1.. SPDX-License-Identifier: GPL-2.0
2
3========================
4ext4 General Information
5========================
6
7Ext4 is an advanced level of the ext3 filesystem which incorporates
8scalability and reliability enhancements for supporting large filesystems
9(64 bit) in keeping with increasing disk capacities and state-of-the-art
10feature requirements.
11
12Mailing list:	linux-ext4@vger.kernel.org
13Web site:	http://ext4.wiki.kernel.org
14
15
16Quick usage instructions
17========================
18
19Note: More extensive information for getting started with ext4 can be
20found at the ext4 wiki site at the URL:
21http://ext4.wiki.kernel.org/index.php/Ext4_Howto
22
23  - The latest version of e2fsprogs can be found at:
24
25    https://www.kernel.org/pub/linux/kernel/people/tytso/e2fsprogs/
26
27	or
28
29    http://sourceforge.net/project/showfiles.php?group_id=2406
30
31	or grab the latest git repository from:
32
33   https://git.kernel.org/pub/scm/fs/ext2/e2fsprogs.git
34
35  - Create a new filesystem using the ext4 filesystem type:
36
37        # mke2fs -t ext4 /dev/hda1
38
39    Or to configure an existing ext3 filesystem to support extents:
40
41	# tune2fs -O extents /dev/hda1
42
43    If the filesystem was created with 128 byte inodes, it can be
44    converted to use 256 byte for greater efficiency via:
45
46        # tune2fs -I 256 /dev/hda1
47
48  - Mounting:
49
50	# mount -t ext4 /dev/hda1 /wherever
51
52  - When comparing performance with other filesystems, it's always
53    important to try multiple workloads; very often a subtle change in a
54    workload parameter can completely change the ranking of which
55    filesystems do well compared to others.  When comparing versus ext3,
56    note that ext4 enables write barriers by default, while ext3 does
57    not enable write barriers by default.  So it is useful to use
58    explicitly specify whether barriers are enabled or not when via the
59    '-o barriers=[0|1]' mount option for both ext3 and ext4 filesystems
60    for a fair comparison.  When tuning ext3 for best benchmark numbers,
61    it is often worthwhile to try changing the data journaling mode; '-o
62    data=writeback' can be faster for some workloads.  (Note however that
63    running mounted with data=writeback can potentially leave stale data
64    exposed in recently written files in case of an unclean shutdown,
65    which could be a security exposure in some situations.)  Configuring
66    the filesystem with a large journal can also be helpful for
67    metadata-intensive workloads.
68
69Features
70========
71
72Currently Available
73-------------------
74
75* ability to use filesystems > 16TB (e2fsprogs support not available yet)
76* extent format reduces metadata overhead (RAM, IO for access, transactions)
77* extent format more robust in face of on-disk corruption due to magics,
78* internal redundancy in tree
79* improved file allocation (multi-block alloc)
80* lift 32000 subdirectory limit imposed by i_links_count[1]
81* nsec timestamps for mtime, atime, ctime, create time
82* inode version field on disk (NFSv4, Lustre)
83* reduced e2fsck time via uninit_bg feature
84* journal checksumming for robustness, performance
85* persistent file preallocation (e.g for streaming media, databases)
86* ability to pack bitmaps and inode tables into larger virtual groups via the
87  flex_bg feature
88* large file support
89* inode allocation using large virtual block groups via flex_bg
90* delayed allocation
91* large block (up to pagesize) support
92* efficient new ordered mode in JBD2 and ext4 (avoid using buffer head to force
93  the ordering)
94* Case-insensitive file name lookups
95* file-based encryption support (fscrypt)
96* file-based verity support (fsverity)
97
98[1] Filesystems with a block size of 1k may see a limit imposed by the
99directory hash tree having a maximum depth of two.
100
101case-insensitive file name lookups
102======================================================
103
104The case-insensitive file name lookup feature is supported on a
105per-directory basis, allowing the user to mix case-insensitive and
106case-sensitive directories in the same filesystem.  It is enabled by
107flipping the +F inode attribute of an empty directory.  The
108case-insensitive string match operation is only defined when we know how
109text in encoded in a byte sequence.  For that reason, in order to enable
110case-insensitive directories, the filesystem must have the
111casefold feature, which stores the filesystem-wide encoding
112model used.  By default, the charset adopted is the latest version of
113Unicode (12.1.0, by the time of this writing), encoded in the UTF-8
114form.  The comparison algorithm is implemented by normalizing the
115strings to the Canonical decomposition form, as defined by Unicode,
116followed by a byte per byte comparison.
117
118The case-awareness is name-preserving on the disk, meaning that the file
119name provided by userspace is a byte-per-byte match to what is actually
120written in the disk.  The Unicode normalization format used by the
121kernel is thus an internal representation, and not exposed to the
122userspace nor to the disk, with the important exception of disk hashes,
123used on large case-insensitive directories with DX feature.  On DX
124directories, the hash must be calculated using the casefolded version of
125the filename, meaning that the normalization format used actually has an
126impact on where the directory entry is stored.
127
128When we change from viewing filenames as opaque byte sequences to seeing
129them as encoded strings we need to address what happens when a program
130tries to create a file with an invalid name.  The Unicode subsystem
131within the kernel leaves the decision of what to do in this case to the
132filesystem, which select its preferred behavior by enabling/disabling
133the strict mode.  When Ext4 encounters one of those strings and the
134filesystem did not require strict mode, it falls back to considering the
135entire string as an opaque byte sequence, which still allows the user to
136operate on that file, but the case-insensitive lookups won't work.
137
138Options
139=======
140
141When mounting an ext4 filesystem, the following option are accepted:
142(*) == default
143
144  ro
145        Mount filesystem read only. Note that ext4 will replay the journal (and
146        thus write to the partition) even when mounted "read only". The mount
147        options "ro,noload" can be used to prevent writes to the filesystem.
148
149  journal_checksum
150        Enable checksumming of the journal transactions.  This will allow the
151        recovery code in e2fsck and the kernel to detect corruption in the
152        kernel.  It is a compatible change and will be ignored by older
153        kernels.
154
155  journal_async_commit
156        Commit block can be written to disk without waiting for descriptor
157        blocks. If enabled older kernels cannot mount the device. This will
158        enable 'journal_checksum' internally.
159
160  journal_path=path, journal_dev=devnum
161        When the external journal device's major/minor numbers have changed,
162        these options allow the user to specify the new journal location.  The
163        journal device is identified through either its new major/minor numbers
164        encoded in devnum, or via a path to the device.
165
166  norecovery, noload
167        Don't load the journal on mounting.  Note that if the filesystem was
168        not unmounted cleanly, skipping the journal replay will lead to the
169        filesystem containing inconsistencies that can lead to any number of
170        problems.
171
172  data=journal
173        All data are committed into the journal prior to being written into the
174        main file system.  Enabling this mode will disable delayed allocation
175        and O_DIRECT support.
176
177  data=ordered	(*)
178        All data are forced directly out to the main file system prior to its
179        metadata being committed to the journal.
180
181  data=writeback
182        Data ordering is not preserved, data may be written into the main file
183        system after its metadata has been committed to the journal.
184
185  commit=nrsec	(*)
186        This setting limits the maximum age of the running transaction to
187        'nrsec' seconds.  The default value is 5 seconds.  This means that if
188        you lose your power, you will lose as much as the latest 5 seconds of
189        metadata changes (your filesystem will not be damaged though, thanks
190        to the journaling). This default value (or any low value) will hurt
191        performance, but it's good for data-safety.  Setting it to 0 will have
192        the same effect as leaving it at the default (5 seconds).  Setting it
193        to very large values will improve performance.  Note that due to
194        delayed allocation even older data can be lost on power failure since
195        writeback of those data begins only after time set in
196        /proc/sys/vm/dirty_expire_centisecs.
197
198  barrier=<0|1(*)>, barrier(*), nobarrier
199        This enables/disables the use of write barriers in the jbd code.
200        barrier=0 disables, barrier=1 enables.  This also requires an IO stack
201        which can support barriers, and if jbd gets an error on a barrier
202        write, it will disable again with a warning.  Write barriers enforce
203        proper on-disk ordering of journal commits, making volatile disk write
204        caches safe to use, at some performance penalty.  If your disks are
205        battery-backed in one way or another, disabling barriers may safely
206        improve performance.  The mount options "barrier" and "nobarrier" can
207        also be used to enable or disable barriers, for consistency with other
208        ext4 mount options.
209
210  inode_readahead_blks=n
211        This tuning parameter controls the maximum number of inode table blocks
212        that ext4's inode table readahead algorithm will pre-read into the
213        buffer cache.  The default value is 32 blocks.
214
215  nouser_xattr
216        Disables Extended User Attributes.  See the attr(5) manual page for
217        more information about extended attributes.
218
219  noacl
220        This option disables POSIX Access Control List support. If ACL support
221        is enabled in the kernel configuration (CONFIG_EXT4_FS_POSIX_ACL), ACL
222        is enabled by default on mount. See the acl(5) manual page for more
223        information about acl.
224
225  bsddf	(*)
226        Make 'df' act like BSD.
227
228  minixdf
229        Make 'df' act like Minix.
230
231  debug
232        Extra debugging information is sent to syslog.
233
234  abort
235        Simulate the effects of calling ext4_abort() for debugging purposes.
236        This is normally used while remounting a filesystem which is already
237        mounted.
238
239  errors=remount-ro
240        Remount the filesystem read-only on an error.
241
242  errors=continue
243        Keep going on a filesystem error.
244
245  errors=panic
246        Panic and halt the machine if an error occurs.  (These mount options
247        override the errors behavior specified in the superblock, which can be
248        configured using tune2fs)
249
250  data_err=ignore(*)
251        Just print an error message if an error occurs in a file data buffer in
252        ordered mode.
253  data_err=abort
254        Abort the journal if an error occurs in a file data buffer in ordered
255        mode.
256
257  grpid | bsdgroups
258        New objects have the group ID of their parent.
259
260  nogrpid (*) | sysvgroups
261        New objects have the group ID of their creator.
262
263  resgid=n
264        The group ID which may use the reserved blocks.
265
266  resuid=n
267        The user ID which may use the reserved blocks.
268
269  sb=
270        Use alternate superblock at this location.
271
272  quota, noquota, grpquota, usrquota
273        These options are ignored by the filesystem. They are used only by
274        quota tools to recognize volumes where quota should be turned on. See
275        documentation in the quota-tools package for more details
276        (http://sourceforge.net/projects/linuxquota).
277
278  jqfmt=<quota type>, usrjquota=<file>, grpjquota=<file>
279        These options tell filesystem details about quota so that quota
280        information can be properly updated during journal replay. They replace
281        the above quota options. See documentation in the quota-tools package
282        for more details (http://sourceforge.net/projects/linuxquota).
283
284  stripe=n
285        Number of filesystem blocks that mballoc will try to use for allocation
286        size and alignment. For RAID5/6 systems this should be the number of
287        data disks *  RAID chunk size in file system blocks.
288
289  delalloc	(*)
290        Defer block allocation until just before ext4 writes out the block(s)
291        in question.  This allows ext4 to better allocation decisions more
292        efficiently.
293
294  nodelalloc
295        Disable delayed allocation.  Blocks are allocated when the data is
296        copied from userspace to the page cache, either via the write(2) system
297        call or when an mmap'ed page which was previously unallocated is
298        written for the first time.
299
300  max_batch_time=usec
301        Maximum amount of time ext4 should wait for additional filesystem
302        operations to be batch together with a synchronous write operation.
303        Since a synchronous write operation is going to force a commit and then
304        a wait for the I/O complete, it doesn't cost much, and can be a huge
305        throughput win, we wait for a small amount of time to see if any other
306        transactions can piggyback on the synchronous write.   The algorithm
307        used is designed to automatically tune for the speed of the disk, by
308        measuring the amount of time (on average) that it takes to finish
309        committing a transaction.  Call this time the "commit time".  If the
310        time that the transaction has been running is less than the commit
311        time, ext4 will try sleeping for the commit time to see if other
312        operations will join the transaction.   The commit time is capped by
313        the max_batch_time, which defaults to 15000us (15ms).   This
314        optimization can be turned off entirely by setting max_batch_time to 0.
315
316  min_batch_time=usec
317        This parameter sets the commit time (as described above) to be at least
318        min_batch_time.  It defaults to zero microseconds.  Increasing this
319        parameter may improve the throughput of multi-threaded, synchronous
320        workloads on very fast disks, at the cost of increasing latency.
321
322  journal_ioprio=prio
323        The I/O priority (from 0 to 7, where 0 is the highest priority) which
324        should be used for I/O operations submitted by kjournald2 during a
325        commit operation.  This defaults to 3, which is a slightly higher
326        priority than the default I/O priority.
327
328  auto_da_alloc(*), noauto_da_alloc
329        Many broken applications don't use fsync() when replacing existing
330        files via patterns such as fd = open("foo.new")/write(fd,..)/close(fd)/
331        rename("foo.new", "foo"), or worse yet, fd = open("foo",
332        O_TRUNC)/write(fd,..)/close(fd).  If auto_da_alloc is enabled, ext4
333        will detect the replace-via-rename and replace-via-truncate patterns
334        and force that any delayed allocation blocks are allocated such that at
335        the next journal commit, in the default data=ordered mode, the data
336        blocks of the new file are forced to disk before the rename() operation
337        is committed.  This provides roughly the same level of guarantees as
338        ext3, and avoids the "zero-length" problem that can happen when a
339        system crashes before the delayed allocation blocks are forced to disk.
340
341  noinit_itable
342        Do not initialize any uninitialized inode table blocks in the
343        background.  This feature may be used by installation CD's so that the
344        install process can complete as quickly as possible; the inode table
345        initialization process would then be deferred until the next time the
346        file system is unmounted.
347
348  init_itable=n
349        The lazy itable init code will wait n times the number of milliseconds
350        it took to zero out the previous block group's inode table.  This
351        minimizes the impact on the system performance while file system's
352        inode table is being initialized.
353
354  discard, nodiscard(*)
355        Controls whether ext4 should issue discard/TRIM commands to the
356        underlying block device when blocks are freed.  This is useful for SSD
357        devices and sparse/thinly-provisioned LUNs, but it is off by default
358        until sufficient testing has been done.
359
360  nouid32
361        Disables 32-bit UIDs and GIDs.  This is for interoperability  with
362        older kernels which only store and expect 16-bit values.
363
364  block_validity(*), noblock_validity
365        These options enable or disable the in-kernel facility for tracking
366        filesystem metadata blocks within internal data structures.  This
367        allows multi- block allocator and other routines to notice bugs or
368        corrupted allocation bitmaps which cause blocks to be allocated which
369        overlap with filesystem metadata blocks.
370
371  dioread_lock, dioread_nolock
372        Controls whether or not ext4 should use the DIO read locking. If the
373        dioread_nolock option is specified ext4 will allocate uninitialized
374        extent before buffer write and convert the extent to initialized after
375        IO completes. This approach allows ext4 code to avoid using inode
376        mutex, which improves scalability on high speed storages. However this
377        does not work with data journaling and dioread_nolock option will be
378        ignored with kernel warning. Note that dioread_nolock code path is only
379        used for extent-based files.  Because of the restrictions this options
380        comprises it is off by default (e.g. dioread_lock).
381
382  max_dir_size_kb=n
383        This limits the size of directories so that any attempt to expand them
384        beyond the specified limit in kilobytes will cause an ENOSPC error.
385        This is useful in memory constrained environments, where a very large
386        directory can cause severe performance problems or even provoke the Out
387        Of Memory killer.  (For example, if there is only 512mb memory
388        available, a 176mb directory may seriously cramp the system's style.)
389
390  i_version
391        Enable 64-bit inode version support. This option is off by default.
392
393  dax
394        Use direct access (no page cache).  See
395        Documentation/filesystems/dax.rst.  Note that this option is
396        incompatible with data=journal.
397
398  inlinecrypt
399        When possible, encrypt/decrypt the contents of encrypted files using the
400        blk-crypto framework rather than filesystem-layer encryption. This
401        allows the use of inline encryption hardware. The on-disk format is
402        unaffected. For more details, see
403        Documentation/block/inline-encryption.rst.
404
405Data Mode
406=========
407There are 3 different data modes:
408
409* writeback mode
410
411  In data=writeback mode, ext4 does not journal data at all.  This mode provides
412  a similar level of journaling as that of XFS, JFS, and ReiserFS in its default
413  mode - metadata journaling.  A crash+recovery can cause incorrect data to
414  appear in files which were written shortly before the crash.  This mode will
415  typically provide the best ext4 performance.
416
417* ordered mode
418
419  In data=ordered mode, ext4 only officially journals metadata, but it logically
420  groups metadata information related to data changes with the data blocks into
421  a single unit called a transaction.  When it's time to write the new metadata
422  out to disk, the associated data blocks are written first.  In general, this
423  mode performs slightly slower than writeback but significantly faster than
424  journal mode.
425
426* journal mode
427
428  data=journal mode provides full data and metadata journaling.  All new data is
429  written to the journal first, and then to its final location.  In the event of
430  a crash, the journal can be replayed, bringing both data and metadata into a
431  consistent state.  This mode is the slowest except when data needs to be read
432  from and written to disk at the same time where it outperforms all others
433  modes.  Enabling this mode will disable delayed allocation and O_DIRECT
434  support.
435
436/proc entries
437=============
438
439Information about mounted ext4 file systems can be found in
440/proc/fs/ext4.  Each mounted filesystem will have a directory in
441/proc/fs/ext4 based on its device name (i.e., /proc/fs/ext4/hdc or
442/proc/fs/ext4/dm-0).   The files in each per-device directory are shown
443in table below.
444
445Files in /proc/fs/ext4/<devname>
446
447  mb_groups
448        details of multiblock allocator buddy cache of free blocks
449
450/sys entries
451============
452
453Information about mounted ext4 file systems can be found in
454/sys/fs/ext4.  Each mounted filesystem will have a directory in
455/sys/fs/ext4 based on its device name (i.e., /sys/fs/ext4/hdc or
456/sys/fs/ext4/dm-0).   The files in each per-device directory are shown
457in table below.
458
459Files in /sys/fs/ext4/<devname>:
460
461(see also Documentation/ABI/testing/sysfs-fs-ext4)
462
463  delayed_allocation_blocks
464        This file is read-only and shows the number of blocks that are dirty in
465        the page cache, but which do not have their location in the filesystem
466        allocated yet.
467
468  inode_goal
469        Tuning parameter which (if non-zero) controls the goal inode used by
470        the inode allocator in preference to all other allocation heuristics.
471        This is intended for debugging use only, and should be 0 on production
472        systems.
473
474  inode_readahead_blks
475        Tuning parameter which controls the maximum number of inode table
476        blocks that ext4's inode table readahead algorithm will pre-read into
477        the buffer cache.
478
479  lifetime_write_kbytes
480        This file is read-only and shows the number of kilobytes of data that
481        have been written to this filesystem since it was created.
482
483  max_writeback_mb_bump
484        The maximum number of megabytes the writeback code will try to write
485        out before move on to another inode.
486
487  mb_group_prealloc
488        The multiblock allocator will round up allocation requests to a
489        multiple of this tuning parameter if the stripe size is not set in the
490        ext4 superblock
491
492  mb_max_to_scan
493        The maximum number of extents the multiblock allocator will search to
494        find the best extent.
495
496  mb_min_to_scan
497        The minimum number of extents the multiblock allocator will search to
498        find the best extent.
499
500  mb_order2_req
501        Tuning parameter which controls the minimum size for requests (as a
502        power of 2) where the buddy cache is used.
503
504  mb_stats
505        Controls whether the multiblock allocator should collect statistics,
506        which are shown during the unmount. 1 means to collect statistics, 0
507        means not to collect statistics.
508
509  mb_stream_req
510        Files which have fewer blocks than this tunable parameter will have
511        their blocks allocated out of a block group specific preallocation
512        pool, so that small files are packed closely together.  Each large file
513        will have its blocks allocated out of its own unique preallocation
514        pool.
515
516  session_write_kbytes
517        This file is read-only and shows the number of kilobytes of data that
518        have been written to this filesystem since it was mounted.
519
520  reserved_clusters
521        This is RW file and contains number of reserved clusters in the file
522        system which will be used in the specific situations to avoid costly
523        zeroout, unexpected ENOSPC, or possible data loss. The default is 2% or
524        4096 clusters, whichever is smaller and this can be changed however it
525        can never exceed number of clusters in the file system. If there is not
526        enough space for the reserved space when mounting the file mount will
527        _not_ fail.
528
529Ioctls
530======
531
532Ext4 implements various ioctls which can be used by applications to access
533ext4-specific functionality. An incomplete list of these ioctls is shown in the
534table below. This list includes truly ext4-specific ioctls (``EXT4_IOC_*``) as
535well as ioctls that may have been ext4-specific originally but are now supported
536by some other filesystem(s) too (``FS_IOC_*``).
537
538Table of Ext4 ioctls
539
540  FS_IOC_GETFLAGS
541        Get additional attributes associated with inode.  The ioctl argument is
542        an integer bitfield, with bit values described in ext4.h.
543
544  FS_IOC_SETFLAGS
545        Set additional attributes associated with inode.  The ioctl argument is
546        an integer bitfield, with bit values described in ext4.h.
547
548  EXT4_IOC_GETVERSION, EXT4_IOC_GETVERSION_OLD
549        Get the inode i_generation number stored for each inode. The
550        i_generation number is normally changed only when new inode is created
551        and it is particularly useful for network filesystems. The '_OLD'
552        version of this ioctl is an alias for FS_IOC_GETVERSION.
553
554  EXT4_IOC_SETVERSION, EXT4_IOC_SETVERSION_OLD
555        Set the inode i_generation number stored for each inode. The '_OLD'
556        version of this ioctl is an alias for FS_IOC_SETVERSION.
557
558  EXT4_IOC_GROUP_EXTEND
559        This ioctl has the same purpose as the resize mount option. It allows
560        to resize filesystem to the end of the last existing block group,
561        further resize has to be done with resize2fs, either online, or
562        offline. The argument points to the unsigned logn number representing
563        the filesystem new block count.
564
565  EXT4_IOC_MOVE_EXT
566        Move the block extents from orig_fd (the one this ioctl is pointing to)
567        to the donor_fd (the one specified in move_extent structure passed as
568        an argument to this ioctl). Then, exchange inode metadata between
569        orig_fd and donor_fd.  This is especially useful for online
570        defragmentation, because the allocator has the opportunity to allocate
571        moved blocks better, ideally into one contiguous extent.
572
573  EXT4_IOC_GROUP_ADD
574        Add a new group descriptor to an existing or new group descriptor
575        block. The new group descriptor is described by ext4_new_group_input
576        structure, which is passed as an argument to this ioctl. This is
577        especially useful in conjunction with EXT4_IOC_GROUP_EXTEND, which
578        allows online resize of the filesystem to the end of the last existing
579        block group.  Those two ioctls combined is used in userspace online
580        resize tool (e.g. resize2fs).
581
582  EXT4_IOC_MIGRATE
583        This ioctl operates on the filesystem itself.  It converts (migrates)
584        ext3 indirect block mapped inode to ext4 extent mapped inode by walking
585        through indirect block mapping of the original inode and converting
586        contiguous block ranges into ext4 extents of the temporary inode. Then,
587        inodes are swapped. This ioctl might help, when migrating from ext3 to
588        ext4 filesystem, however suggestion is to create fresh ext4 filesystem
589        and copy data from the backup. Note, that filesystem has to support
590        extents for this ioctl to work.
591
592  EXT4_IOC_ALLOC_DA_BLKS
593        Force all of the delay allocated blocks to be allocated to preserve
594        application-expected ext3 behaviour. Note that this will also start
595        triggering a write of the data blocks, but this behaviour may change in
596        the future as it is not necessary and has been done this way only for
597        sake of simplicity.
598
599  EXT4_IOC_RESIZE_FS
600        Resize the filesystem to a new size.  The number of blocks of resized
601        filesystem is passed in via 64 bit integer argument.  The kernel
602        allocates bitmaps and inode table, the userspace tool thus just passes
603        the new number of blocks.
604
605  EXT4_IOC_SWAP_BOOT
606        Swap i_blocks and associated attributes (like i_blocks, i_size,
607        i_flags, ...) from the specified inode with inode EXT4_BOOT_LOADER_INO
608        (#5). This is typically used to store a boot loader in a secure part of
609        the filesystem, where it can't be changed by a normal user by accident.
610        The data blocks of the previous boot loader will be associated with the
611        given inode.
612
613References
614==========
615
616kernel source:	<file:fs/ext4/>
617		<file:fs/jbd2/>
618
619programs:	http://e2fsprogs.sourceforge.net/
620
621useful links:	https://fedoraproject.org/wiki/ext3-devel
622		http://www.bullopensource.org/ext4/
623		http://ext4.wiki.kernel.org/index.php/Main_Page
624		https://fedoraproject.org/wiki/Features/Ext4
625