src/disk.c

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