blob: 47f614c47f567bc48de4d4ea98394a489abefcfc [file] [log] [blame]
Darrick J. Wong489fcb92018-07-29 15:36:00 -04001.. SPDX-License-Identifier: GPL-2.0
Dave Kleikampfc513a32006-10-11 01:21:25 -07002
Darrick J. Wong489fcb92018-07-29 15:36:00 -04003========================
4General Information
5========================
Dave Kleikampfc513a32006-10-11 01:21:25 -07006
Masanari Iidac9f3f2d2013-07-18 01:29:12 +09007Ext4 is an advanced level of the ext3 filesystem which incorporates
Diego Calleja22359f52008-10-17 09:15:14 -04008scalability and reliability enhancements for supporting large filesystems
9(64 bit) in keeping with increasing disk capacities and state-of-the-art
10feature requirements.
Dave Kleikampfc513a32006-10-11 01:21:25 -070011
Diego Calleja22359f52008-10-17 09:15:14 -040012Mailing list: linux-ext4@vger.kernel.org
13Web site: http://ext4.wiki.kernel.org
Dave Kleikampfc513a32006-10-11 01:21:25 -070014
15
Darrick J. Wong489fcb92018-07-29 15:36:00 -040016Quick usage instructions
17========================
Dave Kleikampfc513a32006-10-11 01:21:25 -070018
Diego Calleja22359f52008-10-17 09:15:14 -040019Note: More extensive information for getting started with ext4 can be
Darrick J. Wong489fcb92018-07-29 15:36:00 -040020found at the ext4 wiki site at the URL:
21http://ext4.wiki.kernel.org/index.php/Ext4_Howto
Diego Calleja22359f52008-10-17 09:15:14 -040022
Jose R. Santos93e32702008-07-11 19:27:31 -040023 - Compile and install the latest version of e2fsprogs (as of this
Diego Calleja22359f52008-10-17 09:15:14 -040024 writing version 1.41.3) from:
Jose R. Santos93e32702008-07-11 19:27:31 -040025
26 http://sourceforge.net/project/showfiles.php?group_id=2406
Darrick J. Wong489fcb92018-07-29 15:36:00 -040027
Jose R. Santos93e32702008-07-11 19:27:31 -040028 or
29
SeongJae Park3bdadc862017-03-27 22:05:34 +090030 https://www.kernel.org/pub/linux/kernel/people/tytso/e2fsprogs/
Dave Kleikampfc513a32006-10-11 01:21:25 -070031
Jose R. Santos93e32702008-07-11 19:27:31 -040032 or grab the latest git repository from:
Dave Kleikampfc513a32006-10-11 01:21:25 -070033
Jose R. Santos93e32702008-07-11 19:27:31 -040034 git://git.kernel.org/pub/scm/fs/ext2/e2fsprogs.git
Dave Kleikampfc513a32006-10-11 01:21:25 -070035
Theodore Ts'o45373982008-07-27 19:59:21 -040036 - Note that it is highly important to install the mke2fs.conf file
37 that comes with the e2fsprogs 1.41.x sources in /etc/mke2fs.conf. If
38 you have edited the /etc/mke2fs.conf file installed on your system,
39 you will need to merge your changes with the version from e2fsprogs
40 1.41.x.
41
Darrick J. Wong489fcb92018-07-29 15:36:00 -040042 - Create a new filesystem using the ext4 filesystem type:::
Dave Kleikampfc513a32006-10-11 01:21:25 -070043
Darrick J. Wong489fcb92018-07-29 15:36:00 -040044 # mke2fs -t ext4 /dev/hda1
Dave Kleikampfc513a32006-10-11 01:21:25 -070045
Darrick J. Wong489fcb92018-07-29 15:36:00 -040046 Or to configure an existing ext3 filesystem to support extents:::
Dave Kleikampfc513a32006-10-11 01:21:25 -070047
Diego Calleja22359f52008-10-17 09:15:14 -040048 # tune2fs -O extents /dev/hda1
Jose R. Santos93e32702008-07-11 19:27:31 -040049
50 If the filesystem was created with 128 byte inodes, it can be
Darrick J. Wong489fcb92018-07-29 15:36:00 -040051 converted to use 256 byte for greater efficiency via:::
Jose R. Santos93e32702008-07-11 19:27:31 -040052
53 # tune2fs -I 256 /dev/hda1
54
Theodore Ts'o03010a32008-10-10 20:02:48 -040055 (Note: we currently do not have tools to convert an ext4
Jose R. Santos93e32702008-07-11 19:27:31 -040056 filesystem back to ext3; so please do not do try this on production
57 filesystems.)
58
Darrick J. Wong489fcb92018-07-29 15:36:00 -040059 - Mounting:::
Jose R. Santos93e32702008-07-11 19:27:31 -040060
Theodore Ts'o03010a32008-10-10 20:02:48 -040061 # mount -t ext4 /dev/hda1 /wherever
Dave Kleikampfc513a32006-10-11 01:21:25 -070062
Theodore Ts'o8e1a4852009-01-06 14:53:06 -050063 - When comparing performance with other filesystems, it's always
64 important to try multiple workloads; very often a subtle change in a
65 workload parameter can completely change the ranking of which
66 filesystems do well compared to others. When comparing versus ext3,
67 note that ext4 enables write barriers by default, while ext3 does
68 not enable write barriers by default. So it is useful to use
69 explicitly specify whether barriers are enabled or not when via the
70 '-o barriers=[0|1]' mount option for both ext3 and ext4 filesystems
71 for a fair comparison. When tuning ext3 for best benchmark numbers,
72 it is often worthwhile to try changing the data journaling mode; '-o
Lukas Czernerad434012011-06-07 12:27:05 +020073 data=writeback' can be faster for some workloads. (Note however that
74 running mounted with data=writeback can potentially leave stale data
75 exposed in recently written files in case of an unclean shutdown,
76 which could be a security exposure in some situations.) Configuring
77 the filesystem with a large journal can also be helpful for
78 metadata-intensive workloads.
Dave Kleikampfc513a32006-10-11 01:21:25 -070079
Darrick J. Wong489fcb92018-07-29 15:36:00 -040080Features
81========
Dave Kleikampfc513a32006-10-11 01:21:25 -070082
Darrick J. Wong489fcb92018-07-29 15:36:00 -040083Currently Available
84-------------------
Dave Kleikampfc513a32006-10-11 01:21:25 -070085
Jose R. Santos93e32702008-07-11 19:27:31 -040086* ability to use filesystems > 16TB (e2fsprogs support not available yet)
Dave Kleikampfc513a32006-10-11 01:21:25 -070087* extent format reduces metadata overhead (RAM, IO for access, transactions)
88* extent format more robust in face of on-disk corruption due to magics,
Theodore Ts'o8e1a4852009-01-06 14:53:06 -050089* internal redundancy in tree
Mingming Cao49f14872008-07-11 19:27:31 -040090* improved file allocation (multi-block alloc)
Theodore Ts'o722bde62009-02-23 00:51:57 -050091* lift 32000 subdirectory limit imposed by i_links_count[1]
Jose R. Santos93e32702008-07-11 19:27:31 -040092* nsec timestamps for mtime, atime, ctime, create time
93* inode version field on disk (NFSv4, Lustre)
94* reduced e2fsck time via uninit_bg feature
95* journal checksumming for robustness, performance
96* persistent file preallocation (e.g for streaming media, databases)
97* ability to pack bitmaps and inode tables into larger virtual groups via the
98 flex_bg feature
99* large file support
Pavel Machek98bfa342017-09-16 13:48:37 +0200100* inode allocation using large virtual block groups via flex_bg
Mingming Cao49f14872008-07-11 19:27:31 -0400101* delayed allocation
102* large block (up to pagesize) support
Pavel Machek98bfa342017-09-16 13:48:37 +0200103* efficient new ordered mode in JBD2 and ext4 (avoid using buffer head to force
Mingming Cao49f14872008-07-11 19:27:31 -0400104 the ordering)
Dave Kleikampfc513a32006-10-11 01:21:25 -0700105
Theodore Ts'o722bde62009-02-23 00:51:57 -0500106[1] Filesystems with a block size of 1k may see a limit imposed by the
107directory hash tree having a maximum depth of two.
108
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400109Candidate Features for Future Inclusion
110---------------------------------------
Dave Kleikampfc513a32006-10-11 01:21:25 -0700111
Pavel Machek98bfa342017-09-16 13:48:37 +0200112* online defrag (patches available but not well tested)
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300113* reduced mke2fs time via lazy itable initialization in conjunction with
Jose R. Santos93e32702008-07-11 19:27:31 -0400114 the uninit_bg feature (capability to do this is available in e2fsprogs
115 but a kernel thread to do lazy zeroing of unused inode table blocks
116 after filesystem is first mounted is required for safety)
Dave Kleikampfc513a32006-10-11 01:21:25 -0700117
Jose R. Santos93e32702008-07-11 19:27:31 -0400118There are several others under discussion, whether they all make it in is
119partly a function of how much time everyone has to work on them. Features like
120metadata checksumming have been discussed and planned for a bit but no patches
121exist yet so I'm not sure they're in the near-term roadmap.
Dave Kleikampfc513a32006-10-11 01:21:25 -0700122
Jose R. Santos93e32702008-07-11 19:27:31 -0400123The big performance win will come with mballoc, delalloc and flex_bg
124grouping of bitmaps and inode tables. Some test results available here:
Dave Kleikampfc513a32006-10-11 01:21:25 -0700125
Diego Calleja22359f52008-10-17 09:15:14 -0400126 - http://www.bullopensource.org/ext4/20080818-ffsb/ffsb-write-2.6.27-rc1.html
127 - http://www.bullopensource.org/ext4/20080818-ffsb/ffsb-readwrite-2.6.27-rc1.html
Dave Kleikampfc513a32006-10-11 01:21:25 -0700128
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400129Options
130=======
Dave Kleikampfc513a32006-10-11 01:21:25 -0700131
132When mounting an ext4 filesystem, the following option are accepted:
133(*) == default
134
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400135======================= =======================================================
136Mount Option Description
137======================= =======================================================
Theodore Ts'o8e1a4852009-01-06 14:53:06 -0500138ro Mount filesystem read only. Note that ext4 will
139 replay the journal (and thus write to the
140 partition) even when mounted "read only". The
141 mount options "ro,noload" can be used to prevent
142 writes to the filesystem.
143
Linus Torvaldsd4da6c92009-11-02 10:15:27 -0800144journal_checksum Enable checksumming of the journal transactions.
145 This will allow the recovery code in e2fsck and the
146 kernel to detect corruption in the kernel. It is a
147 compatible change and will be ignored by older kernels.
148
Girish Shilamkar818d2762008-01-28 23:58:27 -0500149journal_async_commit Commit block can be written to disk without waiting
150 for descriptor blocks. If enabled older kernels cannot
Linus Torvaldsd4da6c92009-11-02 10:15:27 -0800151 mount the device. This will enable 'journal_checksum'
152 internally.
Girish Shilamkar818d2762008-01-28 23:58:27 -0500153
Eric Sandeenad4eec62013-08-28 19:05:07 -0400154journal_path=path
Dave Kleikampfc513a32006-10-11 01:21:25 -0700155journal_dev=devnum When the external journal device's major/minor numbers
Eric Sandeenad4eec62013-08-28 19:05:07 -0400156 have changed, these options allow the user to specify
Dave Kleikampfc513a32006-10-11 01:21:25 -0700157 the new journal location. The journal device is
Eric Sandeenad4eec62013-08-28 19:05:07 -0400158 identified through either its new major/minor numbers
159 encoded in devnum, or via a path to the device.
Dave Kleikampfc513a32006-10-11 01:21:25 -0700160
Eric Sandeene3bb52a2009-11-19 14:28:50 -0500161norecovery Don't load the journal on mounting. Note that
162noload if the filesystem was not unmounted cleanly,
Theodore Ts'o8e1a4852009-01-06 14:53:06 -0500163 skipping the journal replay will lead to the
164 filesystem containing inconsistencies that can
165 lead to any number of problems.
Dave Kleikampfc513a32006-10-11 01:21:25 -0700166
167data=journal All data are committed into the journal prior to being
Theodore Ts'o56889782011-09-03 18:22:38 -0400168 written into the main file system. Enabling
169 this mode will disable delayed allocation and
170 O_DIRECT support.
Dave Kleikampfc513a32006-10-11 01:21:25 -0700171
172data=ordered (*) All data are forced directly out to the main file
173 system prior to its metadata being committed to the
174 journal.
175
176data=writeback Data ordering is not preserved, data may be written
177 into the main file system after its metadata has been
178 committed to the journal.
179
180commit=nrsec (*) Ext4 can be told to sync all its data and metadata
181 every 'nrsec' seconds. The default value is 5 seconds.
182 This means that if you lose your power, you will lose
183 as much as the latest 5 seconds of work (your
184 filesystem will not be damaged though, thanks to the
185 journaling). This default value (or any low value)
186 will hurt performance, but it's good for data-safety.
187 Setting it to 0 will have the same effect as leaving
188 it at the default (5 seconds).
189 Setting it to very large values will improve
190 performance.
191
Eric Sandeen571640c2008-05-26 12:29:46 -0400192barrier=<0|1(*)> This enables/disables the use of write barriers in
Theodore Ts'o06705bf2009-03-28 10:59:57 -0400193barrier(*) the jbd code. barrier=0 disables, barrier=1 enables.
194nobarrier This also requires an IO stack which can support
Eric Sandeen571640c2008-05-26 12:29:46 -0400195 barriers, and if jbd gets an error on a barrier
196 write, it will disable again with a warning.
197 Write barriers enforce proper on-disk ordering
198 of journal commits, making volatile disk write caches
199 safe to use, at some performance penalty. If
200 your disks are battery-backed in one way or another,
201 disabling barriers may safely improve performance.
Theodore Ts'o06705bf2009-03-28 10:59:57 -0400202 The mount options "barrier" and "nobarrier" can
203 also be used to enable or disable barriers, for
204 consistency with other ext4 mount options.
Dave Kleikampfc513a32006-10-11 01:21:25 -0700205
Fang Wenqi6d3b82f2009-12-24 17:51:42 -0500206inode_readahead_blks=n This tuning parameter controls the maximum
Theodore Ts'o240799c2008-10-09 23:53:47 -0400207 number of inode table blocks that ext4's inode
208 table readahead algorithm will pre-read into
209 the buffer cache. The default value is 32 blocks.
210
Tao Ma939da102012-12-10 16:30:43 -0500211nouser_xattr Disables Extended User Attributes. See the
Adam Borowski91581e42017-12-20 14:58:52 +0100212 attr(5) manual page for more information about
213 extended attributes.
Dave Kleikampfc513a32006-10-11 01:21:25 -0700214
215noacl This option disables POSIX Access Control List
Theodore Ts'oaf909a52011-10-08 14:01:08 -0400216 support. If ACL support is enabled in the kernel
217 configuration (CONFIG_EXT4_FS_POSIX_ACL), ACL is
218 enabled by default on mount. See the acl(5) manual
Adam Borowski91581e42017-12-20 14:58:52 +0100219 page for more information about acl.
Dave Kleikampfc513a32006-10-11 01:21:25 -0700220
Dave Kleikampfc513a32006-10-11 01:21:25 -0700221bsddf (*) Make 'df' act like BSD.
222minixdf Make 'df' act like Minix.
223
Dave Kleikampfc513a32006-10-11 01:21:25 -0700224debug Extra debugging information is sent to syslog.
225
Theodore Ts'o8a8a2052009-06-13 10:08:59 -0400226abort Simulate the effects of calling ext4_abort() for
227 debugging purposes. This is normally used while
228 remounting a filesystem which is already mounted.
229
Theodore Ts'o8e1a4852009-01-06 14:53:06 -0500230errors=remount-ro Remount the filesystem read-only on an error.
Dave Kleikampfc513a32006-10-11 01:21:25 -0700231errors=continue Keep going on a filesystem error.
232errors=panic Panic and halt the machine if an error occurs.
Theodore Ts'o8e1a4852009-01-06 14:53:06 -0500233 (These mount options override the errors behavior
234 specified in the superblock, which can be configured
235 using tune2fs)
Dave Kleikampfc513a32006-10-11 01:21:25 -0700236
Hidehiro Kawai5bf56832008-10-10 22:12:43 -0400237data_err=ignore(*) Just print an error message if an error occurs
238 in a file data buffer in ordered mode.
239data_err=abort Abort the journal if an error occurs in a file
240 data buffer in ordered mode.
241
Ernesto A. Fernández9f037242018-01-11 13:43:33 -0500242grpid New objects have the group ID of their parent.
Dave Kleikampfc513a32006-10-11 01:21:25 -0700243bsdgroups
244
245nogrpid (*) New objects have the group ID of their creator.
246sysvgroups
247
248resgid=n The group ID which may use the reserved blocks.
249
250resuid=n The user ID which may use the reserved blocks.
251
252sb=n Use alternate superblock at this location.
253
Jan Kara13588702009-09-18 12:22:29 -0400254quota These options are ignored by the filesystem. They
255noquota are used only by quota tools to recognize volumes
256grpquota where quota should be turned on. See documentation
257usrquota in the quota-tools package for more details
258 (http://sourceforge.net/projects/linuxquota).
259
260jqfmt=<quota type> These options tell filesystem details about quota
261usrjquota=<file> so that quota information can be properly updated
262grpjquota=<file> during journal replay. They replace the above
263 quota options. See documentation in the quota-tools
264 package for more details
265 (http://sourceforge.net/projects/linuxquota).
Dave Kleikampfc513a32006-10-11 01:21:25 -0700266
Alex Tomasc9de5602008-01-29 00:19:52 -0500267stripe=n Number of filesystem blocks that mballoc will try
268 to use for allocation size and alignment. For RAID5/6
269 systems this should be the number of data
270 disks * RAID chunk size in file system blocks.
Jan Kara83653882009-09-29 15:59:34 -0400271
272delalloc (*) Defer block allocation until just before ext4
273 writes out the block(s) in question. This
274 allows ext4 to better allocation decisions
275 more efficiently.
276nodelalloc Disable delayed allocation. Blocks are allocated
277 when the data is copied from userspace to the
278 page cache, either via the write(2) system call
279 or when an mmap'ed page which was previously
280 unallocated is written for the first time.
Theodore Ts'o240799c2008-10-09 23:53:47 -0400281
Theodore Ts'o30773842009-01-03 20:27:38 -0500282max_batch_time=usec Maximum amount of time ext4 should wait for
283 additional filesystem operations to be batch
284 together with a synchronous write operation.
285 Since a synchronous write operation is going to
286 force a commit and then a wait for the I/O
287 complete, it doesn't cost much, and can be a
288 huge throughput win, we wait for a small amount
289 of time to see if any other transactions can
290 piggyback on the synchronous write. The
291 algorithm used is designed to automatically tune
292 for the speed of the disk, by measuring the
293 amount of time (on average) that it takes to
294 finish committing a transaction. Call this time
295 the "commit time". If the time that the
Matt LaPlante19f59462009-04-27 15:06:31 +0200296 transaction has been running is less than the
Theodore Ts'o30773842009-01-03 20:27:38 -0500297 commit time, ext4 will try sleeping for the
298 commit time to see if other operations will join
299 the transaction. The commit time is capped by
300 the max_batch_time, which defaults to 15000us
301 (15ms). This optimization can be turned off
302 entirely by setting max_batch_time to 0.
303
304min_batch_time=usec This parameter sets the commit time (as
305 described above) to be at least min_batch_time.
306 It defaults to zero microseconds. Increasing
307 this parameter may improve the throughput of
308 multi-threaded, synchronous workloads on very
309 fast disks, at the cost of increasing latency.
310
Theodore Ts'ob3881f72009-01-05 22:46:26 -0500311journal_ioprio=prio The I/O priority (from 0 to 7, where 0 is the
Masanari Iida40e47122012-03-04 23:16:11 +0900312 highest priority) which should be used for I/O
Theodore Ts'ob3881f72009-01-05 22:46:26 -0500313 operations submitted by kjournald2 during a
314 commit operation. This defaults to 3, which is
315 a slightly higher priority than the default I/O
316 priority.
317
Theodore Ts'o06705bf2009-03-28 10:59:57 -0400318auto_da_alloc(*) Many broken applications don't use fsync() when
319noauto_da_alloc replacing existing files via patterns such as
320 fd = open("foo.new")/write(fd,..)/close(fd)/
321 rename("foo.new", "foo"), or worse yet,
322 fd = open("foo", O_TRUNC)/write(fd,..)/close(fd).
323 If auto_da_alloc is enabled, ext4 will detect
324 the replace-via-rename and replace-via-truncate
325 patterns and force that any delayed allocation
326 blocks are allocated such that at the next
327 journal commit, in the default data=ordered
328 mode, the data blocks of the new file are forced
329 to disk before the rename() operation is
Matt LaPlante19f59462009-04-27 15:06:31 +0200330 committed. This provides roughly the same level
Theodore Ts'o06705bf2009-03-28 10:59:57 -0400331 of guarantees as ext3, and avoids the
332 "zero-length" problem that can happen when a
333 system crashes before the delayed allocation
334 blocks are forced to disk.
335
Lukas Czernerbfff6872010-10-27 21:30:05 -0400336noinit_itable Do not initialize any uninitialized inode table
337 blocks in the background. This feature may be
338 used by installation CD's so that the install
339 process can complete as quickly as possible; the
340 inode table initialization process would then be
341 deferred until the next time the file system
342 is unmounted.
343
344init_itable=n The lazy itable init code will wait n times the
345 number of milliseconds it took to zero out the
346 previous block group's inode table. This
Masanari Iida40e47122012-03-04 23:16:11 +0900347 minimizes the impact on the system performance
Lukas Czernerbfff6872010-10-27 21:30:05 -0400348 while file system's inode table is being initialized.
349
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500350discard Controls whether ext4 should issue discard/TRIM
Eric Sandeen5328e632009-11-19 14:25:42 -0500351nodiscard(*) commands to the underlying block device when
352 blocks are freed. This is useful for SSD devices
353 and sparse/thinly-provisioned LUNs, but it is off
354 by default until sufficient testing has been done.
355
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500356nouid32 Disables 32-bit UIDs and GIDs. This is for
357 interoperability with older kernels which only
358 store and expect 16-bit values.
359
Fabian Frederick9e645ab72016-12-03 16:44:45 -0500360block_validity(*) These options enable or disable the in-kernel
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500361noblock_validity facility for tracking filesystem metadata blocks
Fabian Frederick9e645ab72016-12-03 16:44:45 -0500362 within internal data structures. This allows multi-
363 block allocator and other routines to notice
364 bugs or corrupted allocation bitmaps which cause
365 blocks to be allocated which overlap with
366 filesystem metadata blocks.
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500367
368dioread_lock Controls whether or not ext4 should use the DIO read
369dioread_nolock locking. If the dioread_nolock option is specified
370 ext4 will allocate uninitialized extent before buffer
371 write and convert the extent to initialized after IO
372 completes. This approach allows ext4 code to avoid
373 using inode mutex, which improves scalability on high
Lukas Czernerad434012011-06-07 12:27:05 +0200374 speed storages. However this does not work with
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500375 data journaling and dioread_nolock option will be
376 ignored with kernel warning. Note that dioread_nolock
377 code path is only used for extent-based files.
378 Because of the restrictions this options comprises
379 it is off by default (e.g. dioread_lock).
380
Theodore Ts'odf981d02012-08-17 09:48:17 -0400381max_dir_size_kb=n This limits the size of directories so that any
382 attempt to expand them beyond the specified
383 limit in kilobytes will cause an ENOSPC error.
384 This is useful in memory constrained
385 environments, where a very large directory can
386 cause severe performance problems or even
387 provoke the Out Of Memory killer. (For example,
388 if there is only 512mb memory available, a 176mb
389 directory may seriously cramp the system's style.)
390
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500391i_version Enable 64-bit inode version support. This option is
392 off by default.
393
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800394dax Use direct access (no page cache). See
395 Documentation/filesystems/dax.txt. Note that
396 this option is incompatible with data=journal.
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400397======================= =======================================================
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800398
Dave Kleikampfc513a32006-10-11 01:21:25 -0700399Data Mode
Jose R. Santos93e32702008-07-11 19:27:31 -0400400=========
Dave Kleikampfc513a32006-10-11 01:21:25 -0700401There are 3 different data modes:
402
403* writeback mode
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400404
405 In data=writeback mode, ext4 does not journal data at all. This mode provides
406 a similar level of journaling as that of XFS, JFS, and ReiserFS in its default
407 mode - metadata journaling. A crash+recovery can cause incorrect data to
408 appear in files which were written shortly before the crash. This mode will
409 typically provide the best ext4 performance.
Dave Kleikampfc513a32006-10-11 01:21:25 -0700410
411* ordered mode
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400412
413 In data=ordered mode, ext4 only officially journals metadata, but it logically
414 groups metadata information related to data changes with the data blocks into
415 a single unit called a transaction. When it's time to write the new metadata
416 out to disk, the associated data blocks are written first. In general, this
417 mode performs slightly slower than writeback but significantly faster than
418 journal mode.
Dave Kleikampfc513a32006-10-11 01:21:25 -0700419
420* journal mode
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400421
422 data=journal mode provides full data and metadata journaling. All new data is
423 written to the journal first, and then to its final location. In the event of
424 a crash, the journal can be replayed, bringing both data and metadata into a
425 consistent state. This mode is the slowest except when data needs to be read
426 from and written to disk at the same time where it outperforms all others
427 modes. Enabling this mode will disable delayed allocation and O_DIRECT
428 support.
Dave Kleikampfc513a32006-10-11 01:21:25 -0700429
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500430/proc entries
431=============
432
433Information about mounted ext4 file systems can be found in
434/proc/fs/ext4. Each mounted filesystem will have a directory in
435/proc/fs/ext4 based on its device name (i.e., /proc/fs/ext4/hdc or
436/proc/fs/ext4/dm-0). The files in each per-device directory are shown
437in table below.
438
439Files in /proc/fs/ext4/<devname>
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400440
441================ =======
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500442 File Content
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400443================ =======
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500444 mb_groups details of multiblock allocator buddy cache of free blocks
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400445================ =======
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500446
447/sys entries
448============
449
450Information about mounted ext4 file systems can be found in
451/sys/fs/ext4. Each mounted filesystem will have a directory in
452/sys/fs/ext4 based on its device name (i.e., /sys/fs/ext4/hdc or
453/sys/fs/ext4/dm-0). The files in each per-device directory are shown
454in table below.
455
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400456Files in /sys/fs/ext4/<devname>:
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500457
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400458(see also Documentation/ABI/testing/sysfs-fs-ext4)
459
460============================= =================================================
461File Content
462============================= =================================================
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500463 delayed_allocation_blocks This file is read-only and shows the number of
464 blocks that are dirty in the page cache, but
465 which do not have their location in the
466 filesystem allocated yet.
467
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400468inode_goal Tuning parameter which (if non-zero) controls
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500469 the goal inode used by the inode allocator in
470 preference to all other allocation heuristics.
471 This is intended for debugging use only, and
472 should be 0 on production systems.
473
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400474inode_readahead_blks Tuning parameter which controls the maximum
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500475 number of inode table blocks that ext4's inode
476 table readahead algorithm will pre-read into
477 the buffer cache
478
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400479lifetime_write_kbytes This file is read-only and shows the number of
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500480 kilobytes of data that have been written to this
481 filesystem since it was created.
482
483 max_writeback_mb_bump The maximum number of megabytes the writeback
484 code will try to write out before move on to
485 another inode.
486
487 mb_group_prealloc The multiblock allocator will round up allocation
488 requests to a multiple of this tuning parameter if
489 the stripe size is not set in the ext4 superblock
490
491 mb_max_to_scan The maximum number of extents the multiblock
492 allocator will search to find the best extent
493
494 mb_min_to_scan The minimum number of extents the multiblock
495 allocator will search to find the best extent
496
497 mb_order2_req Tuning parameter which controls the minimum size
498 for requests (as a power of 2) where the buddy
499 cache is used
500
501 mb_stats Controls whether the multiblock allocator should
502 collect statistics, which are shown during the
503 unmount. 1 means to collect statistics, 0 means
504 not to collect statistics
505
506 mb_stream_req Files which have fewer blocks than this tunable
507 parameter will have their blocks allocated out
508 of a block group specific preallocation pool, so
509 that small files are packed closely together.
510 Each large file will have its blocks allocated
511 out of its own unique preallocation pool.
512
513 session_write_kbytes This file is read-only and shows the number of
514 kilobytes of data that have been written to this
515 filesystem since it was mounted.
Lukas Czerner27dd4382013-04-09 22:11:22 -0400516
517 reserved_clusters This is RW file and contains number of reserved
518 clusters in the file system which will be used
519 in the specific situations to avoid costly
520 zeroout, unexpected ENOSPC, or possible data
521 loss. The default is 2% or 4096 clusters,
522 whichever is smaller and this can be changed
523 however it can never exceed number of clusters
524 in the file system. If there is not enough space
525 for the reserved space when mounting the file
526 mount will _not_ fail.
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400527============================= =================================================
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500528
529Ioctls
530======
531
532There is some Ext4 specific functionality which can be accessed by applications
533through the system call interfaces. The list of all Ext4 specific ioctls are
534shown in the table below.
535
536Table of Ext4 specific ioctls
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400537
538============================= =================================================
539Ioctl Description
540============================= =================================================
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500541 EXT4_IOC_GETFLAGS Get additional attributes associated with inode.
542 The ioctl argument is an integer bitfield, with
543 bit values described in ext4.h. This ioctl is an
544 alias for FS_IOC_GETFLAGS.
545
546 EXT4_IOC_SETFLAGS Set additional attributes associated with inode.
547 The ioctl argument is an integer bitfield, with
548 bit values described in ext4.h. This ioctl is an
549 alias for FS_IOC_SETFLAGS.
550
551 EXT4_IOC_GETVERSION
552 EXT4_IOC_GETVERSION_OLD
553 Get the inode i_generation number stored for
554 each inode. The i_generation number is normally
555 changed only when new inode is created and it is
556 particularly useful for network filesystems. The
557 '_OLD' version of this ioctl is an alias for
558 FS_IOC_GETVERSION.
559
560 EXT4_IOC_SETVERSION
561 EXT4_IOC_SETVERSION_OLD
562 Set the inode i_generation number stored for
563 each inode. The '_OLD' version of this ioctl
564 is an alias for FS_IOC_SETVERSION.
565
566 EXT4_IOC_GROUP_EXTEND This ioctl has the same purpose as the resize
567 mount option. It allows to resize filesystem
568 to the end of the last existing block group,
569 further resize has to be done with resize2fs,
570 either online, or offline. The argument points
571 to the unsigned logn number representing the
572 filesystem new block count.
573
574 EXT4_IOC_MOVE_EXT Move the block extents from orig_fd (the one
575 this ioctl is pointing to) to the donor_fd (the
576 one specified in move_extent structure passed
577 as an argument to this ioctl). Then, exchange
578 inode metadata between orig_fd and donor_fd.
579 This is especially useful for online
580 defragmentation, because the allocator has the
581 opportunity to allocate moved blocks better,
582 ideally into one contiguous extent.
583
584 EXT4_IOC_GROUP_ADD Add a new group descriptor to an existing or
585 new group descriptor block. The new group
586 descriptor is described by ext4_new_group_input
587 structure, which is passed as an argument to
588 this ioctl. This is especially useful in
589 conjunction with EXT4_IOC_GROUP_EXTEND,
590 which allows online resize of the filesystem
591 to the end of the last existing block group.
592 Those two ioctls combined is used in userspace
593 online resize tool (e.g. resize2fs).
594
595 EXT4_IOC_MIGRATE This ioctl operates on the filesystem itself.
596 It converts (migrates) ext3 indirect block mapped
597 inode to ext4 extent mapped inode by walking
598 through indirect block mapping of the original
599 inode and converting contiguous block ranges
600 into ext4 extents of the temporary inode. Then,
601 inodes are swapped. This ioctl might help, when
602 migrating from ext3 to ext4 filesystem, however
603 suggestion is to create fresh ext4 filesystem
604 and copy data from the backup. Note, that
605 filesystem has to support extents for this ioctl
606 to work.
607
608 EXT4_IOC_ALLOC_DA_BLKS Force all of the delay allocated blocks to be
609 allocated to preserve application-expected ext3
610 behaviour. Note that this will also start
611 triggering a write of the data blocks, but this
612 behaviour may change in the future as it is
613 not necessary and has been done this way only
614 for sake of simplicity.
Yongqiang Yang19c52462012-01-04 17:09:44 -0500615
616 EXT4_IOC_RESIZE_FS Resize the filesystem to a new size. The number
617 of blocks of resized filesystem is passed in via
618 64 bit integer argument. The kernel allocates
619 bitmaps and inode table, the userspace tool thus
620 just passes the new number of blocks.
621
Pavel Machek98bfa342017-09-16 13:48:37 +0200622 EXT4_IOC_SWAP_BOOT Swap i_blocks and associated attributes
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -0400623 (like i_blocks, i_size, i_flags, ...) from
624 the specified inode with inode
625 EXT4_BOOT_LOADER_INO (#5). This is typically
626 used to store a boot loader in a secure part of
627 the filesystem, where it can't be changed by a
628 normal user by accident.
629 The data blocks of the previous boot loader
630 will be associated with the given inode.
Darrick J. Wong489fcb92018-07-29 15:36:00 -0400631============================= =================================================
Lukas Czerner6f9524e2011-02-21 20:16:21 -0500632
Dave Kleikampfc513a32006-10-11 01:21:25 -0700633References
634==========
635
636kernel source: <file:fs/ext4/>
637 <file:fs/jbd2/>
638
639programs: http://e2fsprogs.sourceforge.net/
Dave Kleikampfc513a32006-10-11 01:21:25 -0700640
641useful links: http://fedoraproject.org/wiki/ext3-devel
642 http://www.bullopensource.org/ext4/
Jose R. Santos93e32702008-07-11 19:27:31 -0400643 http://ext4.wiki.kernel.org/index.php/Main_Page
644 http://fedoraproject.org/wiki/Features/Ext4