요약·해설과 원문, 전문 번역을 서로 분리했습니다. API 이름, symbol, source path는 원문 표기를 사용합니다.
1. 요약·해설
원문의 핵심 논리와 kernel programming 관점의 보충 설명입니다. 아래의 전문 번역과는 별도로 작성했습니다.
Integrity metadata와 partition
sysfs-block:106-218T10 Protection Information과 application tag, partition scan·alignment·통계를 다룹니다.
Queue crypto와 discard
sysfs-block:221-426Inline encryption capability, discard hardware·software limit, DAX, FUA, independent access range를 다룹니다.
Queue poll, timeout과 I/O limit
sysfs-block:429-680Polling, timeout, request·segment·zone 한계, merge, readahead, affinity를 다룹니다.
Scheduler, write와 zoned storage
sysfs-block:681-881I/O scheduler, stable write, writeback throttling, cache, zeroing, zone model과 17개 통계 field를 다룹니다.
2. 영어 원문 전체
번역 기준이 된 Linux v6.18.37 원문입니다. 줄 번호는 이 버전의 파일 좌표입니다.
원문 전체 펼치기
1
What: /sys/block/<disk>/alignment_offset
2
Date: April 2009
3
Contact: Martin K. Petersen <[email protected]>
4
Description:
5
Storage devices may report a physical block size that is
6
bigger than the logical block size (for instance a drive
7
with 4KB physical sectors exposing 512-byte logical
8
blocks to the operating system). This parameter
9
indicates how many bytes the beginning of the device is
10
offset from the disk's natural alignment.
13
What: /sys/block/<disk>/discard_alignment
14
Date: May 2011
15
Contact: Martin K. Petersen <[email protected]>
16
Description:
17
Devices that support discard functionality may
18
internally allocate space in units that are bigger than
19
the exported logical block size. The discard_alignment
20
parameter indicates how many bytes the beginning of the
21
device is offset from the internal allocation unit's
22
natural alignment.
24
What: /sys/block/<disk>/atomic_write_max_bytes
25
Date: February 2024
26
Contact: Himanshu Madhani <[email protected]>
27
Description:
28
[RO] This parameter specifies the maximum atomic write
29
size reported by the device. This parameter is relevant
30
for merging of writes, where a merged atomic write
31
operation must not exceed this number of bytes.
32
This parameter may be greater than the value in
33
atomic_write_unit_max_bytes as
34
atomic_write_unit_max_bytes will be rounded down to a
35
power-of-two and atomic_write_unit_max_bytes may also be
36
limited by some other queue limits, such as max_segments.
37
This parameter - along with atomic_write_unit_min_bytes
38
and atomic_write_unit_max_bytes - will not be larger than
39
max_hw_sectors_kb, but may be larger than max_sectors_kb.
42
What: /sys/block/<disk>/atomic_write_unit_min_bytes
43
Date: February 2024
44
Contact: Himanshu Madhani <[email protected]>
45
Description:
46
[RO] This parameter specifies the smallest block which can
47
be written atomically with an atomic write operation. All
48
atomic write operations must begin at a
49
atomic_write_unit_min boundary and must be multiples of
50
atomic_write_unit_min. This value must be a power-of-two.
53
What: /sys/block/<disk>/atomic_write_unit_max_bytes
54
Date: February 2024
55
Contact: Himanshu Madhani <[email protected]>
56
Description:
57
[RO] This parameter defines the largest block which can be
58
written atomically with an atomic write operation. This
59
value must be a multiple of atomic_write_unit_min and must
60
be a power-of-two. This value will not be larger than
61
atomic_write_max_bytes.
64
What: /sys/block/<disk>/atomic_write_boundary_bytes
65
Date: February 2024
66
Contact: Himanshu Madhani <[email protected]>
67
Description:
68
[RO] A device may need to internally split an atomic write I/O
69
which straddles a given logical block address boundary. This
70
parameter specifies the size in bytes of the atomic boundary if
71
one is reported by the device. This value must be a
72
power-of-two and at least the size as in
73
atomic_write_unit_max_bytes.
74
Any attempt to merge atomic write I/Os must not result in a
75
merged I/O which crosses this boundary (if any).
78
What: /sys/block/<disk>/diskseq
79
Date: February 2021
80
Contact: Matteo Croce <[email protected]>
81
Description:
82
The /sys/block/<disk>/diskseq files reports the disk
83
sequence number, which is a monotonically increasing
84
number assigned to every drive.
85
Some devices, like the loop device, refresh such number
86
every time the backing file is changed.
87
The value type is 64 bit unsigned.
90
What: /sys/block/<disk>/inflight
91
Date: October 2009
92
Contact: Jens Axboe <[email protected]>, Nikanth Karthikesan <[email protected]>
93
Description:
94
Reports the number of I/O requests currently in progress
95
(pending / in flight) in a device driver. This can be less
96
than the number of requests queued in the block device queue.
97
The report contains 2 fields: one for read requests
98
and one for write requests.
99
The value type is unsigned int.
100
Cf. Documentation/block/stat.rst which contains a single value for
101
requests in flight.
102
This is related to /sys/block/<disk>/queue/nr_requests
103
and for SCSI device also its queue_depth.
106
What: /sys/block/<disk>/integrity/device_is_integrity_capable
107
Date: July 2014
108
Contact: Martin K. Petersen <[email protected]>
109
Description:
110
Indicates whether a storage device is capable of storing
111
integrity metadata. Set if the device is T10 PI-capable.
112
This flag is set to 1 if the storage media is formatted
113
with T10 Protection Information. If the storage media is
114
not formatted with T10 Protection Information, this flag
115
is set to 0.
118
What: /sys/block/<disk>/integrity/format
119
Date: June 2008
120
Contact: Martin K. Petersen <[email protected]>
121
Description:
122
Metadata format for integrity capable block device.
123
E.g. T10-DIF-TYPE1-CRC.
124
This field describes the type of T10 Protection Information
125
that the block device can send and receive.
126
If the device can store application integrity metadata but
127
no T10 Protection Information profile is used, this field
128
contains "nop".
129
If the device does not support integrity metadata, this
130
field contains "none".
133
What: /sys/block/<disk>/integrity/protection_interval_bytes
134
Date: July 2015
135
Contact: Martin K. Petersen <[email protected]>
136
Description:
137
Describes the number of data bytes which are protected
138
by one integrity tuple. Typically the device's logical
139
block size.
142
What: /sys/block/<disk>/integrity/read_verify
143
Date: June 2008
144
Contact: Martin K. Petersen <[email protected]>
145
Description:
146
Indicates whether the block layer should verify the
147
integrity of read requests serviced by devices that
148
support sending integrity metadata.
151
What: /sys/block/<disk>/integrity/tag_size
152
Date: June 2008
153
Contact: Martin K. Petersen <[email protected]>
154
Description:
155
Number of bytes of integrity tag space available per
156
protection_interval_bytes, which is typically
157
the device's logical block size.
158
This field describes the size of the application tag
159
if the storage device is formatted with T10 Protection
160
Information and permits use of the application tag.
161
The tag_size is reported in bytes and indicates the
162
space available for adding an opaque tag to each block
163
(protection_interval_bytes).
164
If the device does not support T10 Protection Information
165
(even if the device provides application integrity
166
metadata space), this field is set to 0.
169
What: /sys/block/<disk>/integrity/write_generate
170
Date: June 2008
171
Contact: Martin K. Petersen <[email protected]>
172
Description:
173
Indicates whether the block layer should automatically
174
generate checksums for write requests bound for
175
devices that support receiving integrity metadata.
178
What: /sys/block/<disk>/partscan
179
Date: May 2024
180
Contact: Christoph Hellwig <[email protected]>
181
Description:
182
The /sys/block/<disk>/partscan files reports if partition
183
scanning is enabled for the disk. It returns "1" if partition
184
scanning is enabled, or "0" if not. The value type is a 32-bit
185
unsigned integer, but only "0" and "1" are valid values.
188
What: /sys/block/<disk>/<partition>/alignment_offset
189
Date: April 2009
190
Contact: Martin K. Petersen <[email protected]>
191
Description:
192
Storage devices may report a physical block size that is
193
bigger than the logical block size (for instance a drive
194
with 4KB physical sectors exposing 512-byte logical
195
blocks to the operating system). This parameter
196
indicates how many bytes the beginning of the partition
197
is offset from the disk's natural alignment.
200
What: /sys/block/<disk>/<partition>/discard_alignment
201
Date: May 2011
202
Contact: Martin K. Petersen <[email protected]>
203
Description:
204
Devices that support discard functionality may
205
internally allocate space in units that are bigger than
206
the exported logical block size. The discard_alignment
207
parameter indicates how many bytes the beginning of the
208
partition is offset from the internal allocation unit's
209
natural alignment.
212
What: /sys/block/<disk>/<partition>/stat
213
Date: February 2008
214
Contact: Jerome Marchand <[email protected]>
215
Description:
216
The /sys/block/<disk>/<partition>/stat files display the
217
I/O statistics of partition <partition>. The format is the
218
same as the format of /sys/block/<disk>/stat.
221
What: /sys/block/<disk>/queue/add_random
222
Date: June 2010
223
Contact: [email protected]
224
Description:
225
[RW] This file allows to turn off the disk entropy contribution.
226
Default value of this file is '1'(on).
229
What: /sys/block/<disk>/queue/chunk_sectors
230
Date: September 2016
231
Contact: Hannes Reinecke <[email protected]>
232
Description:
233
[RO] chunk_sectors has different meaning depending on the type
234
of the disk. For a RAID device (dm-raid), chunk_sectors
235
indicates the size in 512B sectors of the RAID volume stripe
236
segment. For a zoned block device, either host-aware or
237
host-managed, chunk_sectors indicates the size in 512B sectors
238
of the zones of the device, with the eventual exception of the
239
last zone of the device which may be smaller.
242
What: /sys/block/<disk>/queue/crypto/
243
Date: February 2022
244
Contact: [email protected]
245
Description:
246
The presence of this subdirectory of /sys/block/<disk>/queue/
247
indicates that the device supports inline encryption. This
248
subdirectory contains files which describe the inline encryption
249
capabilities of the device. For more information about inline
250
encryption, refer to Documentation/block/inline-encryption.rst.
253
What: /sys/block/<disk>/queue/crypto/hw_wrapped_keys
254
Date: February 2025
255
Contact: [email protected]
256
Description:
257
[RO] The presence of this file indicates that the device
258
supports hardware-wrapped inline encryption keys, i.e. key blobs
259
that can only be unwrapped and used by dedicated hardware. For
260
more information about hardware-wrapped inline encryption keys,
261
see Documentation/block/inline-encryption.rst.
264
What: /sys/block/<disk>/queue/crypto/max_dun_bits
265
Date: February 2022
266
Contact: [email protected]
267
Description:
268
[RO] This file shows the maximum length, in bits, of data unit
269
numbers accepted by the device in inline encryption requests.
272
What: /sys/block/<disk>/queue/crypto/modes/<mode>
273
Date: February 2022
274
Contact: [email protected]
275
Description:
276
[RO] For each crypto mode (i.e., encryption/decryption
277
algorithm) the device supports with inline encryption, a file
278
will exist at this location. It will contain a hexadecimal
279
number that is a bitmask of the supported data unit sizes, in
280
bytes, for that crypto mode.
282
Currently, the crypto modes that may be supported are:
284
* AES-256-XTS
285
* AES-128-CBC-ESSIV
286
* Adiantum
288
For example, if a device supports AES-256-XTS inline encryption
289
with data unit sizes of 512 and 4096 bytes, the file
290
/sys/block/<disk>/queue/crypto/modes/AES-256-XTS will exist and
291
will contain "0x1200".
294
What: /sys/block/<disk>/queue/crypto/num_keyslots
295
Date: February 2022
296
Contact: [email protected]
297
Description:
298
[RO] This file shows the number of keyslots the device has for
299
use with inline encryption.
302
What: /sys/block/<disk>/queue/crypto/raw_keys
303
Date: February 2025
304
Contact: [email protected]
305
Description:
306
[RO] The presence of this file indicates that the device
307
supports raw inline encryption keys, i.e. keys that are managed
308
in raw, plaintext form in software.
311
What: /sys/block/<disk>/queue/dax
312
Date: June 2016
313
Contact: [email protected]
314
Description:
315
[RO] This file indicates whether the device supports Direct
316
Access (DAX), used by CPU-addressable storage to bypass the
317
pagecache. It shows '1' if true, '0' if not.
320
What: /sys/block/<disk>/queue/discard_granularity
321
Date: May 2011
322
Contact: Martin K. Petersen <[email protected]>
323
Description:
324
[RO] Devices that support discard functionality may internally
325
allocate space using units that are bigger than the logical
326
block size. The discard_granularity parameter indicates the size
327
of the internal allocation unit in bytes if reported by the
328
device. Otherwise the discard_granularity will be set to match
329
the device's physical block size. A discard_granularity of 0
330
means that the device does not support discard functionality.
333
What: /sys/block/<disk>/queue/discard_max_bytes
334
Date: May 2011
335
Contact: Martin K. Petersen <[email protected]>
336
Description:
337
[RW] While discard_max_hw_bytes is the hardware limit for the
338
device, this setting is the software limit. Some devices exhibit
339
large latencies when large discards are issued, setting this
340
value lower will make Linux issue smaller discards and
341
potentially help reduce latencies induced by large discard
342
operations.
345
What: /sys/block/<disk>/queue/discard_max_hw_bytes
346
Date: July 2015
347
Contact: [email protected]
348
Description:
349
[RO] Devices that support discard functionality may have
350
internal limits on the number of bytes that can be trimmed or
351
unmapped in a single operation. The `discard_max_hw_bytes`
352
parameter is set by the device driver to the maximum number of
353
bytes that can be discarded in a single operation. Discard
354
requests issued to the device must not exceed this limit. A
355
`discard_max_hw_bytes` value of 0 means that the device does not
356
support discard functionality.
359
What: /sys/block/<disk>/queue/discard_zeroes_data
360
Date: May 2011
361
Contact: Martin K. Petersen <[email protected]>
362
Description:
363
[RO] Will always return 0. Don't rely on any specific behavior
364
for discards, and don't read this file.
367
What: /sys/block/<disk>/queue/dma_alignment
368
Date: May 2022
369
Contact: [email protected]
370
Description:
371
Reports the alignment that user space addresses must have to be
372
used for raw block device access with O_DIRECT and other driver
373
specific passthrough mechanisms.
376
What: /sys/block/<disk>/queue/fua
377
Date: May 2018
378
Contact: [email protected]
379
Description:
380
[RO] Whether or not the block driver supports the FUA flag for
381
write requests. FUA stands for Force Unit Access. If the FUA
382
flag is set that means that write requests must bypass the
383
volatile cache of the storage device.
386
What: /sys/block/<disk>/queue/hw_sector_size
387
Date: January 2008
388
Contact: [email protected]
389
Description:
390
[RO] This is the hardware sector size of the device, in bytes.
393
What: /sys/block/<disk>/queue/independent_access_ranges/
394
Date: October 2021
395
Contact: [email protected]
396
Description:
397
[RO] The presence of this sub-directory of the
398
/sys/block/xxx/queue/ directory indicates that the device is
399
capable of executing requests targeting different sector ranges
400
in parallel. For instance, single LUN multi-actuator hard-disks
401
will have an independent_access_ranges directory if the device
402
correctly advertises the sector ranges of its actuators.
404
The independent_access_ranges directory contains one directory
405
per access range, with each range described using the sector
406
(RO) attribute file to indicate the first sector of the range
407
and the nr_sectors (RO) attribute file to indicate the total
408
number of sectors in the range starting from the first sector of
409
the range. For example, a dual-actuator hard-disk will have the
410
following independent_access_ranges entries.::
412
$ tree /sys/block/<disk>/queue/independent_access_ranges/
413
/sys/block/<disk>/queue/independent_access_ranges/
414
|-- 0
415
| |-- nr_sectors
416
| `-- sector
417
`-- 1
418
|-- nr_sectors
419
`-- sector
421
The sector and nr_sectors attributes use 512B sector unit,
422
regardless of the actual block size of the device. Independent
423
access ranges do not overlap and include all sectors within the
424
device capacity. The access ranges are numbered in increasing
425
order of the range start sector, that is, the sector attribute
426
of range 0 always has the value 0.
429
What: /sys/block/<disk>/queue/io_poll
430
Date: November 2015
431
Contact: [email protected]
432
Description:
433
[RW] When read, this file shows whether polling is enabled (1)
434
or disabled (0). Writing '0' to this file will disable polling
435
for this device. Writing any non-zero value will enable this
436
feature.
439
What: /sys/block/<disk>/queue/io_poll_delay
440
Date: November 2016
441
Contact: [email protected]
442
Description:
443
[RW] This was used to control what kind of polling will be
444
performed. It is now fixed to -1, which is classic polling.
445
In this mode, the CPU will repeatedly ask for completions
446
without giving up any time.
447
<deprecated>
450
What: /sys/block/<disk>/queue/io_timeout
451
Date: November 2018
452
Contact: Weiping Zhang <[email protected]>
453
Description:
454
[RW] io_timeout is the request timeout in milliseconds. If a
455
request does not complete in this time then the block driver
456
timeout handler is invoked. That timeout handler can decide to
457
retry the request, to fail it or to start a device recovery
458
strategy.
461
What: /sys/block/<disk>/queue/iostats
462
Date: January 2009
463
Contact: [email protected]
464
Description:
465
[RW] This file is used to control (on/off) the iostats
466
accounting of the disk.
468
What: /sys/block/<disk>/queue/iostats_passthrough
469
Date: October 2024
470
Contact: [email protected]
471
Description:
472
[RW] This file is used to control (on/off) the iostats
473
accounting of the disk for passthrough commands.
476
What: /sys/block/<disk>/queue/logical_block_size
477
Date: May 2009
478
Contact: Martin K. Petersen <[email protected]>
479
Description:
480
[RO] This is the smallest unit the storage device can address.
481
It is typically 512 bytes.
484
What: /sys/block/<disk>/queue/max_active_zones
485
Date: July 2020
486
Contact: Niklas Cassel <[email protected]>
487
Description:
488
[RO] For zoned block devices (zoned attribute indicating
489
"host-managed" or "host-aware"), the sum of zones belonging to
490
any of the zone states: EXPLICIT OPEN, IMPLICIT OPEN or CLOSED,
491
is limited by this value. If this value is 0, there is no limit.
493
If the host attempts to exceed this limit, the driver should
494
report this error with BLK_STS_ZONE_ACTIVE_RESOURCE, which user
495
space may see as the EOVERFLOW errno.
498
What: /sys/block/<disk>/queue/max_discard_segments
499
Date: February 2017
500
Contact: [email protected]
501
Description:
502
[RO] The maximum number of DMA scatter/gather entries in a
503
discard request.
506
What: /sys/block/<disk>/queue/max_hw_sectors_kb
507
Date: September 2004
508
Contact: [email protected]
509
Description:
510
[RO] This is the maximum number of kilobytes supported in a
511
single data transfer.
514
What: /sys/block/<disk>/queue/max_integrity_segments
515
Date: September 2010
516
Contact: [email protected]
517
Description:
518
[RO] Maximum number of elements in a DMA scatter/gather list
519
with integrity data that will be submitted by the block layer
520
core to the associated block driver.
523
What: /sys/block/<disk>/queue/max_open_zones
524
Date: July 2020
525
Contact: Niklas Cassel <[email protected]>
526
Description:
527
[RO] For zoned block devices (zoned attribute indicating
528
"host-managed" or "host-aware"), the sum of zones belonging to
529
any of the zone states: EXPLICIT OPEN or IMPLICIT OPEN, is
530
limited by this value. If this value is 0, there is no limit.
533
What: /sys/block/<disk>/queue/max_sectors_kb
534
Date: September 2004
535
Contact: [email protected]
536
Description:
537
[RW] This is the maximum number of kilobytes that the block
538
layer will allow for a filesystem request. Must be smaller than
539
or equal to the maximum size allowed by the hardware. Write 0
540
to use default kernel settings.
543
What: /sys/block/<disk>/queue/max_segment_size
544
Date: March 2010
545
Contact: [email protected]
546
Description:
547
[RO] Maximum size in bytes of a single element in a DMA
548
scatter/gather list.
550
What: /sys/block/<disk>/queue/max_write_streams
551
Date: November 2024
552
Contact: [email protected]
553
Description:
554
[RO] Maximum number of write streams supported, 0 if not
555
supported. If supported, valid values are 1 through
556
max_write_streams, inclusive.
558
What: /sys/block/<disk>/queue/write_stream_granularity
559
Date: November 2024
560
Contact: [email protected]
561
Description:
562
[RO] Granularity of a write stream in bytes. The granularity
563
of a write stream is the size that should be discarded or
564
overwritten together to avoid write amplification in the device.
566
What: /sys/block/<disk>/queue/max_segments
567
Date: March 2010
568
Contact: [email protected]
569
Description:
570
[RO] Maximum number of elements in a DMA scatter/gather list
571
that is submitted to the associated block driver.
574
What: /sys/block/<disk>/queue/minimum_io_size
575
Date: April 2009
576
Contact: Martin K. Petersen <[email protected]>
577
Description:
578
[RO] Storage devices may report a granularity or preferred
579
minimum I/O size which is the smallest request the device can
580
perform without incurring a performance penalty. For disk
581
drives this is often the physical block size. For RAID arrays
582
it is often the stripe chunk size. A properly aligned multiple
583
of minimum_io_size is the preferred request size for workloads
584
where a high number of I/O operations is desired.
587
What: /sys/block/<disk>/queue/nomerges
588
Date: January 2010
589
Contact: [email protected]
590
Description:
591
[RW] Standard I/O elevator operations include attempts to merge
592
contiguous I/Os. For known random I/O loads these attempts will
593
always fail and result in extra cycles being spent in the
594
kernel. This allows one to turn off this behavior on one of two
595
ways: When set to 1, complex merge checks are disabled, but the
596
simple one-shot merges with the previous I/O request are
597
enabled. When set to 2, all merge tries are disabled. The
598
default value is 0 - which enables all types of merge tries.
601
What: /sys/block/<disk>/queue/nr_requests
602
Date: July 2003
603
Contact: [email protected]
604
Description:
605
[RW] This controls how many requests may be allocated in the
606
block layer. Noted this value only represents the quantity for a
607
single blk_mq_tags instance. The actual number for the entire
608
device depends on the hardware queue count, whether elevator is
609
enabled, and whether tags are shared.
612
What: /sys/block/<disk>/queue/nr_zones
613
Date: November 2018
614
Contact: Damien Le Moal <[email protected]>
615
Description:
616
[RO] nr_zones indicates the total number of zones of a zoned
617
block device ("host-aware" or "host-managed" zone model). For
618
regular block devices, the value is always 0.
621
What: /sys/block/<disk>/queue/optimal_io_size
622
Date: April 2009
623
Contact: Martin K. Petersen <[email protected]>
624
Description:
625
[RO] Storage devices may report an optimal I/O size, which is
626
the device's preferred unit for sustained I/O. This is rarely
627
reported for disk drives. For RAID arrays it is usually the
628
stripe width or the internal track size. A properly aligned
629
multiple of optimal_io_size is the preferred request size for
630
workloads where sustained throughput is desired. If no optimal
631
I/O size is reported this file contains 0.
634
What: /sys/block/<disk>/queue/physical_block_size
635
Date: May 2009
636
Contact: Martin K. Petersen <[email protected]>
637
Description:
638
[RO] This is the smallest unit a physical storage device can
639
write atomically. It is usually the same as the logical block
640
size but may be bigger. One example is SATA drives with 4KB
641
sectors that expose a 512-byte logical block size to the
642
operating system. For stacked block devices the
643
physical_block_size variable contains the maximum
644
physical_block_size of the component devices.
647
What: /sys/block/<disk>/queue/read_ahead_kb
648
Date: May 2004
649
Contact: [email protected]
650
Description:
651
[RW] Maximum number of kilobytes to read-ahead for filesystems
652
on this block device.
654
For MADV_HUGEPAGE, the readahead size may exceed this setting
655
since its granularity is based on the hugepage size.
658
What: /sys/block/<disk>/queue/rotational
659
Date: January 2009
660
Contact: [email protected]
661
Description:
662
[RW] This file is used to stat if the device is of rotational
663
type or non-rotational type.
666
What: /sys/block/<disk>/queue/rq_affinity
667
Date: September 2008
668
Contact: [email protected]
669
Description:
670
[RW] If this option is '1', the block layer will migrate request
671
completions to the cpu "group" that originally submitted the
672
request. For some workloads this provides a significant
673
reduction in CPU cycles due to caching effects.
675
For storage configurations that need to maximize distribution of
676
completion processing setting this option to '2' forces the
677
completion to run on the requesting cpu (bypassing the "group"
678
aggregation logic).
681
What: /sys/block/<disk>/queue/scheduler
682
Date: October 2004
683
Contact: [email protected]
684
Description:
685
[RW] When read, this file will display the current and available
686
IO schedulers for this block device. The currently active IO
687
scheduler will be enclosed in [] brackets. Writing an IO
688
scheduler name to this file will switch control of this block
689
device to that new IO scheduler. Note that writing an IO
690
scheduler name to this file will attempt to load that IO
691
scheduler module, if it isn't already present in the system.
694
What: /sys/block/<disk>/queue/stable_writes
695
Date: September 2020
696
Contact: [email protected]
697
Description:
698
[RW] This file will contain '1' if memory must not be modified
699
while it is being used in a write request to this device. When
700
this is the case and the kernel is performing writeback of a
701
page, the kernel will wait for writeback to complete before
702
allowing the page to be modified again, rather than allowing
703
immediate modification as is normally the case. This
704
restriction arises when the device accesses the memory multiple
705
times where the same data must be seen every time -- for
706
example, once to calculate a checksum and once to actually write
707
the data. If no such restriction exists, this file will contain
708
'0'. This file is writable for testing purposes.
710
What: /sys/block/<disk>/queue/virt_boundary_mask
711
Date: April 2021
712
Contact: [email protected]
713
Description:
714
[RO] This file shows the I/O segment memory alignment mask for
715
the block device. I/O requests to this device will be split
716
between segments wherever either the memory address of the end
717
of the previous segment or the memory address of the beginning
718
of the current segment is not aligned to virt_boundary_mask + 1
719
bytes.
722
What: /sys/block/<disk>/queue/wbt_lat_usec
723
Date: November 2016
724
Contact: [email protected]
725
Description:
726
[RW] If the device is registered for writeback throttling, then
727
this file shows the target minimum read latency. If this latency
728
is exceeded in a given window of time (see curr_win_nsec), then
729
the writeback throttling will start scaling back writes. Writing
730
a value of '0' to this file disables the feature. Writing a
731
value of '-1' to this file resets the value to the default
732
setting.
735
What: /sys/block/<disk>/queue/write_cache
736
Date: April 2016
737
Contact: [email protected]
738
Description:
739
[RW] When read, this file will display whether the device has
740
write back caching enabled or not. It will return "write back"
741
for the former case, and "write through" for the latter. Writing
742
to this file can change the kernels view of the device, but it
743
doesn't alter the device state. This means that it might not be
744
safe to toggle the setting from "write back" to "write through",
745
since that will also eliminate cache flushes issued by the
746
kernel.
749
What: /sys/block/<disk>/queue/write_same_max_bytes
750
Date: January 2012
751
Contact: Martin K. Petersen <[email protected]>
752
Description:
753
[RO] Some devices support a write same operation in which a
754
single data block can be written to a range of several
755
contiguous blocks on storage. This can be used to wipe areas on
756
disk or to initialize drives in a RAID configuration.
757
write_same_max_bytes indicates how many bytes can be written in
758
a single write same command. If write_same_max_bytes is 0, write
759
same is not supported by the device.
762
What: /sys/block/<disk>/queue/write_zeroes_max_bytes
763
Date: November 2016
764
Contact: Chaitanya Kulkarni <[email protected]>
765
Description:
766
[RO] Devices that support write zeroes operation in which a
767
single request can be issued to zero out the range of contiguous
768
blocks on storage without having any payload in the request.
769
This can be used to optimize writing zeroes to the devices.
770
write_zeroes_max_bytes indicates how many bytes can be written
771
in a single write zeroes command. If write_zeroes_max_bytes is
772
0, write zeroes is not supported by the device.
775
What: /sys/block/<disk>/queue/write_zeroes_unmap_max_hw_bytes
776
Date: January 2025
777
Contact: Zhang Yi <[email protected]>
778
Description:
779
[RO] This file indicates whether a device supports zeroing data
780
in a specified block range without incurring the cost of
781
physically writing zeroes to the media for each individual
782
block. If this parameter is set to write_zeroes_max_bytes, the
783
device implements a zeroing operation which opportunistically
784
avoids writing zeroes to media while still guaranteeing that
785
subsequent reads from the specified block range will return
786
zeroed data. This operation is a best-effort optimization, a
787
device may fall back to physically writing zeroes to the media
788
due to other factors such as misalignment or being asked to
789
clear a block range smaller than the device's internal
790
allocation unit. If this parameter is set to 0, the device may
791
have to write each logical block media during a zeroing
792
operation.
795
What: /sys/block/<disk>/queue/write_zeroes_unmap_max_bytes
796
Date: January 2025
797
Contact: Zhang Yi <[email protected]>
798
Description:
799
[RW] While write_zeroes_unmap_max_hw_bytes is the hardware limit
800
for the device, this setting is the software limit. Since the
801
unmap write zeroes operation is a best-effort optimization, some
802
devices may still physically writing zeroes to media. So the
803
speed of this operation is not guaranteed. Writing a value of
804
'0' to this file disables this operation. Otherwise, this
805
parameter should be equal to write_zeroes_unmap_max_hw_bytes.
808
What: /sys/block/<disk>/queue/zone_append_max_bytes
809
Date: May 2020
810
Contact: [email protected]
811
Description:
812
[RO] This is the maximum number of bytes that can be written to
813
a sequential zone of a zoned block device using a zone append
814
write operation (REQ_OP_ZONE_APPEND). This value is always 0 for
815
regular block devices.
818
What: /sys/block/<disk>/queue/zone_write_granularity
819
Date: January 2021
820
Contact: [email protected]
821
Description:
822
[RO] This indicates the alignment constraint, in bytes, for
823
write operations in sequential zones of zoned block devices
824
(devices with a zoned attributed that reports "host-managed" or
825
"host-aware"). This value is always 0 for regular block devices.
828
What: /sys/block/<disk>/queue/zoned
829
Date: September 2016
830
Contact: Damien Le Moal <[email protected]>
831
Description:
832
[RO] zoned indicates if the device is a zoned block device and
833
the zone model of the device if it is indeed zoned. The
834
possible values indicated by zoned are "none" for regular block
835
devices and "host-aware" or "host-managed" for zoned block
836
devices. The characteristics of host-aware and host-managed
837
zoned block devices are described in the ZBC (Zoned Block
838
Commands) and ZAC (Zoned Device ATA Command Set) standards.
839
These standards also define the "drive-managed" zone model.
840
However, since drive-managed zoned block devices do not support
841
zone commands, they will be treated as regular block devices and
842
zoned will report "none".
845
What: /sys/block/<disk>/hidden
846
Date: March 2023
847
Contact: [email protected]
848
Description:
849
[RO] the block device is hidden. it doesn’t produce events, and
850
can’t be opened from userspace or using blkdev_get*.
851
Used for the underlying components of multipath devices.
854
What: /sys/block/<disk>/stat
855
Date: February 2008
856
Contact: Jerome Marchand <[email protected]>
857
Description:
858
The /sys/block/<disk>/stat files displays the I/O
859
statistics of disk <disk>. They contain 11 fields:
861
== ==============================================
862
1 reads completed successfully
863
2 reads merged
864
3 sectors read
865
4 time spent reading (ms)
866
5 writes completed
867
6 writes merged
868
7 sectors written
869
8 time spent writing (ms)
870
9 I/Os currently in progress
871
10 time spent doing I/Os (ms)
872
11 weighted time spent doing I/Os (ms)
873
12 discards completed
874
13 discards merged
875
14 sectors discarded
876
15 time spent discarding (ms)
877
16 flush requests completed
878
17 time spent flushing (ms)
879
== ==============================================
881
For more details refer Documentation/admin-guide/iostats.rst
3. 한국어 전문 번역
영어 원문의 문단 순서와 의미를 유지한 전체 번역입니다. 코드, 함수명, symbol과 URL은 원문 표기를 유지합니다.
Disk alignment, atomic write와 실행 중 I/O
1-103| 경로 | 작성 시점·담당 | 의미 |
|---|---|---|
| /sys/block/<disk>/alignment_offset | 2009년 4월 · Martin K. Petersen | Storage device는 logical block보다 큰 physical block size를 보고할 수 있다. 예를 들어 4KB physical sector drive가 OS에는 512-byte logical block을 노출할 수 있다. 이 값은 device 시작점이 disk의 자연스러운 alignment에서 몇 byte 떨어져 있는지 나타낸다. |
| /sys/block/<disk>/discard_alignment | 2011년 5월 · Martin K. Petersen | Discard 지원 device가 exported logical block보다 큰 단위로 내부 공간을 할당할 때, device 시작점이 내부 allocation unit의 자연스러운 alignment에서 몇 byte 떨어져 있는지 나타낸다. |
| /sys/block/<disk>/atomic_write_max_bytes | 2024년 2월 · Himanshu Madhani · read-only | Device가 보고한 최대 atomic write 크기다. Write를 merge할 때 merged atomic write가 이 값을 넘으면 안 된다. atomic_write_unit_max_bytes는 2의 거듭제곱으로 내림되고 max_segments 같은 queue limit에도 제한될 수 있어 이 값이 더 클 수 있다. 이 값과 두 atomic_write_unit 값은 max_hw_sectors_kb보다 크지 않지만 max_sectors_kb보다 클 수 있다. |
| /sys/block/<disk>/atomic_write_unit_min_bytes | 2024년 2월 · Himanshu Madhani · read-only | Atomic write operation으로 원자적으로 쓸 수 있는 가장 작은 block이다. 모든 atomic write는 이 단위의 경계에서 시작하고 이 단위의 배수여야 한다. 값은 2의 거듭제곱이어야 한다. |
| /sys/block/<disk>/atomic_write_unit_max_bytes | 2024년 2월 · Himanshu Madhani · read-only | Atomic write operation으로 원자적으로 쓸 수 있는 가장 큰 block이다. atomic_write_unit_min의 배수이자 2의 거듭제곱이어야 하며 atomic_write_max_bytes보다 클 수 없다. |
| /sys/block/<disk>/atomic_write_boundary_bytes | 2024년 2월 · Himanshu Madhani · read-only | Device가 특정 logical block address 경계를 가로지르는 atomic write I/O를 내부에서 나눠야 할 때 그 경계 크기를 byte로 나타낸다. 2의 거듭제곱이며 atomic_write_unit_max_bytes 이상이어야 한다. Atomic write I/O를 merge한 결과가 이 경계를 가로질러서는 안 된다. |
| /sys/block/<disk>/diskseq | 2021년 2월 · Matteo Croce | 모든 drive에 할당하는 단조 증가 64-bit unsigned sequence number다. Loop device 같은 일부 device는 backing file이 바뀔 때마다 번호를 갱신한다. |
| /sys/block/<disk>/inflight | 2009년 10월 · Jens Axboe, Nikanth Karthikesan | Device driver에서 현재 진행 중인 pending/in-flight I/O request 수를 보고한다. Block queue에 들어 있는 request 수보다 작을 수 있다. unsigned int인 read request와 write request 두 field를 제공한다. Documentation/block/stat.rst의 단일 in-flight 값, queue/nr_requests, SCSI queue_depth와 관련된다. |
Integrity metadata
106-175아래 항목의 담당자는 Martin K. Petersen <[email protected]>이다.
| 경로 | 작성 시점 | 의미 |
|---|---|---|
| /sys/block/<disk>/integrity/device_is_integrity_capable | 2014년 7월 | Storage device가 integrity metadata를 저장할 수 있는지 나타낸다. T10 PI-capable device이며 media가 T10 Protection Information으로 format돼 있으면 1, 아니면 0이다. |
| /sys/block/<disk>/integrity/format | 2008년 6월 | Integrity capable block device의 metadata format이다. 예시는 T10-DIF-TYPE1-CRC다. Device가 송수신할 수 있는 T10 Protection Information type을 설명한다. Application integrity metadata를 저장할 수 있지만 T10 PI profile을 쓰지 않으면 nop, integrity metadata를 지원하지 않으면 none이다. |
| /sys/block/<disk>/integrity/protection_interval_bytes | 2015년 7월 | Integrity tuple 하나가 보호하는 data byte 수다. 일반적으로 device logical block size다. |
| /sys/block/<disk>/integrity/read_verify | 2008년 6월 | Integrity metadata를 보내는 device가 처리한 read request의 integrity를 block layer가 검증할지 나타낸다. |
| /sys/block/<disk>/integrity/tag_size | 2008년 6월 | protection_interval_bytes마다 사용할 수 있는 integrity tag 공간의 byte 수다. T10 PI로 format되고 application tag를 허용하면 block마다 opaque tag를 추가할 공간을 나타낸다. Application metadata 공간이 있더라도 T10 PI를 지원하지 않으면 0이다. |
| /sys/block/<disk>/integrity/write_generate | 2008년 6월 | Integrity metadata 수신을 지원하는 device로 보낼 write request의 checksum을 block layer가 자동 생성할지 나타낸다. |
Partition scan, alignment와 통계
178-218| 경로 | 작성 시점·담당 | 의미 |
|---|---|---|
| /sys/block/<disk>/partscan | 2024년 5월 · Christoph Hellwig | Disk의 partition scanning이 활성화됐으면 1, 아니면 0을 보고한다. Type은 32-bit unsigned integer지만 0과 1만 유효하다. |
| /sys/block/<disk>/<partition>/alignment_offset | 2009년 4월 · Martin K. Petersen | Partition 시작점이 disk의 자연스러운 alignment에서 몇 byte 떨어져 있는지 나타낸다. 4KB physical sector와 512-byte logical block처럼 physical block이 더 큰 device에서 필요하다. |
| /sys/block/<disk>/<partition>/discard_alignment | 2011년 5월 · Martin K. Petersen | Partition 시작점이 device 내부 discard allocation unit의 자연스러운 alignment에서 몇 byte 떨어져 있는지 나타낸다. |
| /sys/block/<disk>/<partition>/stat | 2008년 2월 · Jerome Marchand | 해당 partition의 I/O 통계를 표시하며 형식은 /sys/block/<disk>/stat과 같다. |
Entropy, chunk와 inline encryption
221-308| 경로 | 작성 시점·접근 | 의미 |
|---|---|---|
| /sys/block/<disk>/queue/add_random | 2010년 6월 · read/write | Disk의 entropy 기여를 끌 수 있다. 기본값은 1, 즉 켜짐이다. Contact는 [email protected]다. |
| /sys/block/<disk>/queue/chunk_sectors | 2016년 9월 · read-only | Disk type에 따라 의미가 다르다. RAID device(dm-raid)에서는 RAID volume stripe segment 크기를 512B sector 단위로 나타낸다. Host-aware 또는 host-managed zoned block device에서는 zone 크기를 512B sector 단위로 나타내며 마지막 zone은 더 작을 수 있다. 담당자는 Hannes Reinecke다. |
| /sys/block/<disk>/queue/crypto/ | 2022년 2월 | 이 subdirectory가 있으면 device가 inline encryption을 지원한다. 내부 file은 device의 inline encryption capability를 설명한다. 자세한 내용은 Documentation/block/inline-encryption.rst를 참조한다. |
| /sys/block/<disk>/queue/crypto/hw_wrapped_keys | 2025년 2월 · read-only | 이 file이 있으면 전용 hardware만 unwrap하고 사용할 수 있는 key blob, 즉 hardware-wrapped inline encryption key를 지원한다. |
| /sys/block/<disk>/queue/crypto/max_dun_bits | 2022년 2월 · read-only | Inline encryption request에서 device가 받는 data unit number의 최대 bit 길이다. |
| /sys/block/<disk>/queue/crypto/num_keyslots | 2022년 2월 · read-only | Inline encryption에 사용할 수 있는 device keyslot 수다. |
| /sys/block/<disk>/queue/crypto/raw_keys | 2025년 2월 · read-only | 이 file이 있으면 software에서 raw plaintext 형태로 관리하는 inline encryption key를 지원한다. |
crypto/modes/<mode>
Device가 inline encryption으로 지원하는 crypto mode, 즉 encryption/decryption algorithm마다 read-only file이 하나 생긴다. File의 16진수 값은 해당 mode가 지원하는 byte 단위 data unit size의 bitmask다. 현재 가능한 mode는 AES-256-XTS, AES-128-CBC-ESSIV, Adiantum이다.
예를 들어 AES-256-XTS에서 512-byte와 4096-byte data unit을 지원하면 /sys/block/<disk>/queue/crypto/modes/AES-256-XTS가 존재하고 0x1200을 담는다. 이 crypto 항목들의 contact는 [email protected]다.
DAX, discard와 independent access range
311-426| 경로 | 작성 시점·접근 | 의미 |
|---|---|---|
| /sys/block/<disk>/queue/dax | 2016년 6월 · read-only | CPU-addressable storage가 page cache를 우회하는 Direct Access(DAX)를 지원하면 1, 아니면 0이다. |
| /sys/block/<disk>/queue/discard_granularity | 2011년 5월 · read-only | Device가 보고한 내부 allocation unit 크기를 byte로 나타낸다. 보고하지 않으면 physical block size와 같게 설정한다. 0이면 discard를 지원하지 않는다. |
| /sys/block/<disk>/queue/discard_max_bytes | 2011년 5월 · read/write | discard_max_hw_bytes가 hardware limit인 반면 이 값은 software limit이다. 큰 discard에서 latency가 큰 device는 값을 낮춰 Linux가 더 작은 discard를 발행하게 하면 latency를 줄일 수 있다. |
| /sys/block/<disk>/queue/discard_max_hw_bytes | 2015년 7월 · read-only | Device driver가 설정한 한 번의 operation에서 discard할 수 있는 최대 byte 수다. Device에 보내는 request가 넘으면 안 된다. 0이면 discard를 지원하지 않는다. |
| /sys/block/<disk>/queue/discard_zeroes_data | 2011년 5월 · read-only | 항상 0을 반환한다. Discard의 특정 동작에 의존해서는 안 되며 이 file을 읽지 말아야 한다. |
| /sys/block/<disk>/queue/dma_alignment | 2022년 5월 | O_DIRECT raw block access와 driver별 passthrough mechanism에 쓸 user space address가 만족해야 하는 alignment를 보고한다. |
| /sys/block/<disk>/queue/fua | 2018년 5월 · read-only | Block driver가 write request의 FUA(Force Unit Access) flag를 지원하는지 나타낸다. FUA가 설정되면 write가 storage device의 volatile cache를 우회해야 한다. |
| /sys/block/<disk>/queue/hw_sector_size | 2008년 1월 · read-only | Device hardware sector size를 byte로 나타낸다. |
independent_access_ranges/
2021년 10월에 추가된 read-only subdirectory로 contact는 [email protected]다. 이 directory가 있으면 서로 다른 sector range를 대상으로 하는 request를 병렬 실행할 수 있다. 예를 들어 single-LUN multi-actuator HDD가 actuator의 sector range를 올바르게 알리면 이 directory가 생긴다.
Access range마다 directory 하나가 있고 sector file은 첫 sector, nr_sectors file은 그 지점부터의 전체 sector 수를 나타낸다. 두 값은 실제 block size와 관계없이 512B sector 단위를 쓴다. Range는 겹치지 않고 device capacity의 모든 sector를 포함한다. 시작 sector가 증가하는 순서로 번호를 붙이므로 range 0의 sector 값은 항상 0이다.
range 0range 1
01
sector = 0sector = range 1 시작 sector
02
nr_sectors = range 0 길이nr_sectors = range 1 길이
03
actuator 0가 독립 처리actuator 1이 독립 처리
원문의 ASCII directory tree를 두 독립 sector range와 각 attribute의 관계로 다시 그렸습니다.
Polling, timeout, accounting과 queue limit
429-540| 경로 | 작성 시점·접근 | 의미 |
|---|---|---|
| /sys/block/<disk>/queue/io_poll | 2015년 11월 · read/write | 읽으면 polling 활성화는 1, 비활성화는 0이다. 0을 쓰면 끄고 0이 아닌 값을 쓰면 켠다. |
| /sys/block/<disk>/queue/io_poll_delay | 2016년 11월 · read/write · deprecated | 과거 polling 종류를 제어했으나 현재 -1, classic polling으로 고정된다. CPU가 시간을 양보하지 않고 completion을 반복 확인한다. |
| /sys/block/<disk>/queue/io_timeout | 2018년 11월 · read/write | Request timeout을 millisecond로 나타낸다. 이 시간 안에 완료되지 않으면 driver timeout handler가 retry, fail 또는 device recovery strategy 시작을 결정한다. |
| /sys/block/<disk>/queue/iostats | 2009년 1월 · read/write | Disk I/O statistics accounting을 켜거나 끈다. |
| /sys/block/<disk>/queue/iostats_passthrough | 2024년 10월 · read/write | Passthrough command의 disk I/O statistics accounting을 켜거나 끈다. |
| /sys/block/<disk>/queue/logical_block_size | 2009년 5월 · read-only | Storage device가 address할 수 있는 가장 작은 단위이며 일반적으로 512 byte다. |
| /sys/block/<disk>/queue/max_active_zones | 2020년 7월 · read-only | Host-managed 또는 host-aware zoned device에서 EXPLICIT OPEN, IMPLICIT OPEN, CLOSED 상태 zone 수의 합을 제한한다. 0이면 제한이 없다. Host가 초과하면 driver는 BLK_STS_ZONE_ACTIVE_RESOURCE를 보고해야 하며 user space에는 EOVERFLOW errno로 보일 수 있다. |
| /sys/block/<disk>/queue/max_discard_segments | 2017년 2월 · read-only | Discard request의 최대 DMA scatter/gather entry 수다. |
| /sys/block/<disk>/queue/max_hw_sectors_kb | 2004년 9월 · read-only | 한 번의 data transfer에서 hardware가 지원하는 최대 kilobyte 수다. |
| /sys/block/<disk>/queue/max_integrity_segments | 2010년 9월 · read-only | Block layer core가 driver에 제출하는 integrity data 포함 DMA scatter/gather list의 최대 element 수다. |
| /sys/block/<disk>/queue/max_open_zones | 2020년 7월 · read-only | Zoned device에서 EXPLICIT OPEN 또는 IMPLICIT OPEN 상태 zone 수의 합을 제한한다. 0이면 제한이 없다. |
| /sys/block/<disk>/queue/max_sectors_kb | 2004년 9월 · read/write | Block layer가 filesystem request에 허용하는 최대 kilobyte 수다. Hardware 최대 크기 이하여야 하며 0을 쓰면 kernel 기본값을 사용한다. |
별도 담당자가 적힌 io_timeout과 zoned limit을 제외한 이 구간의 contact는 [email protected]이며, logical_block_size 담당자는 Martin K. Petersen다.
Segment, request, I/O 크기와 completion affinity
543-678| 경로 | 작성 시점·접근 | 의미 |
|---|---|---|
| /sys/block/<disk>/queue/max_segment_size | 2010년 3월 · read-only | DMA scatter/gather list element 하나의 최대 byte 크기다. |
| /sys/block/<disk>/queue/max_write_streams | 2024년 11월 · read-only | 지원하는 최대 write stream 수이며 미지원이면 0이다. 지원 시 유효한 값은 1부터 max_write_streams까지다. |
| /sys/block/<disk>/queue/write_stream_granularity | 2024년 11월 · read-only | Write stream의 byte 단위 granularity다. Device 내부 write amplification을 피하려면 함께 discard하거나 overwrite해야 하는 크기다. |
| /sys/block/<disk>/queue/max_segments | 2010년 3월 · read-only | 연결된 block driver에 제출하는 DMA scatter/gather list의 최대 element 수다. |
| /sys/block/<disk>/queue/minimum_io_size | 2009년 4월 · read-only | 성능 penalty 없이 처리할 수 있는 선호 최소 request 크기다. Disk에서는 흔히 physical block size, RAID에서는 stripe chunk size다. IOPS가 중요한 workload는 올바르게 정렬된 minimum_io_size 배수를 선호한다. |
| /sys/block/<disk>/queue/nomerges | 2010년 1월 · read/write | 0은 모든 merge 시도, 1은 복잡한 검사만 끄고 직전 request와의 단순 one-shot merge 유지, 2는 모든 merge 시도를 끈다. 알려진 random I/O에서는 실패할 merge 검사의 CPU cycle을 줄일 수 있다. |
| /sys/block/<disk>/queue/nr_requests | 2003년 7월 · read/write | Block layer에서 할당할 수 있는 request 수다. 값은 blk_mq_tags instance 하나의 수만 나타내며 전체 device 수는 hardware queue 수, elevator 활성화, tag 공유 여부에 따라 달라진다. |
| /sys/block/<disk>/queue/nr_zones | 2018년 11월 · read-only | Host-aware 또는 host-managed zoned block device의 전체 zone 수다. 일반 block device에서는 항상 0이다. |
| /sys/block/<disk>/queue/optimal_io_size | 2009년 4월 · read-only | 지속 I/O에 선호하는 단위다. Disk drive는 드물게 보고하며 RAID에서는 보통 stripe width 또는 내부 track size다. Sustained throughput workload는 올바르게 정렬된 배수를 선호한다. 보고하지 않으면 0이다. |
| /sys/block/<disk>/queue/physical_block_size | 2009년 5월 · read-only | Physical storage가 원자적으로 쓸 수 있는 가장 작은 단위다. 보통 logical block과 같지만 4KB sector를 512-byte logical block으로 노출하는 SATA drive처럼 더 클 수 있다. Stacked device에서는 component의 최댓값이다. |
| /sys/block/<disk>/queue/read_ahead_kb | 2004년 5월 · read/write | 이 block device의 filesystem에 적용하는 최대 readahead kilobyte 수다. MADV_HUGEPAGE는 hugepage 크기를 granularity로 사용하므로 이 값을 넘을 수 있다. |
| /sys/block/<disk>/queue/rotational | 2009년 1월 · read/write | Device가 rotational인지 non-rotational인지 나타내는 데 사용한다. |
| /sys/block/<disk>/queue/rq_affinity | 2008년 9월 · read/write | 1이면 completion을 request를 제출한 CPU group으로 이동해 cache 효과로 CPU cycle을 줄일 수 있다. 2이면 group aggregation을 우회하고 요청 CPU에서 completion을 실행해 처리 분산을 극대화한다. |
minimum_io_size, optimal_io_size, physical_block_size의 담당자는 Martin K. Petersen이며, 나머지 항목의 contact는 [email protected]다. nr_zones 담당자는 Damien Le Moal이다.
Scheduler, stable write, throttling과 cache
681-746| 경로 | 작성 시점·접근 | 의미 |
|---|---|---|
| /sys/block/<disk>/queue/scheduler | 2004년 10월 · read/write | 현재 및 사용 가능한 I/O scheduler를 표시하며 활성 scheduler는 []로 감싼다. Scheduler 이름을 쓰면 전환하고 module이 없으면 load를 시도한다. |
| /sys/block/<disk>/queue/stable_writes | 2020년 9월 · read/write | Write request가 memory를 여러 번 읽어 매번 같은 data를 봐야 하면 1이다. 예를 들어 checksum 계산 뒤 실제 write를 수행할 때 kernel은 page writeback이 끝나기 전 수정을 막는다. 제약이 없으면 0이며 test 목적으로 쓸 수 있다. |
| /sys/block/<disk>/queue/virt_boundary_mask | 2021년 4월 · read-only | I/O segment memory alignment mask다. 이전 segment 끝 또는 현재 segment 시작 memory address가 virt_boundary_mask + 1 byte에 정렬되지 않은 지점에서 request를 분할한다. |
| /sys/block/<disk>/queue/wbt_lat_usec | 2016년 11월 · read/write | Writeback throttling 등록 device의 목표 최소 read latency다. 일정 window에서 초과하면 write를 줄이기 시작한다. 0은 기능을 끄고 -1은 기본값으로 reset한다. Window는 curr_win_nsec를 참조한다. |
| /sys/block/<disk>/queue/write_cache | 2016년 4월 · read/write | Write-back cache 활성화 시 write back, 아니면 write through를 반환한다. 쓰기는 kernel이 보는 상태만 바꾸고 device 상태는 바꾸지 않는다. write back에서 write through로 바꾸면 kernel cache flush도 사라지므로 안전하지 않을 수 있다. |
이 구간의 contact는 [email protected]다.
Write same, zeroing, zoned storage와 통계
749-881| 경로 | 작성 시점·접근 | 의미 |
|---|---|---|
| /sys/block/<disk>/queue/write_same_max_bytes | 2012년 1월 · read-only | Data block 하나를 연속 block range에 쓰는 write same command의 최대 byte 수다. Disk 영역 삭제나 RAID drive 초기화에 쓸 수 있으며 0이면 미지원이다. |
| /sys/block/<disk>/queue/write_zeroes_max_bytes | 2016년 11월 · read-only | Payload 없이 연속 block range를 zero로 만드는 write zeroes command의 최대 byte 수다. 0이면 미지원이다. |
| /sys/block/<disk>/queue/write_zeroes_unmap_max_hw_bytes | 2025년 1월 · read-only | 각 block에 zero를 물리적으로 쓰는 비용을 피할 수 있는 hardware capability다. write_zeroes_max_bytes와 같으면 이후 read가 zero를 반환함을 보장하면서 media write를 기회적으로 피한다. Misalignment나 내부 allocation unit보다 작은 range에서는 실제 zero write로 fallback할 수 있다. 0이면 각 logical block을 실제로 써야 할 수 있다. |
| /sys/block/<disk>/queue/write_zeroes_unmap_max_bytes | 2025년 1월 · read/write | Hardware limit에 대응하는 software limit이다. Best-effort라 속도는 보장되지 않는다. 0은 기능을 끄며 그 밖에는 write_zeroes_unmap_max_hw_bytes와 같아야 한다. |
| /sys/block/<disk>/queue/zone_append_max_bytes | 2020년 5월 · read-only | REQ_OP_ZONE_APPEND로 zoned device의 sequential zone에 쓸 수 있는 최대 byte 수다. 일반 block device에서는 항상 0이다. |
| /sys/block/<disk>/queue/zone_write_granularity | 2021년 1월 · read-only | Host-managed 또는 host-aware device의 sequential zone write가 만족해야 하는 byte 단위 alignment다. 일반 device에서는 0이다. |
| /sys/block/<disk>/queue/zoned | 2016년 9월 · read-only | 일반 device는 none, zoned device는 host-aware 또는 host-managed를 반환한다. ZBC와 ZAC standard가 두 model을 정의한다. drive-managed model은 zone command를 지원하지 않아 일반 device로 취급하고 none을 반환한다. |
| /sys/block/<disk>/hidden | 2023년 3월 · read-only | 숨겨진 block device다. Event를 만들지 않고 user space나 blkdev_get*으로 열 수 없다. Multipath device의 underlying component에 사용한다. |
write_same_max_bytes 담당자는 Martin K. Petersen, write_zeroes_max_bytes는 Chaitanya Kulkarni, 두 unmap 항목은 Zhang Yi, zoned는 Damien Le Moal이며 나머지는 [email protected]다.
/sys/block/<disk>/stat
2008년 2월에 추가됐고 담당자는 Jerome Marchand다. Disk I/O 통계를 표시한다. 원문 설명은 11개 field라고 적지만 이어지는 표에는 다음 17개 field가 정의돼 있다.
| 번호 | 통계 |
|---|---|
| 1 | 성공적으로 완료된 read |
| 2 | merge된 read |
| 3 | 읽은 sector |
| 4 | read에 사용한 시간(ms) |
| 5 | 완료된 write |
| 6 | merge된 write |
| 7 | 쓴 sector |
| 8 | write에 사용한 시간(ms) |
| 9 | 현재 진행 중인 I/O |
| 10 | I/O 수행에 사용한 시간(ms) |
| 11 | I/O 수행에 사용한 가중 시간(ms) |
| 12 | 완료된 discard |
| 13 | merge된 discard |
| 14 | discard한 sector |
| 15 | discard에 사용한 시간(ms) |
| 16 | 완료된 flush request |
| 17 | flush에 사용한 시간(ms) |
자세한 내용은 Documentation/admin-guide/iostats.rst를 참조한다.
Disk 기본 속성과 atomic write
sysfs-block:1-103Physical·logical alignment, atomic write 단위와 경계, disk sequence, in-flight read/write 수를 다룹니다.