diff options
author | Adrian Bunk <bunk@stusta.de> | 2005-11-08 21:34:59 -0800 |
---|---|---|
committer | Linus Torvalds <torvalds@g5.osdl.org> | 2005-11-09 07:55:58 -0800 |
commit | 55e64b3003f03355d8e950af58d456b6314a5a62 (patch) | |
tree | 6a51b651a032abc82f880fd275e5272e0c322813 /fs/ext2 | |
parent | 2860b733f114e088b56c20da6145902c16b79a44 (diff) | |
download | op-kernel-dev-55e64b3003f03355d8e950af58d456b6314a5a62.zip op-kernel-dev-55e64b3003f03355d8e950af58d456b6314a5a62.tar.gz |
[PATCH] ext2: remove the ancient CHANGES file
This patch removes an ancient changelog file.
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'fs/ext2')
-rw-r--r-- | fs/ext2/CHANGES | 157 |
1 files changed, 0 insertions, 157 deletions
diff --git a/fs/ext2/CHANGES b/fs/ext2/CHANGES deleted file mode 100644 index aa5aaf0..0000000 --- a/fs/ext2/CHANGES +++ /dev/null @@ -1,157 +0,0 @@ -Changes from version 0.5a to version 0.5b -========================================= - - Now that we have sysctl(), the immutable flag cannot be changed when - the system is running at security level > 0. - - Some cleanups in the code. - - More consistency checks on directories. - - The ext2.diff patch from Tom May <ftom@netcom.com> has been - integrated. This patch replaces expensive "/" and "%" with - cheap ">>" and "&" where possible. - -Changes from version 0.5 to version 0.5a -======================================== - - Zero the partial block following the end of the file when a file - is truncated. - - Dates updated in the copyright. - - More checks when the filesystem is mounted: the count of blocks, - fragments, and inodes per group is checked against the block size. - - The buffers used by the error routines are now static variables, to - avoid using space on the kernel stack, as requested by Linus. - - Some cleanups in the error messages (some versions of syslog contain - a bug which truncates an error message if it contains '\n'). - - Check that no data can be written to a file past the 2GB limit. - - The famous readdir() bug has been fixed by Stephen Tweedie. - - Added a revision level in the superblock. - - Full support for O_SYNC flag of the open system call. - - New mount options: `resuid=#uid' and `resgid=#gid'. `resuid' causes - ext2fs to consider user #uid like root for the reserved blocks. - `resgid' acts the same way with group #gid. New fields in the - superblock contain default values for resuid and resgid and can - be modified by tune2fs. - Idea comes from Rene Cougnenc <cougnenc@renux.frmug.fr.net>. - - New mount options: `bsddf' and `minixdf'. `bsddf' causes ext2fs - to remove the blocks used for FS structures from the total block - count in statfs. With `minixdf', ext2fs mimics Minix behavior - in statfs (i.e. it returns the total number of blocks on the - partition). This is intended to make bde happy :-) - - New file attributes: - - Immutable files cannot be modified. Data cannot be written to - these files. They cannot be removed, renamed and new links cannot - be created. Even root cannot modify the files. He has to remove - the immutable attribute first. - - Append-only files: can only be written in append-mode when writing. - They cannot be removed, renamed and new links cannot be created. - Note: files may only be added to an append-only directory. - - No-dump files: the attribute is not used by the kernel. My port - of dump uses it to avoid backing up files which are not important. - - New check in ext2_check_dir_entry: the inode number is checked. - - Support for big file systems: the copy of the FS descriptor is now - dynamically allocated (previous versions used a fixed size array). - This allows to mount 2GB+ FS. - - Reorganization of the ext2_inode structure to allow other operating - systems to create specific fields if they use ext2fs as their native - file system. Currently, ext2fs is only implemented in Linux but - will soon be part of Gnu Hurd and of Masix. - -Changes from version 0.4b to version 0.5 -======================================== - - New superblock fields: s_lastcheck and s_checkinterval added - by Uwe Ohse <uwe@tirka.gun.de> to implement timedependent checks - of the file system - - Real random numbers for secure rm added by Pierre del Perugia - <delperug@gla.ecoledoc.ibp.fr> - - The mount warnings related to the state of a fs are not printed - if the fs is mounted read-only, idea by Nick Holloway - <alfie@dcs.warwick.ac.uk> - -Changes from version 0.4a to version 0.4b -========================================= - - Copyrights changed to include the name of my laboratory. - - Clean up of balloc.c and ialloc.c. - - More consistency checks. - - Block preallocation added by Stephen Tweedie. - - Direct reads of directories disallowed. - - Readahead implemented in readdir by Stephen Tweedie. - - Bugs in block and inodes allocation fixed. - - Readahead implemented in ext2_find_entry by Chip Salzenberg. - - New mount options: - `check=none|normal|strict' - `debug' - `errors=continue|remount-ro|panic' - `grpid', `bsdgroups' - `nocheck' - `nogrpid', `sysvgroups' - - truncate() now tries to deallocate contiguous blocks in a single call - to ext2_free_blocks(). - - lots of cosmetic changes. - -Changes from version 0.4 to version 0.4a -======================================== - - the `sync' option support is now complete. Version 0.4 was not - supporting it when truncating a file. I have tested the synchronous - writes and they work but they make the system very slow :-( I have - to work again on this to make it faster. - - when detecting an error on a mounted filesystem, version 0.4 used - to try to write a flag in the super block even if the filesystem had - been mounted read-only. This is fixed. - - the `sb=#' option now causes the kernel code to use the filesystem - descriptors located at block #+1. Version 0.4 used the superblock - backup located at block # but used the main copy of the descriptors. - - a new file attribute `S' is supported. This attribute causes - synchronous writes but is applied to a file not to the entire file - system (thanks to Michael Kraehe <kraehe@bakunin.north.de> for - suggesting it). - - the directory cache is inhibited by default. The cache management - code seems to be buggy and I have to look at it carefully before - using it again. - - deleting a file with the `s' attribute (secure deletion) causes its - blocks to be overwritten with random values not with zeros (thanks to - Michael A. Griffith <grif@cs.ucr.edu> for suggesting it). - - lots of cosmetic changes have been made. - -Changes from version 0.3 to version 0.4 -======================================= - - Three new mount options are supported: `check', `sync' and `sb=#'. - `check' tells the kernel code to make more consistency checks - when the file system is mounted. Currently, the kernel code checks - that the blocks and inodes bitmaps are consistent with the free - blocks and inodes counts. More checks will be added in future - releases. - `sync' tells the kernel code to use synchronous writes when updating - an inode, a bitmap, a directory entry or an indirect block. This - can make the file system much slower but can be a big win for files - recovery in case of a crash (and we can now say to the BSD folks - that Linux also supports synchronous updates :-). - `sb=#' tells the kernel code to use an alternate super block instead - of its master copy. `#' is the number of the block (counted in - 1024 bytes blocks) which contains the alternate super block. - An ext2 file system typically contains backups of the super block - at blocks 8193, 16385, and so on. - - I have change the meaning of the valid flag used by e2fsck. it - now contains the state of the file system. If the kernel code - detects an inconsistency while the file system is mounted, it flags - it as erroneous and e2fsck will detect that on next run. - - The super block now contains a mount counter. This counter is - incremented each time the file system is mounted read/write. When - this counter becomes bigger than a maximal mount counts (also stored - in the super block), e2fsck checks the file system, even if it had - been unmounted cleanly, and resets this counter to 0. - - File attributes are now supported. One can associate a set of - attributes to a file. Three attributes are defined: - `c': the file is marked for automatic compression, - `s': the file is marked for secure deletion: when the file is - deleted, its blocks are zeroed and written back to the disk, - `u': the file is marked for undeletion: when the file is deleted, - its contents are saved to allow a future undeletion. - Currently, only the `s' attribute is implemented in the kernel - code. Support for the other attributes will be added in a future - release. - - a few bugs related to times updates have been fixed by Bruce - Evans and me. - - a bug related to the links count of deleted inodes has been fixed. - Previous versions used to keep the links count set to 1 when a file - was deleted. The new version now sets links_count to 0 when deleting - the last link. - - a race condition when deallocating an inode has been fixed by - Stephen Tweedie. - |