ("numfields = %i; => unknown file format.", numfields);
 761                         continue;
 762                 }
 763
 764                 {
 765                         derive_t diff_read_sectors;
 766                         derive_t diff_write_sectors;
 767
 768                 /* If the counter wraps around, it's only 32 bits.. */
 769                         if (read_sectors < ds->read_sectors)
 770                                 diff_read_sectors = 1 + read_sectors
 771                                         + (UINT_MAX - ds->read_sectors);
 772                         else
 773                                 diff_read_sectors = read_sectors - ds->read_sectors;
 774                         if (write_sectors < ds->write_sectors)
 775                                 diff_write_sectors = 1 + write_sectors
 776                                         + (UINT_MAX - ds->write_sectors);
 777                         else
 778                                 diff_write_sectors = write_sectors - ds->write_sectors;
 779
 780                         ds->read_bytes += 512 * diff_read_sectors;
 781                         ds->write_bytes += 512 * diff_write_sectors;
 782                         ds->read_sectors = read_sectors;
 783                         ds->write_sectors = write_sectors;
 784                 }
 785
 786                 /* Calculate the average time an io-op needs to complete */
 787                 if (is_disk)
 788                 {
 789                         derive_t diff_read_ops;
 790                         derive_t diff_write_ops;
 791                         derive_t diff_read_time;
 792                         derive_t diff_write_time;
 793
 794                         if (read_ops < ds->read_ops)
 795                                 diff_read_ops = 1 + read_ops
 796                                         + (UINT_MAX - ds->read_ops);
 797                         else
 798                                 diff_read_ops = read_ops - ds->read_ops;
 799                         DEBUG ("disk plugin: disk_name = %s; read_ops = %"PRIi64"; "
 800                                         "ds->read_ops = %"PRIi64"; diff_read_ops = %"PRIi64";",
 801                                         disk_name,
 802                                         read_ops, ds->read_ops, diff_read_ops);
 803
 804                         if (write_ops < ds->write_ops)
 805                                 diff_write_ops = 1 + write_ops
 806                                         + (UINT_MAX - ds->write_ops);
 807                         else
 808                                 diff_write_ops = write_ops - ds->write_ops;
 809
 810                         if (read_time < ds->read_time)
 811                                 diff_read_time = 1 + read_time
 812                                         + (UINT_MAX - ds->read_time);
 813                         else
 814                                 diff_read_time = read_time - ds->read_time;
 815
 816                         if (write_time < ds->write_time)
 817                                 diff_write_time = 1 + write_time
 818                                         + (UINT_MAX - ds->write_time);
 819                         else
 820                                 diff_write_time = write_time - ds->write_time;
 821
 822                         if (diff_read_ops != 0)
 823                                 ds->avg_read_time += disk_calc_time_incr (
 824                                                 diff_read_time, diff_read_ops);
 825                         if (diff_write_ops != 0)
 826                                 ds->avg_write_time += disk_calc_time_incr (
 827                                                 diff_write_time, diff_write_ops);
 828
 829                         ds->read_ops = read_ops;
 830                         ds->read_time = read_time;
 831                         ds->write_ops = write_ops;
 832                         ds->write_time = write_time;
 833                 } /* if (is_disk) */
 834
 835                 /* Don't write to the RRDs if we've just started.. */
 836                 ds->poll_count++;
 837                 if (ds->poll_count <= 2)
 838                 {
 839                         DEBUG ("disk plugin: (ds->poll_count = %i) <= "
 840                                         "(min_poll_count = 2); => Not writing.",
 841                                         ds->poll_count);
 842                         continue;
 843                 }
 844
 845                 if ((read_ops == 0) && (write_ops == 0))
 846                 {
 847                         DEBUG ("disk plugin: ((read_ops == 0) && "
 848                                         "(write_ops == 0)); => Not writing.");
 849                         continue;
 850                 }
 851
 852                 output_name = disk_name;
 853
 854 #if HAVE_LIBUDEV
 855                 char *alt_name = disk_udev_attr_name (handle_udev, disk_name, conf_udev_name_attr);
 856                 if (alt_name != NULL)
 857                         output_name = alt_name;
 858 #endif
 859
 860                 if ((ds->read_bytes != 0) || (ds->write_bytes != 0))
 861                         disk_submit (output_name, "disk_octets",
 862                                         ds->read_bytes, ds->write_bytes);
 863
 864                 if ((ds->read_ops != 0) || (ds->write_ops != 0))
 865                         disk_submit (output_name, "disk_ops",
 866                                         read_ops, write_ops);
 867
 868                 if ((ds->avg_read_time != 0) || (ds->avg_write_time != 0))
 869                         disk_submit (output_name, "disk_time",
 870                                         ds->avg_read_time, ds->avg_write_time);
 871
 872                 if (is_disk)
 873                 {
 874                         disk_submit (output_name, "disk_merged",
 875                                         read_merged, write_merged);
 876                         submit_in_progress (output_name, in_progress);
 877                         submit_io_time (output_name, io_time, weighted_time);
 878                 } /* if (is_disk) */
 879
 880 #if HAVE_LIBUDEV
 881                 /* release udev-based alternate name, if allocated */
 882                 sfree (alt_name);
 883 #endif
 884         } /* while (fgets (buffer, sizeof (buffer), fh) != NULL) */
 885
 886 #if HAVE_LIBUDEV
 887         udev_unref(handle_udev);
 888 #endif
 889
 890         fclose (fh);
 891 /* #endif defined(KERNEL_LINUX) */
 892
 893 #elif HAVE_LIBKSTAT
 894 # if HAVE_KSTAT_IO_T_WRITES && HAVE_KSTAT_IO_T_NWRITES && HAVE_KSTAT_IO_T_WTIME
 895 #  define KIO_ROCTETS reads
 896 #  define KIO_WOCTETS writes
 897 #  define KIO_ROPS    nreads
 898 #  define KIO_WOPS    nwrites
 899 #  define KIO_RTIME   rtime
 900 #  define KIO_WTIME   wtime
 901 # elif HAVE_KSTAT_IO_T_NWRITTEN && HAVE_KSTAT_IO_T_WRITES && HAVE_KSTAT_IO_T_WTIME
 902 #  define KIO_ROCTETS nread
 903 #  define KIO_WOCTETS nwritten
 904 #  define KIO_ROPS    reads
 905 #  define KIO_WOPS    writes
 906 #  define KIO_RTIME   rtime
 907 #  define KIO_WTIME   wtime
 908 # else
 909 #  error "kstat_io_t does not have the required members"
 910 # endif
 911         static kstat_io_t kio;
 912         int i;
 913
 914         if (kc == NULL)
 915                 return (-1);
 916
 917         for (i = 0; i < numdisk; i++)
 918         {
 919                 if (kstat_read (kc, ksp[i], &kio) == -1)
 920                         continue;
 921
 922                 if (strncmp (ksp[i]->ks_class, "disk", 4) == 0)
 923                 {
 924                         disk_submit (ksp[i]->ks_name, "disk_octets",
 925                                         kio.KIO_ROCTETS, kio.KIO_WOCTETS);
 926                         disk_submit (ksp[i]->ks_name, "disk_ops",
 927                                         kio.KIO_ROPS, kio.KIO_WOPS);
 928                         /* FIXME: Convert this to microseconds if necessary */
 929                         disk_submit (ksp[i]->ks_name, "disk_time",
 930                                         kio.KIO_RTIME, kio.KIO_WTIME);
 931                 }
 932                 else if (strncmp (ksp[i]->ks_class, "partition", 9) == 0)
 933                 {
 934                         disk_submit (ksp[i]->ks_name, "disk_octets",
 935                                         kio.KIO_ROCTETS, kio.KIO_WOCTETS);
 936                         disk_submit (ksp[i]->ks_name, "disk_ops",
 937                                         kio.KIO_ROPS, kio.KIO_WOPS);
 938                 }
 939         }
 940 /* #endif defined(HAVE_LIBKSTAT) */
 941
 942 #elif defined(HAVE_LIBSTATGRAB)
 943         sg_disk_io_stats *ds;
 944 # if HAVE_LIBSTATGRAB_0_90
 945         size_t disks;
 946 # else
 947         int disks;
 948 #endif
 949         int counter;
 950         char name[DATA_MAX_NAME_LEN];
 951
 952         if ((ds = sg_get_disk_io_stats(&disks)) == NULL)
 953                 return (0);
 954
 955         for (counter=0; counter < disks; counter++) {
 956                 strncpy(name, ds->disk_name, sizeof(name));
 957                 name[sizeof(name)-1] = '\0'; /* strncpy doesn't terminate longer strings */
 958                 disk_submit (name, "disk_octets", ds->read_bytes, ds->write_bytes);
 959                 ds++;
 960         }
 961 /* #endif defined(HAVE_LIBSTATGRAB) */
 962
 963 #elif defined(HAVE_PERFSTAT)
 964         derive_t read_sectors;
 965         derive_t write_sectors;
 966         derive_t read_time;
 967         derive_t write_time;
 968         derive_t read_ops;
 969         derive_t write_ops;
 970         perfstat_id_t firstpath;
 971         int rnumdisk;
 972         int i;
 973
 974         if ((numdisk = perfstat_disk(NULL, NULL, sizeof(perfstat_disk_t), 0)) < 0)
 975         {
 976                 char errbuf[1024];
 977                 WARNING ("disk plugin: perfstat_disk: %s",
 978                                 sstrerror (errno, errbuf, sizeof (errbuf)));
 979                 return (-1);
 980         }
 981
 982         if (numdisk != pnumdisk || stat_disk==NULL) {
 983                 if (stat_disk!=NULL)
 984                         free(stat_disk);
 985                 stat_disk = (perfstat_disk_t *)calloc(numdisk, sizeof(perfstat_disk_t));
 986         }
 987         pnumdisk = numdisk;
 988
 989         firstpath.name[0]='\0';
 990         if ((rnumdisk = perfstat_disk(&firstpath, stat_disk, sizeof(perfstat_disk_t), numdisk)) < 0)
 991         {
 992                 char errbuf[1024];
 993                 WARNING ("disk plugin: perfstat_disk : %s",
 994                                 sstrerror (errno, errbuf, sizeof (errbuf)));
 995                 return (-1);
 996         }
 997
 998         for (i = 0; i < rnumdisk; i++)
 999         {
1000                 read_sectors = stat_disk[i].rblks*stat_disk[i].bsize;
1001                 write_sectors = stat_disk[i].wblks*stat_disk[i].bsize;
1002                 disk_submit (stat_disk[i].name, "disk_octets", read_sectors, write_sectors);
1003
1004                 read_ops = stat_disk[i].xrate;
1005                 write_ops = stat_disk[i].xfers - stat_disk[i].xrate;
1006                 disk_submit (stat_disk[i].name, "disk_ops", read_ops, write_ops);
1007
1008                 read_time = stat_disk[i].rserv;
1009                 read_time *= ((double)(_system_configuration.Xint)/(double)(_system_configuration.Xfrac)) / 1000000.0;
1010                 write_time = stat_disk[i].wserv;
1011                 write_time *= ((double)(_system_configuration.Xint)/(double)(_system_configuration.Xfrac)) / 1000000.0;
1012                 disk_submit (stat_disk[i].name, "disk_time", read_time, write_time);
1013         }
1014 #endif /* defined(HAVE_PERFSTAT) */
1015
1016         return (0);
1017 } /* int disk_read */
1018
1019 void module_register (void)
1020 {
1021   plugin_register_config ("disk", disk_config,
1022       config_keys, config_keys_num);
1023   plugin_register_init ("disk", disk_init);
1024   plugin_register_read ("disk", disk_read);
1025 } /* void module_register */