ext4.txt 26 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597
  1. Ext4 Filesystem
  2. ===============
  3. Ext4 is an an advanced level of the ext3 filesystem which incorporates
  4. scalability and reliability enhancements for supporting large filesystems
  5. (64 bit) in keeping with increasing disk capacities and state-of-the-art
  6. feature requirements.
  7. Mailing list: linux-ext4@vger.kernel.org
  8. Web site: http://ext4.wiki.kernel.org
  9. 1. Quick usage instructions:
  10. ===========================
  11. Note: More extensive information for getting started with ext4 can be
  12. found at the ext4 wiki site at the URL:
  13. http://ext4.wiki.kernel.org/index.php/Ext4_Howto
  14. - Compile and install the latest version of e2fsprogs (as of this
  15. writing version 1.41.3) from:
  16. http://sourceforge.net/project/showfiles.php?group_id=2406
  17. or
  18. ftp://ftp.kernel.org/pub/linux/kernel/people/tytso/e2fsprogs/
  19. or grab the latest git repository from:
  20. git://git.kernel.org/pub/scm/fs/ext2/e2fsprogs.git
  21. - Note that it is highly important to install the mke2fs.conf file
  22. that comes with the e2fsprogs 1.41.x sources in /etc/mke2fs.conf. If
  23. you have edited the /etc/mke2fs.conf file installed on your system,
  24. you will need to merge your changes with the version from e2fsprogs
  25. 1.41.x.
  26. - Create a new filesystem using the ext4 filesystem type:
  27. # mke2fs -t ext4 /dev/hda1
  28. Or to configure an existing ext3 filesystem to support extents:
  29. # tune2fs -O extents /dev/hda1
  30. If the filesystem was created with 128 byte inodes, it can be
  31. converted to use 256 byte for greater efficiency via:
  32. # tune2fs -I 256 /dev/hda1
  33. (Note: we currently do not have tools to convert an ext4
  34. filesystem back to ext3; so please do not do try this on production
  35. filesystems.)
  36. - Mounting:
  37. # mount -t ext4 /dev/hda1 /wherever
  38. - When comparing performance with other filesystems, it's always
  39. important to try multiple workloads; very often a subtle change in a
  40. workload parameter can completely change the ranking of which
  41. filesystems do well compared to others. When comparing versus ext3,
  42. note that ext4 enables write barriers by default, while ext3 does
  43. not enable write barriers by default. So it is useful to use
  44. explicitly specify whether barriers are enabled or not when via the
  45. '-o barriers=[0|1]' mount option for both ext3 and ext4 filesystems
  46. for a fair comparison. When tuning ext3 for best benchmark numbers,
  47. it is often worthwhile to try changing the data journaling mode; '-o
  48. data=writeback' can be faster for some workloads. (Note however that
  49. running mounted with data=writeback can potentially leave stale data
  50. exposed in recently written files in case of an unclean shutdown,
  51. which could be a security exposure in some situations.) Configuring
  52. the filesystem with a large journal can also be helpful for
  53. metadata-intensive workloads.
  54. 2. Features
  55. ===========
  56. 2.1 Currently available
  57. * ability to use filesystems > 16TB (e2fsprogs support not available yet)
  58. * extent format reduces metadata overhead (RAM, IO for access, transactions)
  59. * extent format more robust in face of on-disk corruption due to magics,
  60. * internal redundancy in tree
  61. * improved file allocation (multi-block alloc)
  62. * lift 32000 subdirectory limit imposed by i_links_count[1]
  63. * nsec timestamps for mtime, atime, ctime, create time
  64. * inode version field on disk (NFSv4, Lustre)
  65. * reduced e2fsck time via uninit_bg feature
  66. * journal checksumming for robustness, performance
  67. * persistent file preallocation (e.g for streaming media, databases)
  68. * ability to pack bitmaps and inode tables into larger virtual groups via the
  69. flex_bg feature
  70. * large file support
  71. * Inode allocation using large virtual block groups via flex_bg
  72. * delayed allocation
  73. * large block (up to pagesize) support
  74. * efficient new ordered mode in JBD2 and ext4(avoid using buffer head to force
  75. the ordering)
  76. [1] Filesystems with a block size of 1k may see a limit imposed by the
  77. directory hash tree having a maximum depth of two.
  78. 2.2 Candidate features for future inclusion
  79. * Online defrag (patches available but not well tested)
  80. * reduced mke2fs time via lazy itable initialization in conjunction with
  81. the uninit_bg feature (capability to do this is available in e2fsprogs
  82. but a kernel thread to do lazy zeroing of unused inode table blocks
  83. after filesystem is first mounted is required for safety)
  84. There are several others under discussion, whether they all make it in is
  85. partly a function of how much time everyone has to work on them. Features like
  86. metadata checksumming have been discussed and planned for a bit but no patches
  87. exist yet so I'm not sure they're in the near-term roadmap.
  88. The big performance win will come with mballoc, delalloc and flex_bg
  89. grouping of bitmaps and inode tables. Some test results available here:
  90. - http://www.bullopensource.org/ext4/20080818-ffsb/ffsb-write-2.6.27-rc1.html
  91. - http://www.bullopensource.org/ext4/20080818-ffsb/ffsb-readwrite-2.6.27-rc1.html
  92. 3. Options
  93. ==========
  94. When mounting an ext4 filesystem, the following option are accepted:
  95. (*) == default
  96. ro Mount filesystem read only. Note that ext4 will
  97. replay the journal (and thus write to the
  98. partition) even when mounted "read only". The
  99. mount options "ro,noload" can be used to prevent
  100. writes to the filesystem.
  101. journal_checksum Enable checksumming of the journal transactions.
  102. This will allow the recovery code in e2fsck and the
  103. kernel to detect corruption in the kernel. It is a
  104. compatible change and will be ignored by older kernels.
  105. journal_async_commit Commit block can be written to disk without waiting
  106. for descriptor blocks. If enabled older kernels cannot
  107. mount the device. This will enable 'journal_checksum'
  108. internally.
  109. journal_dev=devnum When the external journal device's major/minor numbers
  110. have changed, this option allows the user to specify
  111. the new journal location. The journal device is
  112. identified through its new major/minor numbers encoded
  113. in devnum.
  114. norecovery Don't load the journal on mounting. Note that
  115. noload if the filesystem was not unmounted cleanly,
  116. skipping the journal replay will lead to the
  117. filesystem containing inconsistencies that can
  118. lead to any number of problems.
  119. data=journal All data are committed into the journal prior to being
  120. written into the main file system. Enabling
  121. this mode will disable delayed allocation and
  122. O_DIRECT support.
  123. data=ordered (*) All data are forced directly out to the main file
  124. system prior to its metadata being committed to the
  125. journal.
  126. data=writeback Data ordering is not preserved, data may be written
  127. into the main file system after its metadata has been
  128. committed to the journal.
  129. commit=nrsec (*) Ext4 can be told to sync all its data and metadata
  130. every 'nrsec' seconds. The default value is 5 seconds.
  131. This means that if you lose your power, you will lose
  132. as much as the latest 5 seconds of work (your
  133. filesystem will not be damaged though, thanks to the
  134. journaling). This default value (or any low value)
  135. will hurt performance, but it's good for data-safety.
  136. Setting it to 0 will have the same effect as leaving
  137. it at the default (5 seconds).
  138. Setting it to very large values will improve
  139. performance.
  140. barrier=<0|1(*)> This enables/disables the use of write barriers in
  141. barrier(*) the jbd code. barrier=0 disables, barrier=1 enables.
  142. nobarrier This also requires an IO stack which can support
  143. barriers, and if jbd gets an error on a barrier
  144. write, it will disable again with a warning.
  145. Write barriers enforce proper on-disk ordering
  146. of journal commits, making volatile disk write caches
  147. safe to use, at some performance penalty. If
  148. your disks are battery-backed in one way or another,
  149. disabling barriers may safely improve performance.
  150. The mount options "barrier" and "nobarrier" can
  151. also be used to enable or disable barriers, for
  152. consistency with other ext4 mount options.
  153. inode_readahead_blks=n This tuning parameter controls the maximum
  154. number of inode table blocks that ext4's inode
  155. table readahead algorithm will pre-read into
  156. the buffer cache. The default value is 32 blocks.
  157. nouser_xattr Disables Extended User Attributes. If you have extended
  158. attribute support enabled in the kernel configuration
  159. (CONFIG_EXT4_FS_XATTR), extended attribute support
  160. is enabled by default on mount. See the attr(5) manual
  161. page and http://acl.bestbits.at/ for more information
  162. about extended attributes.
  163. noacl This option disables POSIX Access Control List
  164. support. If ACL support is enabled in the kernel
  165. configuration (CONFIG_EXT4_FS_POSIX_ACL), ACL is
  166. enabled by default on mount. See the acl(5) manual
  167. page and http://acl.bestbits.at/ for more information
  168. about acl.
  169. bsddf (*) Make 'df' act like BSD.
  170. minixdf Make 'df' act like Minix.
  171. debug Extra debugging information is sent to syslog.
  172. abort Simulate the effects of calling ext4_abort() for
  173. debugging purposes. This is normally used while
  174. remounting a filesystem which is already mounted.
  175. errors=remount-ro Remount the filesystem read-only on an error.
  176. errors=continue Keep going on a filesystem error.
  177. errors=panic Panic and halt the machine if an error occurs.
  178. (These mount options override the errors behavior
  179. specified in the superblock, which can be configured
  180. using tune2fs)
  181. data_err=ignore(*) Just print an error message if an error occurs
  182. in a file data buffer in ordered mode.
  183. data_err=abort Abort the journal if an error occurs in a file
  184. data buffer in ordered mode.
  185. grpid Give objects the same group ID as their creator.
  186. bsdgroups
  187. nogrpid (*) New objects have the group ID of their creator.
  188. sysvgroups
  189. resgid=n The group ID which may use the reserved blocks.
  190. resuid=n The user ID which may use the reserved blocks.
  191. sb=n Use alternate superblock at this location.
  192. quota These options are ignored by the filesystem. They
  193. noquota are used only by quota tools to recognize volumes
  194. grpquota where quota should be turned on. See documentation
  195. usrquota in the quota-tools package for more details
  196. (http://sourceforge.net/projects/linuxquota).
  197. jqfmt=<quota type> These options tell filesystem details about quota
  198. usrjquota=<file> so that quota information can be properly updated
  199. grpjquota=<file> during journal replay. They replace the above
  200. quota options. See documentation in the quota-tools
  201. package for more details
  202. (http://sourceforge.net/projects/linuxquota).
  203. stripe=n Number of filesystem blocks that mballoc will try
  204. to use for allocation size and alignment. For RAID5/6
  205. systems this should be the number of data
  206. disks * RAID chunk size in file system blocks.
  207. delalloc (*) Defer block allocation until just before ext4
  208. writes out the block(s) in question. This
  209. allows ext4 to better allocation decisions
  210. more efficiently.
  211. nodelalloc Disable delayed allocation. Blocks are allocated
  212. when the data is copied from userspace to the
  213. page cache, either via the write(2) system call
  214. or when an mmap'ed page which was previously
  215. unallocated is written for the first time.
  216. max_batch_time=usec Maximum amount of time ext4 should wait for
  217. additional filesystem operations to be batch
  218. together with a synchronous write operation.
  219. Since a synchronous write operation is going to
  220. force a commit and then a wait for the I/O
  221. complete, it doesn't cost much, and can be a
  222. huge throughput win, we wait for a small amount
  223. of time to see if any other transactions can
  224. piggyback on the synchronous write. The
  225. algorithm used is designed to automatically tune
  226. for the speed of the disk, by measuring the
  227. amount of time (on average) that it takes to
  228. finish committing a transaction. Call this time
  229. the "commit time". If the time that the
  230. transaction has been running is less than the
  231. commit time, ext4 will try sleeping for the
  232. commit time to see if other operations will join
  233. the transaction. The commit time is capped by
  234. the max_batch_time, which defaults to 15000us
  235. (15ms). This optimization can be turned off
  236. entirely by setting max_batch_time to 0.
  237. min_batch_time=usec This parameter sets the commit time (as
  238. described above) to be at least min_batch_time.
  239. It defaults to zero microseconds. Increasing
  240. this parameter may improve the throughput of
  241. multi-threaded, synchronous workloads on very
  242. fast disks, at the cost of increasing latency.
  243. journal_ioprio=prio The I/O priority (from 0 to 7, where 0 is the
  244. highest priority) which should be used for I/O
  245. operations submitted by kjournald2 during a
  246. commit operation. This defaults to 3, which is
  247. a slightly higher priority than the default I/O
  248. priority.
  249. auto_da_alloc(*) Many broken applications don't use fsync() when
  250. noauto_da_alloc replacing existing files via patterns such as
  251. fd = open("foo.new")/write(fd,..)/close(fd)/
  252. rename("foo.new", "foo"), or worse yet,
  253. fd = open("foo", O_TRUNC)/write(fd,..)/close(fd).
  254. If auto_da_alloc is enabled, ext4 will detect
  255. the replace-via-rename and replace-via-truncate
  256. patterns and force that any delayed allocation
  257. blocks are allocated such that at the next
  258. journal commit, in the default data=ordered
  259. mode, the data blocks of the new file are forced
  260. to disk before the rename() operation is
  261. committed. This provides roughly the same level
  262. of guarantees as ext3, and avoids the
  263. "zero-length" problem that can happen when a
  264. system crashes before the delayed allocation
  265. blocks are forced to disk.
  266. noinit_itable Do not initialize any uninitialized inode table
  267. blocks in the background. This feature may be
  268. used by installation CD's so that the install
  269. process can complete as quickly as possible; the
  270. inode table initialization process would then be
  271. deferred until the next time the file system
  272. is unmounted.
  273. init_itable=n The lazy itable init code will wait n times the
  274. number of milliseconds it took to zero out the
  275. previous block group's inode table. This
  276. minimizes the impact on the system performance
  277. while file system's inode table is being initialized.
  278. discard Controls whether ext4 should issue discard/TRIM
  279. nodiscard(*) commands to the underlying block device when
  280. blocks are freed. This is useful for SSD devices
  281. and sparse/thinly-provisioned LUNs, but it is off
  282. by default until sufficient testing has been done.
  283. nouid32 Disables 32-bit UIDs and GIDs. This is for
  284. interoperability with older kernels which only
  285. store and expect 16-bit values.
  286. block_validity This options allows to enables/disables the in-kernel
  287. noblock_validity facility for tracking filesystem metadata blocks
  288. within internal data structures. This allows multi-
  289. block allocator and other routines to quickly locate
  290. extents which might overlap with filesystem metadata
  291. blocks. This option is intended for debugging
  292. purposes and since it negatively affects the
  293. performance, it is off by default.
  294. dioread_lock Controls whether or not ext4 should use the DIO read
  295. dioread_nolock locking. If the dioread_nolock option is specified
  296. ext4 will allocate uninitialized extent before buffer
  297. write and convert the extent to initialized after IO
  298. completes. This approach allows ext4 code to avoid
  299. using inode mutex, which improves scalability on high
  300. speed storages. However this does not work with
  301. data journaling and dioread_nolock option will be
  302. ignored with kernel warning. Note that dioread_nolock
  303. code path is only used for extent-based files.
  304. Because of the restrictions this options comprises
  305. it is off by default (e.g. dioread_lock).
  306. i_version Enable 64-bit inode version support. This option is
  307. off by default.
  308. Data Mode
  309. =========
  310. There are 3 different data modes:
  311. * writeback mode
  312. In data=writeback mode, ext4 does not journal data at all. This mode provides
  313. a similar level of journaling as that of XFS, JFS, and ReiserFS in its default
  314. mode - metadata journaling. A crash+recovery can cause incorrect data to
  315. appear in files which were written shortly before the crash. This mode will
  316. typically provide the best ext4 performance.
  317. * ordered mode
  318. In data=ordered mode, ext4 only officially journals metadata, but it logically
  319. groups metadata information related to data changes with the data blocks into a
  320. single unit called a transaction. When it's time to write the new metadata
  321. out to disk, the associated data blocks are written first. In general,
  322. this mode performs slightly slower than writeback but significantly faster than journal mode.
  323. * journal mode
  324. data=journal mode provides full data and metadata journaling. All new data is
  325. written to the journal first, and then to its final location.
  326. In the event of a crash, the journal can be replayed, bringing both data and
  327. metadata into a consistent state. This mode is the slowest except when data
  328. needs to be read from and written to disk at the same time where it
  329. outperforms all others modes. Enabling this mode will disable delayed
  330. allocation and O_DIRECT support.
  331. /proc entries
  332. =============
  333. Information about mounted ext4 file systems can be found in
  334. /proc/fs/ext4. Each mounted filesystem will have a directory in
  335. /proc/fs/ext4 based on its device name (i.e., /proc/fs/ext4/hdc or
  336. /proc/fs/ext4/dm-0). The files in each per-device directory are shown
  337. in table below.
  338. Files in /proc/fs/ext4/<devname>
  339. ..............................................................................
  340. File Content
  341. mb_groups details of multiblock allocator buddy cache of free blocks
  342. ..............................................................................
  343. /sys entries
  344. ============
  345. Information about mounted ext4 file systems can be found in
  346. /sys/fs/ext4. Each mounted filesystem will have a directory in
  347. /sys/fs/ext4 based on its device name (i.e., /sys/fs/ext4/hdc or
  348. /sys/fs/ext4/dm-0). The files in each per-device directory are shown
  349. in table below.
  350. Files in /sys/fs/ext4/<devname>
  351. (see also Documentation/ABI/testing/sysfs-fs-ext4)
  352. ..............................................................................
  353. File Content
  354. delayed_allocation_blocks This file is read-only and shows the number of
  355. blocks that are dirty in the page cache, but
  356. which do not have their location in the
  357. filesystem allocated yet.
  358. inode_goal Tuning parameter which (if non-zero) controls
  359. the goal inode used by the inode allocator in
  360. preference to all other allocation heuristics.
  361. This is intended for debugging use only, and
  362. should be 0 on production systems.
  363. inode_readahead_blks Tuning parameter which controls the maximum
  364. number of inode table blocks that ext4's inode
  365. table readahead algorithm will pre-read into
  366. the buffer cache
  367. lifetime_write_kbytes This file is read-only and shows the number of
  368. kilobytes of data that have been written to this
  369. filesystem since it was created.
  370. max_writeback_mb_bump The maximum number of megabytes the writeback
  371. code will try to write out before move on to
  372. another inode.
  373. mb_group_prealloc The multiblock allocator will round up allocation
  374. requests to a multiple of this tuning parameter if
  375. the stripe size is not set in the ext4 superblock
  376. mb_max_to_scan The maximum number of extents the multiblock
  377. allocator will search to find the best extent
  378. mb_min_to_scan The minimum number of extents the multiblock
  379. allocator will search to find the best extent
  380. mb_order2_req Tuning parameter which controls the minimum size
  381. for requests (as a power of 2) where the buddy
  382. cache is used
  383. mb_stats Controls whether the multiblock allocator should
  384. collect statistics, which are shown during the
  385. unmount. 1 means to collect statistics, 0 means
  386. not to collect statistics
  387. mb_stream_req Files which have fewer blocks than this tunable
  388. parameter will have their blocks allocated out
  389. of a block group specific preallocation pool, so
  390. that small files are packed closely together.
  391. Each large file will have its blocks allocated
  392. out of its own unique preallocation pool.
  393. session_write_kbytes This file is read-only and shows the number of
  394. kilobytes of data that have been written to this
  395. filesystem since it was mounted.
  396. ..............................................................................
  397. Ioctls
  398. ======
  399. There is some Ext4 specific functionality which can be accessed by applications
  400. through the system call interfaces. The list of all Ext4 specific ioctls are
  401. shown in the table below.
  402. Table of Ext4 specific ioctls
  403. ..............................................................................
  404. Ioctl Description
  405. EXT4_IOC_GETFLAGS Get additional attributes associated with inode.
  406. The ioctl argument is an integer bitfield, with
  407. bit values described in ext4.h. This ioctl is an
  408. alias for FS_IOC_GETFLAGS.
  409. EXT4_IOC_SETFLAGS Set additional attributes associated with inode.
  410. The ioctl argument is an integer bitfield, with
  411. bit values described in ext4.h. This ioctl is an
  412. alias for FS_IOC_SETFLAGS.
  413. EXT4_IOC_GETVERSION
  414. EXT4_IOC_GETVERSION_OLD
  415. Get the inode i_generation number stored for
  416. each inode. The i_generation number is normally
  417. changed only when new inode is created and it is
  418. particularly useful for network filesystems. The
  419. '_OLD' version of this ioctl is an alias for
  420. FS_IOC_GETVERSION.
  421. EXT4_IOC_SETVERSION
  422. EXT4_IOC_SETVERSION_OLD
  423. Set the inode i_generation number stored for
  424. each inode. The '_OLD' version of this ioctl
  425. is an alias for FS_IOC_SETVERSION.
  426. EXT4_IOC_GROUP_EXTEND This ioctl has the same purpose as the resize
  427. mount option. It allows to resize filesystem
  428. to the end of the last existing block group,
  429. further resize has to be done with resize2fs,
  430. either online, or offline. The argument points
  431. to the unsigned logn number representing the
  432. filesystem new block count.
  433. EXT4_IOC_MOVE_EXT Move the block extents from orig_fd (the one
  434. this ioctl is pointing to) to the donor_fd (the
  435. one specified in move_extent structure passed
  436. as an argument to this ioctl). Then, exchange
  437. inode metadata between orig_fd and donor_fd.
  438. This is especially useful for online
  439. defragmentation, because the allocator has the
  440. opportunity to allocate moved blocks better,
  441. ideally into one contiguous extent.
  442. EXT4_IOC_GROUP_ADD Add a new group descriptor to an existing or
  443. new group descriptor block. The new group
  444. descriptor is described by ext4_new_group_input
  445. structure, which is passed as an argument to
  446. this ioctl. This is especially useful in
  447. conjunction with EXT4_IOC_GROUP_EXTEND,
  448. which allows online resize of the filesystem
  449. to the end of the last existing block group.
  450. Those two ioctls combined is used in userspace
  451. online resize tool (e.g. resize2fs).
  452. EXT4_IOC_MIGRATE This ioctl operates on the filesystem itself.
  453. It converts (migrates) ext3 indirect block mapped
  454. inode to ext4 extent mapped inode by walking
  455. through indirect block mapping of the original
  456. inode and converting contiguous block ranges
  457. into ext4 extents of the temporary inode. Then,
  458. inodes are swapped. This ioctl might help, when
  459. migrating from ext3 to ext4 filesystem, however
  460. suggestion is to create fresh ext4 filesystem
  461. and copy data from the backup. Note, that
  462. filesystem has to support extents for this ioctl
  463. to work.
  464. EXT4_IOC_ALLOC_DA_BLKS Force all of the delay allocated blocks to be
  465. allocated to preserve application-expected ext3
  466. behaviour. Note that this will also start
  467. triggering a write of the data blocks, but this
  468. behaviour may change in the future as it is
  469. not necessary and has been done this way only
  470. for sake of simplicity.
  471. EXT4_IOC_RESIZE_FS Resize the filesystem to a new size. The number
  472. of blocks of resized filesystem is passed in via
  473. 64 bit integer argument. The kernel allocates
  474. bitmaps and inode table, the userspace tool thus
  475. just passes the new number of blocks.
  476. ..............................................................................
  477. References
  478. ==========
  479. kernel source: <file:fs/ext4/>
  480. <file:fs/jbd2/>
  481. programs: http://e2fsprogs.sourceforge.net/
  482. useful links: http://fedoraproject.org/wiki/ext3-devel
  483. http://www.bullopensource.org/ext4/
  484. http://ext4.wiki.kernel.org/index.php/Main_Page
  485. http://fedoraproject.org/wiki/Features/Ext4