Loading...
1.. SPDX-License-Identifier: GPL-2.0
2
3
4==============================
5The Second Extended Filesystem
6==============================
7
8ext2 was originally released in January 1993. Written by R\'emy Card,
9Theodore Ts'o and Stephen Tweedie, it was a major rewrite of the
10Extended Filesystem. It is currently still (April 2001) the predominant
11filesystem in use by Linux. There are also implementations available
12for NetBSD, FreeBSD, the GNU HURD, Windows 95/98/NT, OS/2 and RISC OS.
13
14Options
15=======
16
17Most defaults are determined by the filesystem superblock, and can be
18set using tune2fs(8). Kernel-determined defaults are indicated by (*).
19
20==================== === ================================================
21bsddf (*) Makes ``df`` act like BSD.
22minixdf Makes ``df`` act like Minix.
23
24check=none, nocheck (*) Don't do extra checking of bitmaps on mount
25 (check=normal and check=strict options removed)
26
27dax Use direct access (no page cache). See
28 Documentation/filesystems/dax.rst.
29
30debug Extra debugging information is sent to the
31 kernel syslog. Useful for developers.
32
33errors=continue Keep going on a filesystem error.
34errors=remount-ro Remount the filesystem read-only on an error.
35errors=panic Panic and halt the machine if an error occurs.
36
37grpid, bsdgroups Give objects the same group ID as their parent.
38nogrpid, sysvgroups New objects have the group ID of their creator.
39
40nouid32 Use 16-bit UIDs and GIDs.
41
42oldalloc Enable the old block allocator. Orlov should
43 have better performance, we'd like to get some
44 feedback if it's the contrary for you.
45orlov (*) Use the Orlov block allocator.
46 (See http://lwn.net/Articles/14633/ and
47 http://lwn.net/Articles/14446/.)
48
49resuid=n The user ID which may use the reserved blocks.
50resgid=n The group ID which may use the reserved blocks.
51
52sb=n Use alternate superblock at this location.
53
54user_xattr Enable "user." POSIX Extended Attributes
55 (requires CONFIG_EXT2_FS_XATTR).
56nouser_xattr Don't support "user." extended attributes.
57
58acl Enable POSIX Access Control Lists support
59 (requires CONFIG_EXT2_FS_POSIX_ACL).
60noacl Don't support POSIX ACLs.
61
62quota, usrquota Enable user disk quota support
63 (requires CONFIG_QUOTA).
64
65grpquota Enable group disk quota support
66 (requires CONFIG_QUOTA).
67==================== === ================================================
68
69noquota option ls silently ignored by ext2.
70
71
72Specification
73=============
74
75ext2 shares many properties with traditional Unix filesystems. It has
76the concepts of blocks, inodes and directories. It has space in the
77specification for Access Control Lists (ACLs), fragments, undeletion and
78compression though these are not yet implemented (some are available as
79separate patches). There is also a versioning mechanism to allow new
80features (such as journalling) to be added in a maximally compatible
81manner.
82
83Blocks
84------
85
86The space in the device or file is split up into blocks. These are
87a fixed size, of 1024, 2048 or 4096 bytes (8192 bytes on Alpha systems),
88which is decided when the filesystem is created. Smaller blocks mean
89less wasted space per file, but require slightly more accounting overhead,
90and also impose other limits on the size of files and the filesystem.
91
92Block Groups
93------------
94
95Blocks are clustered into block groups in order to reduce fragmentation
96and minimise the amount of head seeking when reading a large amount
97of consecutive data. Information about each block group is kept in a
98descriptor table stored in the block(s) immediately after the superblock.
99Two blocks near the start of each group are reserved for the block usage
100bitmap and the inode usage bitmap which show which blocks and inodes
101are in use. Since each bitmap is limited to a single block, this means
102that the maximum size of a block group is 8 times the size of a block.
103
104The block(s) following the bitmaps in each block group are designated
105as the inode table for that block group and the remainder are the data
106blocks. The block allocation algorithm attempts to allocate data blocks
107in the same block group as the inode which contains them.
108
109The Superblock
110--------------
111
112The superblock contains all the information about the configuration of
113the filing system. The primary copy of the superblock is stored at an
114offset of 1024 bytes from the start of the device, and it is essential
115to mounting the filesystem. Since it is so important, backup copies of
116the superblock are stored in block groups throughout the filesystem.
117The first version of ext2 (revision 0) stores a copy at the start of
118every block group, along with backups of the group descriptor block(s).
119Because this can consume a considerable amount of space for large
120filesystems, later revisions can optionally reduce the number of backup
121copies by only putting backups in specific groups (this is the sparse
122superblock feature). The groups chosen are 0, 1 and powers of 3, 5 and 7.
123
124The information in the superblock contains fields such as the total
125number of inodes and blocks in the filesystem and how many are free,
126how many inodes and blocks are in each block group, when the filesystem
127was mounted (and if it was cleanly unmounted), when it was modified,
128what version of the filesystem it is (see the Revisions section below)
129and which OS created it.
130
131If the filesystem is revision 1 or higher, then there are extra fields,
132such as a volume name, a unique identification number, the inode size,
133and space for optional filesystem features to store configuration info.
134
135All fields in the superblock (as in all other ext2 structures) are stored
136on the disc in little endian format, so a filesystem is portable between
137machines without having to know what machine it was created on.
138
139Inodes
140------
141
142The inode (index node) is a fundamental concept in the ext2 filesystem.
143Each object in the filesystem is represented by an inode. The inode
144structure contains pointers to the filesystem blocks which contain the
145data held in the object and all of the metadata about an object except
146its name. The metadata about an object includes the permissions, owner,
147group, flags, size, number of blocks used, access time, change time,
148modification time, deletion time, number of links, fragments, version
149(for NFS) and extended attributes (EAs) and/or Access Control Lists (ACLs).
150
151There are some reserved fields which are currently unused in the inode
152structure and several which are overloaded. One field is reserved for the
153directory ACL if the inode is a directory and alternately for the top 32
154bits of the file size if the inode is a regular file (allowing file sizes
155larger than 2GB). The translator field is unused under Linux, but is used
156by the HURD to reference the inode of a program which will be used to
157interpret this object. Most of the remaining reserved fields have been
158used up for both Linux and the HURD for larger owner and group fields,
159The HURD also has a larger mode field so it uses another of the remaining
160fields to store the extra more bits.
161
162There are pointers to the first 12 blocks which contain the file's data
163in the inode. There is a pointer to an indirect block (which contains
164pointers to the next set of blocks), a pointer to a doubly-indirect
165block (which contains pointers to indirect blocks) and a pointer to a
166trebly-indirect block (which contains pointers to doubly-indirect blocks).
167
168The flags field contains some ext2-specific flags which aren't catered
169for by the standard chmod flags. These flags can be listed with lsattr
170and changed with the chattr command, and allow specific filesystem
171behaviour on a per-file basis. There are flags for secure deletion,
172undeletable, compression, synchronous updates, immutability, append-only,
173dumpable, no-atime, indexed directories, and data-journaling. Not all
174of these are supported yet.
175
176Directories
177-----------
178
179A directory is a filesystem object and has an inode just like a file.
180It is a specially formatted file containing records which associate
181each name with an inode number. Later revisions of the filesystem also
182encode the type of the object (file, directory, symlink, device, fifo,
183socket) to avoid the need to check the inode itself for this information
184(support for taking advantage of this feature does not yet exist in
185Glibc 2.2).
186
187The inode allocation code tries to assign inodes which are in the same
188block group as the directory in which they are first created.
189
190The current implementation of ext2 uses a singly-linked list to store
191the filenames in the directory; a pending enhancement uses hashing of the
192filenames to allow lookup without the need to scan the entire directory.
193
194The current implementation never removes empty directory blocks once they
195have been allocated to hold more files.
196
197Special files
198-------------
199
200Symbolic links are also filesystem objects with inodes. They deserve
201special mention because the data for them is stored within the inode
202itself if the symlink is less than 60 bytes long. It uses the fields
203which would normally be used to store the pointers to data blocks.
204This is a worthwhile optimisation as it we avoid allocating a full
205block for the symlink, and most symlinks are less than 60 characters long.
206
207Character and block special devices never have data blocks assigned to
208them. Instead, their device number is stored in the inode, again reusing
209the fields which would be used to point to the data blocks.
210
211Reserved Space
212--------------
213
214In ext2, there is a mechanism for reserving a certain number of blocks
215for a particular user (normally the super-user). This is intended to
216allow for the system to continue functioning even if non-privileged users
217fill up all the space available to them (this is independent of filesystem
218quotas). It also keeps the filesystem from filling up entirely which
219helps combat fragmentation.
220
221Filesystem check
222----------------
223
224At boot time, most systems run a consistency check (e2fsck) on their
225filesystems. The superblock of the ext2 filesystem contains several
226fields which indicate whether fsck should actually run (since checking
227the filesystem at boot can take a long time if it is large). fsck will
228run if the filesystem was not cleanly unmounted, if the maximum mount
229count has been exceeded or if the maximum time between checks has been
230exceeded.
231
232Feature Compatibility
233---------------------
234
235The compatibility feature mechanism used in ext2 is sophisticated.
236It safely allows features to be added to the filesystem, without
237unnecessarily sacrificing compatibility with older versions of the
238filesystem code. The feature compatibility mechanism is not supported by
239the original revision 0 (EXT2_GOOD_OLD_REV) of ext2, but was introduced in
240revision 1. There are three 32-bit fields, one for compatible features
241(COMPAT), one for read-only compatible (RO_COMPAT) features and one for
242incompatible (INCOMPAT) features.
243
244These feature flags have specific meanings for the kernel as follows:
245
246A COMPAT flag indicates that a feature is present in the filesystem,
247but the on-disk format is 100% compatible with older on-disk formats, so
248a kernel which didn't know anything about this feature could read/write
249the filesystem without any chance of corrupting the filesystem (or even
250making it inconsistent). This is essentially just a flag which says
251"this filesystem has a (hidden) feature" that the kernel or e2fsck may
252want to be aware of (more on e2fsck and feature flags later). The ext3
253HAS_JOURNAL feature is a COMPAT flag because the ext3 journal is simply
254a regular file with data blocks in it so the kernel does not need to
255take any special notice of it if it doesn't understand ext3 journaling.
256
257An RO_COMPAT flag indicates that the on-disk format is 100% compatible
258with older on-disk formats for reading (i.e. the feature does not change
259the visible on-disk format). However, an old kernel writing to such a
260filesystem would/could corrupt the filesystem, so this is prevented. The
261most common such feature, SPARSE_SUPER, is an RO_COMPAT feature because
262sparse groups allow file data blocks where superblock/group descriptor
263backups used to live, and ext2_free_blocks() refuses to free these blocks,
264which would leading to inconsistent bitmaps. An old kernel would also
265get an error if it tried to free a series of blocks which crossed a group
266boundary, but this is a legitimate layout in a SPARSE_SUPER filesystem.
267
268An INCOMPAT flag indicates the on-disk format has changed in some
269way that makes it unreadable by older kernels, or would otherwise
270cause a problem if an old kernel tried to mount it. FILETYPE is an
271INCOMPAT flag because older kernels would think a filename was longer
272than 256 characters, which would lead to corrupt directory listings.
273The COMPRESSION flag is an obvious INCOMPAT flag - if the kernel
274doesn't understand compression, you would just get garbage back from
275read() instead of it automatically decompressing your data. The ext3
276RECOVER flag is needed to prevent a kernel which does not understand the
277ext3 journal from mounting the filesystem without replaying the journal.
278
279For e2fsck, it needs to be more strict with the handling of these
280flags than the kernel. If it doesn't understand ANY of the COMPAT,
281RO_COMPAT, or INCOMPAT flags it will refuse to check the filesystem,
282because it has no way of verifying whether a given feature is valid
283or not. Allowing e2fsck to succeed on a filesystem with an unknown
284feature is a false sense of security for the user. Refusing to check
285a filesystem with unknown features is a good incentive for the user to
286update to the latest e2fsck. This also means that anyone adding feature
287flags to ext2 also needs to update e2fsck to verify these features.
288
289Metadata
290--------
291
292It is frequently claimed that the ext2 implementation of writing
293asynchronous metadata is faster than the ffs synchronous metadata
294scheme but less reliable. Both methods are equally resolvable by their
295respective fsck programs.
296
297If you're exceptionally paranoid, there are 3 ways of making metadata
298writes synchronous on ext2:
299
300- per-file if you have the program source: use the O_SYNC flag to open()
301- per-file if you don't have the source: use "chattr +S" on the file
302- per-filesystem: add the "sync" option to mount (or in /etc/fstab)
303
304the first and last are not ext2 specific but do force the metadata to
305be written synchronously. See also Journaling below.
306
307Limitations
308-----------
309
310There are various limits imposed by the on-disk layout of ext2. Other
311limits are imposed by the current implementation of the kernel code.
312Many of the limits are determined at the time the filesystem is first
313created, and depend upon the block size chosen. The ratio of inodes to
314data blocks is fixed at filesystem creation time, so the only way to
315increase the number of inodes is to increase the size of the filesystem.
316No tools currently exist which can change the ratio of inodes to blocks.
317
318Most of these limits could be overcome with slight changes in the on-disk
319format and using a compatibility flag to signal the format change (at
320the expense of some compatibility).
321
322===================== ======= ======= ======= ========
323Filesystem block size 1kB 2kB 4kB 8kB
324===================== ======= ======= ======= ========
325File size limit 16GB 256GB 2048GB 2048GB
326Filesystem size limit 2047GB 8192GB 16384GB 32768GB
327===================== ======= ======= ======= ========
328
329There is a 2.4 kernel limit of 2048GB for a single block device, so no
330filesystem larger than that can be created at this time. There is also
331an upper limit on the block size imposed by the page size of the kernel,
332so 8kB blocks are only allowed on Alpha systems (and other architectures
333which support larger pages).
334
335There is an upper limit of 32000 subdirectories in a single directory.
336
337There is a "soft" upper limit of about 10-15k files in a single directory
338with the current linear linked-list directory implementation. This limit
339stems from performance problems when creating and deleting (and also
340finding) files in such large directories. Using a hashed directory index
341(under development) allows 100k-1M+ files in a single directory without
342performance problems (although RAM size becomes an issue at this point).
343
344The (meaningless) absolute upper limit of files in a single directory
345(imposed by the file size, the realistic limit is obviously much less)
346is over 130 trillion files. It would be higher except there are not
347enough 4-character names to make up unique directory entries, so they
348have to be 8 character filenames, even then we are fairly close to
349running out of unique filenames.
350
351Journaling
352----------
353
354A journaling extension to the ext2 code has been developed by Stephen
355Tweedie. It avoids the risks of metadata corruption and the need to
356wait for e2fsck to complete after a crash, without requiring a change
357to the on-disk ext2 layout. In a nutshell, the journal is a regular
358file which stores whole metadata (and optionally data) blocks that have
359been modified, prior to writing them into the filesystem. This means
360it is possible to add a journal to an existing ext2 filesystem without
361the need for data conversion.
362
363When changes to the filesystem (e.g. a file is renamed) they are stored in
364a transaction in the journal and can either be complete or incomplete at
365the time of a crash. If a transaction is complete at the time of a crash
366(or in the normal case where the system does not crash), then any blocks
367in that transaction are guaranteed to represent a valid filesystem state,
368and are copied into the filesystem. If a transaction is incomplete at
369the time of the crash, then there is no guarantee of consistency for
370the blocks in that transaction so they are discarded (which means any
371filesystem changes they represent are also lost).
372Check Documentation/filesystems/ext4/ if you want to read more about
373ext4 and journaling.
374
375References
376==========
377
378======================= ===============================================
379The kernel source file:/usr/src/linux/fs/ext2/
380e2fsprogs (e2fsck) http://e2fsprogs.sourceforge.net/
381Design & Implementation http://e2fsprogs.sourceforge.net/ext2intro.html
382Journaling (ext3) ftp://ftp.uk.linux.org/pub/linux/sct/fs/jfs/
383Filesystem Resizing http://ext2resize.sourceforge.net/
384Compression [1]_ http://e2compr.sourceforge.net/
385======================= ===============================================
386
387Implementations for:
388
389======================= ===========================================================
390Windows 95/98/NT/2000 http://www.chrysocome.net/explore2fs
391Windows 95 [1]_ http://www.yipton.net/content.html#FSDEXT2
392DOS client [1]_ ftp://metalab.unc.edu/pub/Linux/system/filesystems/ext2/
393OS/2 [2]_ ftp://metalab.unc.edu/pub/Linux/system/filesystems/ext2/
394RISC OS client http://www.esw-heim.tu-clausthal.de/~marco/smorbrod/IscaFS/
395======================= ===========================================================
396
397.. [1] no longer actively developed/supported (as of Apr 2001)
398.. [2] no longer actively developed/supported (as of Mar 2009)
1.. SPDX-License-Identifier: GPL-2.0
2
3
4The Second Extended Filesystem
5==============================
6
7ext2 was originally released in January 1993. Written by R\'emy Card,
8Theodore Ts'o and Stephen Tweedie, it was a major rewrite of the
9Extended Filesystem. It is currently still (April 2001) the predominant
10filesystem in use by Linux. There are also implementations available
11for NetBSD, FreeBSD, the GNU HURD, Windows 95/98/NT, OS/2 and RISC OS.
12
13Options
14=======
15
16Most defaults are determined by the filesystem superblock, and can be
17set using tune2fs(8). Kernel-determined defaults are indicated by (*).
18
19==================== === ================================================
20bsddf (*) Makes ``df`` act like BSD.
21minixdf Makes ``df`` act like Minix.
22
23check=none, nocheck (*) Don't do extra checking of bitmaps on mount
24 (check=normal and check=strict options removed)
25
26dax Use direct access (no page cache). See
27 Documentation/filesystems/dax.txt.
28
29debug Extra debugging information is sent to the
30 kernel syslog. Useful for developers.
31
32errors=continue Keep going on a filesystem error.
33errors=remount-ro Remount the filesystem read-only on an error.
34errors=panic Panic and halt the machine if an error occurs.
35
36grpid, bsdgroups Give objects the same group ID as their parent.
37nogrpid, sysvgroups New objects have the group ID of their creator.
38
39nouid32 Use 16-bit UIDs and GIDs.
40
41oldalloc Enable the old block allocator. Orlov should
42 have better performance, we'd like to get some
43 feedback if it's the contrary for you.
44orlov (*) Use the Orlov block allocator.
45 (See http://lwn.net/Articles/14633/ and
46 http://lwn.net/Articles/14446/.)
47
48resuid=n The user ID which may use the reserved blocks.
49resgid=n The group ID which may use the reserved blocks.
50
51sb=n Use alternate superblock at this location.
52
53user_xattr Enable "user." POSIX Extended Attributes
54 (requires CONFIG_EXT2_FS_XATTR).
55nouser_xattr Don't support "user." extended attributes.
56
57acl Enable POSIX Access Control Lists support
58 (requires CONFIG_EXT2_FS_POSIX_ACL).
59noacl Don't support POSIX ACLs.
60
61nobh Do not attach buffer_heads to file pagecache.
62
63quota, usrquota Enable user disk quota support
64 (requires CONFIG_QUOTA).
65
66grpquota Enable group disk quota support
67 (requires CONFIG_QUOTA).
68==================== === ================================================
69
70noquota option ls silently ignored by ext2.
71
72
73Specification
74=============
75
76ext2 shares many properties with traditional Unix filesystems. It has
77the concepts of blocks, inodes and directories. It has space in the
78specification for Access Control Lists (ACLs), fragments, undeletion and
79compression though these are not yet implemented (some are available as
80separate patches). There is also a versioning mechanism to allow new
81features (such as journalling) to be added in a maximally compatible
82manner.
83
84Blocks
85------
86
87The space in the device or file is split up into blocks. These are
88a fixed size, of 1024, 2048 or 4096 bytes (8192 bytes on Alpha systems),
89which is decided when the filesystem is created. Smaller blocks mean
90less wasted space per file, but require slightly more accounting overhead,
91and also impose other limits on the size of files and the filesystem.
92
93Block Groups
94------------
95
96Blocks are clustered into block groups in order to reduce fragmentation
97and minimise the amount of head seeking when reading a large amount
98of consecutive data. Information about each block group is kept in a
99descriptor table stored in the block(s) immediately after the superblock.
100Two blocks near the start of each group are reserved for the block usage
101bitmap and the inode usage bitmap which show which blocks and inodes
102are in use. Since each bitmap is limited to a single block, this means
103that the maximum size of a block group is 8 times the size of a block.
104
105The block(s) following the bitmaps in each block group are designated
106as the inode table for that block group and the remainder are the data
107blocks. The block allocation algorithm attempts to allocate data blocks
108in the same block group as the inode which contains them.
109
110The Superblock
111--------------
112
113The superblock contains all the information about the configuration of
114the filing system. The primary copy of the superblock is stored at an
115offset of 1024 bytes from the start of the device, and it is essential
116to mounting the filesystem. Since it is so important, backup copies of
117the superblock are stored in block groups throughout the filesystem.
118The first version of ext2 (revision 0) stores a copy at the start of
119every block group, along with backups of the group descriptor block(s).
120Because this can consume a considerable amount of space for large
121filesystems, later revisions can optionally reduce the number of backup
122copies by only putting backups in specific groups (this is the sparse
123superblock feature). The groups chosen are 0, 1 and powers of 3, 5 and 7.
124
125The information in the superblock contains fields such as the total
126number of inodes and blocks in the filesystem and how many are free,
127how many inodes and blocks are in each block group, when the filesystem
128was mounted (and if it was cleanly unmounted), when it was modified,
129what version of the filesystem it is (see the Revisions section below)
130and which OS created it.
131
132If the filesystem is revision 1 or higher, then there are extra fields,
133such as a volume name, a unique identification number, the inode size,
134and space for optional filesystem features to store configuration info.
135
136All fields in the superblock (as in all other ext2 structures) are stored
137on the disc in little endian format, so a filesystem is portable between
138machines without having to know what machine it was created on.
139
140Inodes
141------
142
143The inode (index node) is a fundamental concept in the ext2 filesystem.
144Each object in the filesystem is represented by an inode. The inode
145structure contains pointers to the filesystem blocks which contain the
146data held in the object and all of the metadata about an object except
147its name. The metadata about an object includes the permissions, owner,
148group, flags, size, number of blocks used, access time, change time,
149modification time, deletion time, number of links, fragments, version
150(for NFS) and extended attributes (EAs) and/or Access Control Lists (ACLs).
151
152There are some reserved fields which are currently unused in the inode
153structure and several which are overloaded. One field is reserved for the
154directory ACL if the inode is a directory and alternately for the top 32
155bits of the file size if the inode is a regular file (allowing file sizes
156larger than 2GB). The translator field is unused under Linux, but is used
157by the HURD to reference the inode of a program which will be used to
158interpret this object. Most of the remaining reserved fields have been
159used up for both Linux and the HURD for larger owner and group fields,
160The HURD also has a larger mode field so it uses another of the remaining
161fields to store the extra more bits.
162
163There are pointers to the first 12 blocks which contain the file's data
164in the inode. There is a pointer to an indirect block (which contains
165pointers to the next set of blocks), a pointer to a doubly-indirect
166block (which contains pointers to indirect blocks) and a pointer to a
167trebly-indirect block (which contains pointers to doubly-indirect blocks).
168
169The flags field contains some ext2-specific flags which aren't catered
170for by the standard chmod flags. These flags can be listed with lsattr
171and changed with the chattr command, and allow specific filesystem
172behaviour on a per-file basis. There are flags for secure deletion,
173undeletable, compression, synchronous updates, immutability, append-only,
174dumpable, no-atime, indexed directories, and data-journaling. Not all
175of these are supported yet.
176
177Directories
178-----------
179
180A directory is a filesystem object and has an inode just like a file.
181It is a specially formatted file containing records which associate
182each name with an inode number. Later revisions of the filesystem also
183encode the type of the object (file, directory, symlink, device, fifo,
184socket) to avoid the need to check the inode itself for this information
185(support for taking advantage of this feature does not yet exist in
186Glibc 2.2).
187
188The inode allocation code tries to assign inodes which are in the same
189block group as the directory in which they are first created.
190
191The current implementation of ext2 uses a singly-linked list to store
192the filenames in the directory; a pending enhancement uses hashing of the
193filenames to allow lookup without the need to scan the entire directory.
194
195The current implementation never removes empty directory blocks once they
196have been allocated to hold more files.
197
198Special files
199-------------
200
201Symbolic links are also filesystem objects with inodes. They deserve
202special mention because the data for them is stored within the inode
203itself if the symlink is less than 60 bytes long. It uses the fields
204which would normally be used to store the pointers to data blocks.
205This is a worthwhile optimisation as it we avoid allocating a full
206block for the symlink, and most symlinks are less than 60 characters long.
207
208Character and block special devices never have data blocks assigned to
209them. Instead, their device number is stored in the inode, again reusing
210the fields which would be used to point to the data blocks.
211
212Reserved Space
213--------------
214
215In ext2, there is a mechanism for reserving a certain number of blocks
216for a particular user (normally the super-user). This is intended to
217allow for the system to continue functioning even if non-privileged users
218fill up all the space available to them (this is independent of filesystem
219quotas). It also keeps the filesystem from filling up entirely which
220helps combat fragmentation.
221
222Filesystem check
223----------------
224
225At boot time, most systems run a consistency check (e2fsck) on their
226filesystems. The superblock of the ext2 filesystem contains several
227fields which indicate whether fsck should actually run (since checking
228the filesystem at boot can take a long time if it is large). fsck will
229run if the filesystem was not cleanly unmounted, if the maximum mount
230count has been exceeded or if the maximum time between checks has been
231exceeded.
232
233Feature Compatibility
234---------------------
235
236The compatibility feature mechanism used in ext2 is sophisticated.
237It safely allows features to be added to the filesystem, without
238unnecessarily sacrificing compatibility with older versions of the
239filesystem code. The feature compatibility mechanism is not supported by
240the original revision 0 (EXT2_GOOD_OLD_REV) of ext2, but was introduced in
241revision 1. There are three 32-bit fields, one for compatible features
242(COMPAT), one for read-only compatible (RO_COMPAT) features and one for
243incompatible (INCOMPAT) features.
244
245These feature flags have specific meanings for the kernel as follows:
246
247A COMPAT flag indicates that a feature is present in the filesystem,
248but the on-disk format is 100% compatible with older on-disk formats, so
249a kernel which didn't know anything about this feature could read/write
250the filesystem without any chance of corrupting the filesystem (or even
251making it inconsistent). This is essentially just a flag which says
252"this filesystem has a (hidden) feature" that the kernel or e2fsck may
253want to be aware of (more on e2fsck and feature flags later). The ext3
254HAS_JOURNAL feature is a COMPAT flag because the ext3 journal is simply
255a regular file with data blocks in it so the kernel does not need to
256take any special notice of it if it doesn't understand ext3 journaling.
257
258An RO_COMPAT flag indicates that the on-disk format is 100% compatible
259with older on-disk formats for reading (i.e. the feature does not change
260the visible on-disk format). However, an old kernel writing to such a
261filesystem would/could corrupt the filesystem, so this is prevented. The
262most common such feature, SPARSE_SUPER, is an RO_COMPAT feature because
263sparse groups allow file data blocks where superblock/group descriptor
264backups used to live, and ext2_free_blocks() refuses to free these blocks,
265which would leading to inconsistent bitmaps. An old kernel would also
266get an error if it tried to free a series of blocks which crossed a group
267boundary, but this is a legitimate layout in a SPARSE_SUPER filesystem.
268
269An INCOMPAT flag indicates the on-disk format has changed in some
270way that makes it unreadable by older kernels, or would otherwise
271cause a problem if an old kernel tried to mount it. FILETYPE is an
272INCOMPAT flag because older kernels would think a filename was longer
273than 256 characters, which would lead to corrupt directory listings.
274The COMPRESSION flag is an obvious INCOMPAT flag - if the kernel
275doesn't understand compression, you would just get garbage back from
276read() instead of it automatically decompressing your data. The ext3
277RECOVER flag is needed to prevent a kernel which does not understand the
278ext3 journal from mounting the filesystem without replaying the journal.
279
280For e2fsck, it needs to be more strict with the handling of these
281flags than the kernel. If it doesn't understand ANY of the COMPAT,
282RO_COMPAT, or INCOMPAT flags it will refuse to check the filesystem,
283because it has no way of verifying whether a given feature is valid
284or not. Allowing e2fsck to succeed on a filesystem with an unknown
285feature is a false sense of security for the user. Refusing to check
286a filesystem with unknown features is a good incentive for the user to
287update to the latest e2fsck. This also means that anyone adding feature
288flags to ext2 also needs to update e2fsck to verify these features.
289
290Metadata
291--------
292
293It is frequently claimed that the ext2 implementation of writing
294asynchronous metadata is faster than the ffs synchronous metadata
295scheme but less reliable. Both methods are equally resolvable by their
296respective fsck programs.
297
298If you're exceptionally paranoid, there are 3 ways of making metadata
299writes synchronous on ext2:
300
301- per-file if you have the program source: use the O_SYNC flag to open()
302- per-file if you don't have the source: use "chattr +S" on the file
303- per-filesystem: add the "sync" option to mount (or in /etc/fstab)
304
305the first and last are not ext2 specific but do force the metadata to
306be written synchronously. See also Journaling below.
307
308Limitations
309-----------
310
311There are various limits imposed by the on-disk layout of ext2. Other
312limits are imposed by the current implementation of the kernel code.
313Many of the limits are determined at the time the filesystem is first
314created, and depend upon the block size chosen. The ratio of inodes to
315data blocks is fixed at filesystem creation time, so the only way to
316increase the number of inodes is to increase the size of the filesystem.
317No tools currently exist which can change the ratio of inodes to blocks.
318
319Most of these limits could be overcome with slight changes in the on-disk
320format and using a compatibility flag to signal the format change (at
321the expense of some compatibility).
322
323===================== ======= ======= ======= ========
324Filesystem block size 1kB 2kB 4kB 8kB
325===================== ======= ======= ======= ========
326File size limit 16GB 256GB 2048GB 2048GB
327Filesystem size limit 2047GB 8192GB 16384GB 32768GB
328===================== ======= ======= ======= ========
329
330There is a 2.4 kernel limit of 2048GB for a single block device, so no
331filesystem larger than that can be created at this time. There is also
332an upper limit on the block size imposed by the page size of the kernel,
333so 8kB blocks are only allowed on Alpha systems (and other architectures
334which support larger pages).
335
336There is an upper limit of 32000 subdirectories in a single directory.
337
338There is a "soft" upper limit of about 10-15k files in a single directory
339with the current linear linked-list directory implementation. This limit
340stems from performance problems when creating and deleting (and also
341finding) files in such large directories. Using a hashed directory index
342(under development) allows 100k-1M+ files in a single directory without
343performance problems (although RAM size becomes an issue at this point).
344
345The (meaningless) absolute upper limit of files in a single directory
346(imposed by the file size, the realistic limit is obviously much less)
347is over 130 trillion files. It would be higher except there are not
348enough 4-character names to make up unique directory entries, so they
349have to be 8 character filenames, even then we are fairly close to
350running out of unique filenames.
351
352Journaling
353----------
354
355A journaling extension to the ext2 code has been developed by Stephen
356Tweedie. It avoids the risks of metadata corruption and the need to
357wait for e2fsck to complete after a crash, without requiring a change
358to the on-disk ext2 layout. In a nutshell, the journal is a regular
359file which stores whole metadata (and optionally data) blocks that have
360been modified, prior to writing them into the filesystem. This means
361it is possible to add a journal to an existing ext2 filesystem without
362the need for data conversion.
363
364When changes to the filesystem (e.g. a file is renamed) they are stored in
365a transaction in the journal and can either be complete or incomplete at
366the time of a crash. If a transaction is complete at the time of a crash
367(or in the normal case where the system does not crash), then any blocks
368in that transaction are guaranteed to represent a valid filesystem state,
369and are copied into the filesystem. If a transaction is incomplete at
370the time of the crash, then there is no guarantee of consistency for
371the blocks in that transaction so they are discarded (which means any
372filesystem changes they represent are also lost).
373Check Documentation/filesystems/ext4/ if you want to read more about
374ext4 and journaling.
375
376References
377==========
378
379======================= ===============================================
380The kernel source file:/usr/src/linux/fs/ext2/
381e2fsprogs (e2fsck) http://e2fsprogs.sourceforge.net/
382Design & Implementation http://e2fsprogs.sourceforge.net/ext2intro.html
383Journaling (ext3) ftp://ftp.uk.linux.org/pub/linux/sct/fs/jfs/
384Filesystem Resizing http://ext2resize.sourceforge.net/
385Compression [1]_ http://e2compr.sourceforge.net/
386======================= ===============================================
387
388Implementations for:
389
390======================= ===========================================================
391Windows 95/98/NT/2000 http://www.chrysocome.net/explore2fs
392Windows 95 [1]_ http://www.yipton.net/content.html#FSDEXT2
393DOS client [1]_ ftp://metalab.unc.edu/pub/Linux/system/filesystems/ext2/
394OS/2 [2]_ ftp://metalab.unc.edu/pub/Linux/system/filesystems/ext2/
395RISC OS client http://www.esw-heim.tu-clausthal.de/~marco/smorbrod/IscaFS/
396======================= ===========================================================
397
398.. [1] no longer actively developed/supported (as of Apr 2001)
399.. [2] no longer actively developed/supported (as of Mar 2009)