E2fsprogs 1.28 (August 31, 2002) ================================ Add support for the Hashed-Tree Directory Indexing to e2fsck. Support for setting the htree flag is not included yet, although it can be manually turned on via the debugfs program. Clarified e2fsck error message which is printed when it cannot find sufficient contiguous block when relcating filesystem metadata. Added support for building an EVMS plugin module for ext2/3. This module is substantially the same as the EVMS module shipping with EVMS 1.1, with one or two bugfixes. E2fsprogs can also build this plugin for use with EVMS 1.0 (which did not include the ext2 plugin module), if the configure --enable-old-evms flag is given. Fsck will search through EVMS volumes when trying to resolve filesystem specifications such as LABEL=xxx or UUID=xxx. Added a new utility program, /sbin/findfs, which will return filesystem specifications such as LABEL=xxx or UUID=xxx, and prints the device name. Update and clarified various man pages. (addresses Debian Bug #145044, #146437, #131350, #151990, #144621, #141938) If there are no filesystems specified on fsck's command line, fsck now treat that as if the -As options were given. Previously it would simply do nothing. (Addresses Debian Bug #153102) Mke2fs no longer treats a failure to be able to clear the MD signature at the end of the filesystem as a fatal error. (Addresses Debian Bug #155007) The e2p library functions (which are used by lsattr and chattr) now double check to make sure the file is a regular file or directory before attempting to use the ext2 ioctls. Some device drivers unfortunately respond to the ext2 ioctl's with unknown behaviour. (Addresses Debian Bug #152029). The extended attribute handling has been updated to correspond with the latest V2 bestbits ACL code. Fixed bug in e2fsck which caused it to not clear the dtime field when processing truncated inodes on the orphan list. This could cause data loss(!) if a filesystem is rebooted before a truncate has been committed. E2fsck now uses red/black trees in pass1b, which removes some O(n**2) algorithms. This makes e2fsck much faster in the case of severely corrupted filesystems where a large number of blocks are claimed by a large number of inodes. (Thanks to the 2.5 IDE device driver for inspiring this work.) Resize2fs has been significantly sped up when shrinking and expanding a filesystem by a very small number of blocks (for example, when EVMS is converting a partition to be an EVMS legacy volume). Added a new option to e2fsck, -D, which will optimize or compress all of the directories in the filesystem. E2fsck now catches SIGINT and SIGTERM to make sure it will can properly clean up and only exit at safe points. Fsck will pass SIGINT/SIGTERM to its child processes, and wait until they have all exited before it exits. The uuid parsing code in the uuid library now properly complains when an illegally formated uuid is presented to it. (Addresses Debian bug #152891) Restrict use of the 2.4 setrlimit ABI f*ckup to kernels between 2.4.10 and 2.4.17, since the workaround can cause problems when using a 2.4 kernel with an old version of glibc built with the 2.2 headers. Fixed a bug in mke2fs where it wasn't properly clearing the initial superblock used by other filesystems. (Addresses Debian bug #147256.) Added support for the synchronous directory feature written by Andrew Morton. The debugfs program can delete directories using the rmdir command. Add support for 8k and 16k filesystems (for systems with page sizes that are greater or equal to 8k or 16k, respectively). Note that these filesystems can not be mounted on x86 systems, or other systems with only 4k page sizes, due to limitations in the current Linux VM code. Resize2fs requires that the filesystem state be valid and have no errors; otherwise, e2fsck -f must be run first. (Previously it simply required that the last fsck time be greater than the last mount time.) Configure now defaults the man pages directory to /usr/share/man on Linux systems. E2fsck now offers to truncate inodes which contain too many blocks (so that i_blocks would overflow. Also fixed handling of large sparse files. E2fsck now more completely checks for symlink validity, including requiring NULL termination and length checks. E2fsck will offer to try forcing a disk write to remap a bad block after finding a read error when reading a filesystem metadata block. Fixed a bug in debugfs which caused the -b and -s options to crash debugfs, as well as breaking the testb, setb, and clearb functions. Added a bmap command to debugfs which calculates the logical to physical block mapping for a particular inode. Fixed a bug in code which checked to see if a device was mounted which sometimes (rarely) failed in the case of a plain file. Fixed a bug in resize2fs where when it reported an error, it would print a message erroneously indicating that the filesystem had been resized before it aborted. When resizing a plain file which is smaller than the requested size, resize2fs will attempt to extended the file so that programs like e2fsck will not complain that the file is too small. Resize2fs will print the actual new size of the filesystem when it is finished resizing. Fixed a bug in debugfs where "ls -l" would report incorrectl file type information on big-endian systems. Programmer's notes: ------------------- Fixed collisions with C++ reserved words. Added portability fixes for building e2fsprogs on the HURD and AIX. Added the ext2ed program for creation of test cases. (ext2ed has many limitations and bugs which make it unsuitable for production use.) The ext2fs_read_dir_block2 and ext2fs_write_dir_block now take a new flag, EXT2_DIRBLOCK_V2_STRUCT, which will will reverse when the name_len field is byte swampped on big-endian machines, since in the V2 structure, name_len is a char field which is doesn't need to be byte swapped --- except if an old-style kernel had byte-swapped the name_len field as part of the V1 structure. E2fsprogs 1.27 (March 8, 2002) ============================== The warning messages for mke2fs now go to standard error. Fixed to make sure "make check" runs all of the test programs with LD_LIBRARY_PATH set, so that we test the libraries in the build tree. The mke2fs program checks the boot sector for the BSD disk label, and avoids erasing it if it is there. Fixed a bug in e2fsck which caused it to core dump if the journal inode was missing when it was supposed to be there. E2fsck now prints ranges in pass 5 when printing deltas for the block and inode bitmaps. Debugfs's "ls -l" command now will print out the file type information in the directory entry. Create man pages and hard links for fsck.ext3 and mkfs.ext3. If mke2fs is invoked as mkfs.ext3, create the filesystem with a journal. Debugfs can now examine the experimental directory indexing information. Fixed bug in debugfs which caused it to core dump if modify_inode is called without an open filesystem. The debugfs lsdel command now runs its output through a pager first. When installing manual pages, remove the compressed manual pages first. Synchronized with Debian's packaging information for e2fsprogs-1.26-1. Fix the 2.4 resource limitation workaround introduced in 1.26 which actually broke things on mips32, sparc32, and Alpha platforms. Updated the I18N code so that calls to setlocate(LC_CTYPE, 0) are made (which is required by the newer libintl libraries). Programmer's notes: ------------------- Fixed various gcc -Wall complaints. Fixed a few memory leaks in the e2fsck journaling code, and in the ismounted code checking for a swap device. Add new inode I/O abstraction interface which exports an inode as an I/O object. Exported ext2_file_flush as a public interface. In ext2_file_write, we now mark the buffer void * argument as a const, since ext2_file_write doesn't modify the buffer. Lots of small random portability fixes to make e2fsprogs build under AIX --- even without the 5L compatibility toolkit, and even using the uber-crippled AIX native C compiler. E2fsprogs 1.26 (February 3, 2002) ================================= Dumpe2fs will keep going now if the bad block inode can't be read. (Previously it stopped with a fatal error.) E2fsck will now give an opportunity to run the journal even if the recovery flag is not set. This is the default behaviour if e2fsck is using a backup superblock, since the needs_recovery flag will never be set in the backup superblock copies. E2fsck now automatically finds the backup superblock/group descriptors even when the primary superblock is completely destroyed for 2k and 4k filesystems. (Previously it just guessed/assumed that we were dealing with a 1k filesystem in that case, and users had to manually specify the backup superblock number.) Fixed a bug in e2fsck where it dereferences a null pointer when there is a problem opening a filesystem in preen mode. E2fsck now handles external journals correctly; previously it trashed the external journal device if the journal needed to be replayed. Work around ulimit incompatibility problem caused by recent 2.4 kernels; the unix IO module will automatically try to set any resource limits to be infinite on startup. Fixed a bug in e2fsck where it wasn't allocating a new block or inode bitmap if it wasn't present and the blocksize was 2k or 4k. (Addresses Debian Bug #116975) E2fsck will check and fix botched ext3 V1 to V2 superblock updates by clearing the new V2 fields if they do not make sense or if the ext3 superblock is version 1 superblock. E2fsck will automatically relocate the ext3 journal from a visible file (i.e., /.journal) to an hidden inode if the filesystem has been opened read/write. This allows the users to add a journal while the filesystem is mounted, but the next time the system is rebooted, the journal file will disappear. This avoids problems with backups, stupid operators with superuser bits, etc. Fix a bug in fsck where it would not support a filesystem type of "auto" if the device was specified in terms of LABEL=xxx or UUID=xxx. Fsck now supports fstab entries of the form "ext3,ext2". It will also automatically identify reiserfs filesystems. The number of processes spawned by fsck can now be limited using the FSCK_MAX_INST environment variable. Fsck now searchs the LVM proc hierarchy to find logical volumes which should be searched for UUID and label information. Work around a bug in 2.4.10+ kernels by trying to unset the filesize limit if at all possible, when opening a block device. (The filesize limit shouldn't be applied against writes to a block device.) In mke2fs and e2fsck, specifying the -c option twice will now do a read/write test on the disk. Update the man pages to encourage using the -c option, and to discouraging running badblocks separately, since users tend to forget to set the blocksize when running badblocks. Mke2fs now automatically clears the immutable attribute on a pre-existing .journal file left over from a previous failed attempt to add a journal to an alreadyy-mounted filesystem. Fixed mke2fs's exit codes to consistently indicate when the mke2fs operation failed. Mke2fs now creates the lost+found directory with permissions of 0700, so that files that had previously lived in protected directory are safe if they get dropped in lost+found and the system administrator doesn't deal with immediately. (Addresses Debian #bug 118443) Mke2fs and e2fsck (and all programs that use the ext2fs_check_if_mounted function) will now properly identify that a device is mounted, even in cases where devfs has confused things with multiple devices names with the same identity, or if a dim-witted system administrator has done something stupid like creating device file aliases in their home directory. Also now checks for swap devices by using /proc/swaps. (Addresses Debian bug #106622) Added a new option (-T) to tune2fs which sets the time a filesystem was last checked. Speed up e2image when creating sparse raw image files by optimizing away excess lseek() system calls. Fix support of large (> 2GB) files when used as a filesystem in mke2fs, tune2fs, debugfs, and findsuper. Debugfs's stat and icheck commands now properly deals with large (> 2GB) files. Debugfs's set_super_value command now prints out the list of valid superblock fields which can be set using the command. Debugfs's rm and kill_file command now updates the superblock free block and inode counts, thus keeping the filesystem consistent. Debugfs's lsdel command now takes an optional argument which allows the user to only see the most recently deleted files. A new command (undel) was added to debugfs which automates undeleting a deleted inode and linking it back to a directory. Debugfs's ls command now takes a new option, -d, which lists deleted directory entries. Debugfs's testb, freeb, setb, and find_free_block commands now take an optional count argument. Add support for a new ext2 file attribute, EXT2_NOTAIL_FL, which will be used to signal that a particular inode should not be eligible for tail-merging --- this is needed for programs like LILO. The findsuper (an unreleased, uninstalled utility program) has been improved to add extra validity checks and to add a progress meter. (It is still an unsupported program; the officially supported way to recover from a trashed partition table is to use gpart; findsuper is for wizards only. :-) Debugfs was fixed to compile with "configure --disable-swapfs". Cleaned up various manual pages. (Addresses Debian bug #119624, #120171) Added new translation file for Turkish. Programmer's notes: ------------------- Fix general gcc -Wall complaints. E2fsprogs (mostly) works with the dietlibc. The programming texinfo file has been expanded to include description of additional libext2fs functions. (Still not compeltely done, but a lot of the more important functions have been documented.) Added a umask structure to struct_ext2_filsys, which currently only modifies the behaviour of ext2fs_mkdir(), but if we add any file creation functions to libext2fs, we should also make sure they respect the umask setting. The build-rpm script was fixed to be a bit more safe. The tests' Makefile now has a way of automating test case creation for e2fsck, using "make testnew". Created a new function, ext2fs_dir_iterate2 which passes more information to the callback function (identical to the one used by ext2fs_dblist_dir_iterate). The directory iterator functions take a new flag, DIRENT_FLAG_INCLUDE_REMOVED, which will return deleted directory entries. If the directory entry is deleted, the callback function will be called with the entry paraemter set to DIRENT_DELETED_FILE. Added new functions, ext2fs_inode_alloc_stats and ext2fs_block_alloc_stats, which takes updates block/inode allocation statistics in the bitmaps, block group descriptors, and superblock when inodes or blocks are allocated or deallocated. E2fsprogs 1.25 (September 20, 2001) =================================== This is primarily a bug-fix release; no new features were added, but there are a number of embarassing bug fixes and cleanups applied. Fix a bug mke2fs which causes stack garbage to be written to disk when zapping disk sectors. (This bug could cause mke2fs to core dump for some kernels, I suspect with security enhancement patches.) Remove unneeded #include of which was breaking building e2fsprogs on the Hurd. (Addresses Debian bug #112414.) Updated tune2fs man page to reflect the fact that adding or removing a journal doesn't require running e2fsck. Remove use of AC_REQUIRE from autoconf which had been used to prevent AC_CANONICAL_HOST from being called twice; unfortunately this causes recent autoconf to bomb out since they don't allow AC_REQUIRE to be used outside of autoconf macros. Fortunately, it doesn't seem to be necessary any more, anyway. E2fsck will now not fall back to an alternate superblock if the user specifies the superblock location explicitly on the command-line. This allows easier recovery from a situation where the primary superblock and block groups are slightly corrupted, but the backup superblocks are completely broken. Fix problem which caused compile_et and mk_cmds to blow up if e2fsprogs was in a directory with a pathname that contained spaces. We are now more paranoid about checking the entry in /etc/mtab to make sure the filesystem is really mounted, since some broken distributions (read: Slackware) don't initialize /etc/mtab before checking non-root filesystems. (Slackware also doesn't check the root filesystem separately, and reboot if the root filesystem had changes applied to it, which is dangerous and broken, but there's nothing I can do about that.) Make UUID library C++ friendly by adding appropriate extern "C" declarations and using const in the function declarations. Fix up the com_err texinfo file so that it can product a valid info file (previously, it could only be used to generate paper documentation using texinfo.tex). E2fsprogs 1.24a (September 2, 2001) =================================== Fix brown-paper bug in mke2fs which caused it to segfault when printing the version string. E2fsprogs 1.24 (August 30, 2001) ================================ Revert the BLKGETSIZE64 support, since for some distributions, this ioctl has been used by an unofficial kernel patch to update the last sector on the disk, and this was causing disk corruption problems as a result. Mke2fs is now more careful about zapping swap space signatures and other filesystem/raid superblock magic values so that programs like mount who try to use hueristics to determine the filesystem type of a partition are less likely to be confused. E2fsck will no longer complain if the the mode of EXT2_RESIZE_INO is a regular file (since Andreas Dilger's on-line resizing tools will set its mode bits). Fixed some minor texinfo, man pages nits for spelling errors, texinfo warnings, etc. (Addresses Debian bug #110621.) E2fsprogs program no longer print the filesystem version number (i.e. 0.5b), since it only confuses people and doesn't serve any real purpose. E2fsck will once again compile under libc5, since it will supply its own version of strnlen if necesssary. mke2fs and tune2fs will allow the use of UUID= or LABEL= specifiers when specifying the external journal device. tune2fs will also search devices looking for the externla journal debice when removing. E2fsprogs 1.23 (August 15, 2001) ================================ Add initial support for extended attributes (EA); e2fsck will correctly handle a filesystem with EA's and check the EA blocks for corruptions. E2fsck's symlink sanity checking has been cleaned up. It now checks the i_size value of fast symlinks, and checks for immutable flags being set of symlinks, etc. E2fsck now offers to clear inodes which are special files that have a non-zero i_blocks or i_size field. (The i_size field check was in the previous version of the code, but due to a bug it didn't offer to clear the inode unless i_size and i_size_high were both non-zero.) E2image can now create "raw" image files, which only contain the filesystem metadata placed in a spare file so that e2fsck, dumpe2fs, debugfs, etc., can be run directly on the raw image file. Add support for the 64-bit block device patches. Fixed bugs in creating external journals with a 1k blocksize. Add initial support for external journals (so long as the external journal only supports a single filesystem) in e2fsck. Remove requirement for needing to run fsck on a filesystem after removing a journal (either internal or external). The man pages now document how to create and manage external journals. Speed up the check of an ext3 filesystems by avoiding a needless flush of all of the superblock and block group descriptors. Speed up creating an internal journal using tune2fs in the case where the filesystem has a lot blocks already allocated. Tune2fs has been fixed to make sure that only error messages go to stderr, and normal message go to stdout. (Addresses Debian bug #108555) Fixed a minor bug in mke2fs; if -O none is passed to mke2fs, it will now not set the sparse_super feature. (Addresses Debian bug #108165) Add support in fsck for the filesystem type "auto". Fsck -A will not try to interpret device names for filesystems which have a pass number is 0. (Addresses Debian bug #106696). Fsck prints a warning message if now valid filesystems are passed to it. (Addresses Debian Bug #107458.) E2fsck now gives an explicit warning if there filesystem still has errors at the end of the run. (Addresses Debian bug #104502) E2fsck will set the EXT2_ERROR_FS flag if the journal superblock reflects an error. E2fsck will also not run the orphan list if the filesystem contains errors, since the orphan list might be corrupted. E2fsck now prints the number of large files when given the -v option. Fixed minor memory leaks in e2fsck. Some minor man pages updates. (Addresses Debian bug #30833, #108174) E2fsprogs 1.22 (June 22, 2001) ============================== Fixed a bug in e2fsck's handling of orphan inodes which are special files (i.e., block/character device files, named FIFO's, etc.). Fixed a bug accidentally introduced in e2fsprogs 1.21 which caused tune2fs to fail at adding a journal to a mounted filesystem. Fixed a few big-endian bugs in e2fsprogs * The directory block functions were accidentally reporting some directories as corrupted when they weren't. * If e2fsprogs is compiled --disable-swapfs, the C language equivalents weren't being included for big-endian platforms. (Fixes Debian bug #101686). Fixed a Hurd compilation problem. (Addresses Debian bug #101361) Programmer's notes: ------------------- Use platform independent method of defining the BLKFLSBUF and FDFLUSH ioctls, and allow them to be defined for non-i386 platforms. The uuid.h header file is now protected against multiple inclusions. E2fsprogs is now being developed using BitKeeper. Changed the test scripts to deal with BK's stripping CR characgters from text files, and changed the top-level Makefile.in to avoid including BitKeeper files when generating the source tarball. E2fsprogs 1.21 (June 15, 2001) ============================== Added new configure flags which allow a subset e2fsprogs to be built; this is most useful for boot floopies, since the resulting shared libraries and programs are slimmed down by removing features that aren't necessary for a boot floppy. The new flags that were added are: --disable-swapfs, which removes support for byte swapping old legacy PPC/68k filesystems, --disable-debugfs, which removes support for debugfs from the libext2fs library, --disable-imager, which removes support for the e2image program, and --disable-resizer, which removes support for resize2fs. E2fsck now prints the number of mounts or days elapsed since the last check when e2fsck is forced to check an otherwise clean filesystem. Tune2fs now prints an informative message about how often a filesystem will be checked when adding a journal to the filesystem, to remind the user that he/she may want to adjust those parameters using tune2fs -c/-i. Worked around hurd brain-damage which causes e2fsck to sometimes believe a filesystem is the root filesystem based on device numbers (since Hurd doesn't have dev_t's, which is arguably a POSIX.1 violation). Fixed a bug introduced in 1.20 which caused e2fsck to abort with an erroneous error with the -F option was specified. Fixed a ext3 recovery bug in the revoke handling; synchronized with ext3 0.7a. Fixed two bugs in e2fsck's handling of dup block handling, dealing with relatively uncommon edge cases: a directory with an indirect block which is claimed by another file, and when the last inode in the filesystem has blocks claimed by another file. E2fsck now checks to see if the i_size field of a fast symlink is too big, and offers to clear the symlink if so. E2fsck now checks to see if i_size_high of special files is non-zero, and offers to clear i_size_high. Fix e2fsck's handling of incompatible journal flags so that the user has chance to abort, and then has the option to clear out the journal entirely. (Addresses Debian bug #98527.) Fixed a bug in fsck which could cause it to core dump if a mix of standard and non-standard device names are used in /etc/fstab. (Debian bug #100559) Fixed a bug in debugfs which caused read errors when copying a file to not be noticed. The debugfs set_super_value command can now modify the s_lastcheck field. Fixed a bug in lsattr and chattr which was accidentally introduced in 1.20 to support > 2GB files; both lsattr and chattr wasn't reading directories correctly because the change modified the layout of struct dirent to be incompatible with the libe2p shared library. Cleaned up the mke2fs manual page and included a discussion about why it's good to periodically check the filesystem even when journaling is enabled. Programmer's notes: ------------------- Fix general gcc -Wall complaints. The types needed by the ext2 header files are now provided by lib/ext2fs/ext2_types.h, instead of include/asm/types.h. Integers are now preferred to longs when trying to find a 32-bit type in ext2_types.h. Also, if linux/types.h has already been defined, don't try to redefine the types. Fixed make depend script so that it automatically corrects the pathname cleanups performed by make -M, so I don't have to fix them up by hand. Fixed the d_loaddump test case to be more robust, and not depend on bash'isms. Removed debugfs's dependence on pread(), which was accidentally intrudced in e2fsprogs 1.20 Fixed a performance bug in the libext2fs's icount routine; the size estimate of the icount array was incorrectly being calculated. Removed use of the badblocks compatibility functions in the e2fsprogs programs. Added paranoia code which protects against strange cases where /etc isn't on the root filesystem, or if /etc/mtab doesn't exist. The header file ext2_types.h is now installed. Autoconf is used to determine when we are on big-endian machines, instead of doing run-time tests, to save a few bytes of code. The ext2fs_mark_generic_bitmap and ext2fs_unmark_generic_bitmap functions are no longer inline functions, which saves space and doesn't really cost any real performance. The ext2fs library no longer depends on the e2p library. (What need there was of it --- namely, fsetflags, was coded in-line). Fixed the makefile so that lib/ext2fs/ext2_types.h is generated even when the user is stupid and tries compiling the package using "make install" as root. Miscellaneous code cleanups: * Added missing files from Makefile.in's SRCS file, so that their dependencies would be properly calculated. * Removed redundant code * Fixed comments in code * Removed no-longer unneeded argsused #pragma. E2fsprogs 1.20 (May 20, 2001) ============================= Add support for replaying the ext3 journal purely in user mode, including handling the orphaned inode list. Used code contributed by Andreas Dilger, with bug fixes and the orphaned inode handling done by Theodore Ts'o. The mke2fs and tune2fs programs can create or modify a filesystem to include an ext3 journal. Tune2fs also can be used to remove an ext3 journal from a filesystem. E2fsck will now check for the existence of a linked list of orphan inodes in the superblock, and clear those inodes before starting the rest of the filesystem check (but after the journal playback). E2fsck now validates the file descriptor passed to the -C option, which saves against the completion bar getting written to an unexpected location, such as the disk being checked. (Debian bug/wishlist #55220) E2fsck will now bump the filesystem revision number from zero to one if any of the compatibility bits are set. Fixed a bug where a badly corrupted directory on a big endian system could cause e2fsck to die with a bus error. The ext2fs_read_dir_block() and ext2fs_process_dir_block() functions in the ext2 library now does alignment sanity checks on the rec_len field of the directory entry before using it. The ext2 library has been enhanced to make tune2fs safe to run on mounted filesystems. (Users could usually get away with using tune2fs on mounted filesystems before, but with the advent of ext3 and journaling, it became important to make tune2fs was *really* safe for use even when the filesystem being modified is mounted.) E2label is now implemented by tune2fs using an argv[0] dispatch, so that e2label is also now safe for use on mounted filesystems. Added a new program, e2image, which creates a backup of critical ext2 filesystem data structures. The generated image file can be examined using dumpe2fs and debugfs. In the future, e2fsck will be able to use the image file to help recover very badly damaged filesystems. Fixed a number of LFS bugs in e2fsck; very, very large (> 2**42) files no longer cause e2fsck to bomb out. Also treat files > 2GB as being large file when deciding whether or not the filesystem has large files. Fixed lsattr and chattr so that they work correctly on large files. (Fixes Debian bug #72690.) Removed limitation in get_device_size() which imposed a one terrabyte filesystem limitation. (Most 2.2 kernels still have a signed int problem which cause 1 TB block device limitation. Fortunately, the kernel patches to fix this are much easier than fixing the 2TB limitation in the kernel. :-) A max_mount_count of zero is now treated as if no mount count were set. (Previously, no mount count was indicated by using -1, and a mount count of zero caused e2fsck to always be run.) Mke2fs supports two new filesystem types largefile and largefile4. Mke2fs now adds some randomness in s_max_mount_count so that multiple filesystems won't be all checked at the same time under normal operations. Fixed bug in the progress bar printing code which could cause e2fsck to core dump on an illegal filesystem. Fixed bug in fsck which could allow more than one instance of e2fsck to be printing a progress bar. (Debian bug #65267) Fsck using a UUID or a LABEL specifier will work even if devfs is compiled into the kernel and not mounted. If the pathnames in /proc/partitions are incorrect, fsck will search /dev for the correct device (using the new ext2fs_find_block_device library function). Fsck now also checks the RAID devices first so that they are properly found when they are in use. Support has also been added to support additional IDE disks and the DAC 960 device names. (Debian bug #94159) Fixed a bug in fsck which caused it not deal properly with 16 byte long filesystem labels. Fsck's -t option has been made a lot more flexible. The semantics for what happens if a comma-separated list to fsck has been regularized, and it is now possible to filter what filesystems will get checked based what is in the filesystem's fstab entry's option field. (Debian bug #89483.) The dumpe2fs program can now print out the group description information in hex, and also prints the location of the superblock and block group descriptor for those block groups that have them. Mke2fs now clears the ext2 superblock before it starts creating a filesystem, so that the superblock magic number is only written if the filesystem creation process successfully completes. The debugfs program's stat command now pretty-prints the blocks used by an inode so that it's more compact and informative. The debugfs stats command now uses the same libe2p code (which is used by dumpe2fs) to print the superblock header information. This is more complete, and it avoids a bit of code duplication. Added a new debugfs command, set_super_value (ssv) which allows the user to set arbitrary superblock fields. Debugfs was extended to support inode numbers in hex (by prefixing them with 0x), and so that modify_inode can set the inode generation number. Also, there is now a new function command called logdump which will dump an ext3 journal. Fixed a bug in debugfs so that quitting out of the pager doesn't kill debugfs. Debugfs's dump command now stops immediately upon reporting a disk read error. (Fixed a bug in ext2fs_file_read library routine which caused debugfs not to stop.) (Debian bug #79163) On systems with /proc/mounts (mainly Linux systems), /proc/mounts is checked before /etc/mtab is used, since /proc/mounts is more likely to be accurate. Added portability fixes for Solaris and Linux/ia64. Various manual pages were clarified and cleaned up. (Fixed debian bugs #63442, #67446, and #87216) Programmer's notes: ------------------- The e2fsck message printer now supports %Iu and %Ig, which will print out the inode's user and group owners, respectively. E2fsprogs now includes its own version of include/linux/ext2_fs.h, so that no longer dependent on the system having the correct version of the kernel header files. Added a new function to libext2, ext2fs_find_block_device(), which searches the system (i.e., /dev, /devfs, /devices) for a pathname to a device given its device number. Added a new function to libext2, ext2fs_sync_device, which centralizes all of the places which might try to use the BLKFLSBUF or FDFLUSH ioctls (and usually failing to define them since the system header files don't usually do this for us, and we're trying to avoid usage of kernel include files now). Added new utility programs in tests/progs: random_exercise and hold_inode. They aren't built by default; they're useful for exercising ext3 filesystem code. Added a new ext2 filesystem flag, EXT2_FLAG_SUPER_ONLY, which causes the filesystem close functions to only update the superblock, and to not touch the block group descriptors. Needed by tune2fs when modifying a mounted filesystem. Got rid of struct ext2fs_sb and replaced it with the standard struct ext2_super_block from include/linux/ext2_fs.h. Note: this may break source (but not binary) compatibility of some users of the ext2 library. Those applications should just simply do a global search and replace of struct ext2fs_sb with struct ext2_super_block, and include the new header file which defines it. The ino_t type has been renamed ext2_ino_t to protect applications that attempt to compile -D_FILE_OFFSET_BITS=64, since this inexplicably changes ino_t(!?). So we use ext2_ino_t to avoid an unexpected ABI change. The Makefiles have been reworked so that "make check" can be run from the top-level Makefile. Fix general gcc -Wall complaints and removed dead code. Remove use of NOARGS, because we assume everyone does ANSI C these days. Added build-rpm script from sct. New functions ext2fs_image_{inode,super,bitmap}_{read,write} added to support e2image. New function ext2fs_flush_icache which must be called if the application program modifies the inode table blocks without going through ext2fs_write_inode() interface. New ext2fs_check_mount_point() function, which will return the mount point of a device if mounted. The io_channel abstraction now has an optional interface, io_channel_write_range, which allows specific byte ranges to be written. The unix_io IO channel now supports write-through caching, so that journal creation is more efficient. Added x86 assembly language routines to support byte swapping, to reduce executable size. Fixed bug in the utility program subst so that it's possible to replace a substitution variable with a zero-length string. Fixed numbering e2fsck pass1 problem numbers; an extra zero had slipped into some of the problem number. E2fsprogs 1.19 (July 13, 2000) ============================== Release the resize2fs program since the timeout before it could be released under the GPL has finally expired. Add experimental support needed for the ext2 compression patches. This requires compiling e2fsprogs with the --enable-compression flag to the configure script. Added ext3 journalling support. E2fsck will run the journal (if necessary) by temporarily mounting the filesystem. /sbin/fsck.ext3 is installed as a symlink to e2fsck. Fsck has been taught about ext3, and treats it the same as ext2 in terms of the progress bar logic. Dumpe2fs will display the superblock journaling information if the filesystem has a journal. The ext2 library will now permit opening an ext3 filesystem with the recovery flag set. This is necessary for on-line dump's to work correctly, but there may be issues with this working well since ext3 is much less agressive about syncing blocks to the filesystem, since they're safe on the journal. Tune2fs and e2fsck have been changed to allow the mount_count check to be disabled by setting max_mount_count to -1. (This was already supported by the kernel.) Create a symbolic link for fsck.ext3, since the e2fsprogs utilities are used for ext3 as well. Added internationalization support for e2fsprogs; must be enabled by passing --enable-nls to configure. Always use the provided ext2fs header files to insulate ourselves from kernel version changes. Which include files are used by e2fsprogs have also been cleaned up to improve portability. Limit the number of times that e2fsck updates the progress bar so that people who are booting using a 9600 baud console don't get swampped by too many updates. Improved the loop detection algorithm in e2sck's pass #3 so that it is much, much faster for large filesystems with a large number of directories. The memory footprint for e2fsck is now slightly smaller than before. E2fsck now checks if special devices have a non-zero size, and offers to clear the size field if it finds such an inode. E2fsck now checks if special devices have the append-only flag set, and offers to clear the inode. E2fsck now properly handles some "should never fail" cases during a bitmap copy in pass5. E2fsck now properly prints control characters in filenames as ^A .. ^Z. E2fsck now calculates the correct location of the backup superblock in the case of filesystem blocksizes > 1k. Fixed a bug in e2fsck's calculation of the number of inodes_per_block which normally didn't cause problems under most filesystem parameters, but could cause a valid superblock to be rejected in extreme cases. Other checks for validating superblock values were made more stringent. Added non-destructive write testing to the badblocks program, courtesy of David Beattie. The badblocks also now has an option to input the current set of bad blocks, so that known bad blocks are skipped to speed up the badblocks test. There is also a persistent rescan feature which causes badblocks to run until it has completed some number of passes without discovering any new bad blocks. Badblocks now checks to see if the device is mounted and refuses to do the tests involving writing to the device if it is mounted. Also, badblocks now allows the number of blocks to be checked to be defaulted to the size of the partition. Fixed a bug in fsck which didn't allow non-root users to be able to check filesystems if there were any LABEL= or UUID= entries in /etc/fstab. The Hurd doesn't support the filetype filesystem feature. The mke2fs program now makes sure that for the Hurd, the filestype feature is turned off. E2fsck will check to see if the filetype feature is turned on for Hurd filesystems, and offer to turn off the feature. Mke2fs now has a safety check to make sure the number of blocks do not exceed 32 bits even on a 64 bit platform. Really fixed a bug in fsck to allow "fsck -As" to run interactive fsck's. (For those people who like to do interactive fsck's in the /etc/rc scripts!?!) Debugfs has a few new features: the rdump command, which will do a recursive dump of a directory and all of its contents, and the lcd command which does a local chdir (much like the ftp command of the same name). In addition, the debugfs program and the open_filesystem command now takes three new options: -b and -s, which allows the blocksize and superblock location to be specified, and the -c option which is used in catastrophic situations where the block group descriptors are corrupt. If the -c option is specified, debugfs will skip trying to read in the block and inode bitmaps. Debufs's lsdel command was fixed to handle bad blocks in the inode table. A Y2K bug in debugfs's "ls -l" handling was fixed by switching to use 4 digit years. General improvements in error messages - Mke2fs prints a sane error message if the partition size is zero (usually because the partition table wasn't reread by the kernel due to the partition being busy), instead of "invalid argument passed to ext2 library while initializing superblock". - Fsck now prints more self-explanatory message if an invalid UUID= or LABEL= specification is passed to it. UUID library changed to use the LGPL. Fixed a bug in the UUID library where very rapid calls to the time-based UUID generator could cause duplicate UUID's to be returned. This was not a problem for e2fsprogs, but it could be a problem for other users of the library. Make the UUID library more robust in the face of missing or an improper /dev/urandom or /dev/random files. Added some random portability fixes for Solaris. Some minor man page updates. Fixed a memory leak in the ss library. Programmer's notes: ------------------- We now try to use lseek64 and open64 from the LFS if possible. The 3rd parameter in e2p's print_flags is now a flags word, instead of a boolean option. The mark and unmark bitmap functions now return the previous state of the bit that was being changed, which is useful for some speed optimizations. The following functions have been added to enhance the badblocks list handling in libext2fs: ext2fs_write_bb_FILE, ext2fs_read_bb_FILE2, and ext2fs_badblocks_equal. The ext2 header files now have the latest journalling fields to the superblock. The ext2fs_mkdir function in libext2fs now properly backs out of error conditions robustly. Cleaned up makefiles: - to cleanly compile with the -j flag. - so distclean removes all generated files. - so in case of an error while installing header files, the make aborts. Fix test_script so that it works correctly when compiling in the source directory. The random UUID generation routine has been made slightly better in the case where /dev/random doesn't exist. (Use of randomly-based UUID is still not recommended if /dev/random doesn't exist, however; it's better to use the time/ethernet MAC address UUID in this case.) Clean up the build process so it's more friendly in case of missing directories. The ext2fs header file can now be #include'd into C++ programs. The e2p.h header file is now installed. Added workaround to a gawk 3.0.5 bug in lib/ss/mk_cmds. E2fsprogs 1.18 (November 10, 1999) ================================== Fix a core dumping bug in e2fsck if an imagic inode is present or (more rarely) if the filesystem is badly corrupted enough that e2fsck has to restart pass 1 processing. E2fsck now closes the filesystem before freeing a large number of its data structures, so in the case of future memory faults, at least the fixed filesystem will be fully written out. If a filesystem doesn't support imagic inodes, and e2fsck discovers an imagic inode, it will offer to clear the imagic flag. E2fsck will now offer to clear the immutable flag on special files (device/socket/fifos) when running it in non-preen mode. E2fsck will now set the filetype when creating /lost+found, and when connected orphaned inodes to /lost+found. Debugfs's ncheck and icheck commands now handles the case where there are bad blocks in the inode table without bombing out. The badblocks list processing code has been made more efficiently for appending a large number of (ordered) badblocks to the badblocks list. Some minor man page updates. Fsck now allows interactive e2fsck's when using fsck -As (not a common mode, but some people like to do this in boot scripts for silly reasons). Programmer's notes: ------------------- The internal e2fsck problem code for PR_2_SPLIT_DOT was fixed to meet with the problem code convention. The badblocks list regression test program has been updated to work with previously made API name changes. The ext2fs_free() command now uses the new badblocks API to avoid using the compatibility layer. Added new regression test cases; the run_e2fsck test script now supports the ability for a test case to run a prepratory command before running e2fsck. E2fsprogs 1.17 (October 26, 1999) ================================= Fixed nasty typo in fsck which caused parallelized fsck's to go into an infinite loop. Fixed a bug in fsck where it used strncmp to compare a binary UUID, thus potentially causing problems if a binary UUID contained a NULL character. E2fsck now uses stricter checks for directory entries in pass 2: zero-length filenames are not allowed; neither are 8 byte long directory entries. The debugfs "dirty" command now clears the filesystem valid bit. (Previously this just set the dirty-as-in-needs-writing-out-to-disk bit in the in-core superblock image. The new functionality is more what the user expects, and is more useful.) Added a debugging hook to test parallel fsck; if the environment variable FSCK_FORCE_ALL_PARALLEL, then filesystems on the same drive will be checked in parallel when they normally would not be. Programmer's notes: ------------------- Fixed some #ifdef's for compilation under the Hurd OS. Fixed minor W2K compatibility problems. Fixed some miscellaneous GCC warnings. E2fsprogs 1.16 (October 22, 1999) ================================= Fixed a race condition bug in fsck; when printing a progress bar, if checking multiple filesystems in parallel, it was possible for fsck to send e2fsck a SIGUSR1 signal before e2fsck had installed its signal handler, which would cause it to terminate with a signal 10. E2fsck now properly handles filesystems that have the INCOMPAT_FILETYPE feature turned on. It can be used to convert a filesystem into using or not using FILETYPE feature. E2fsck now properly handles filesystems that have the IMAGIC feature turned on (this is used on Linux AFS servers). The mke2fs program now creates filesystems that have the filetype and sparse_superblock features enabled by default, unless it is run on a pre-2.2 kernel. These features are not supported by a pre-2.2 kernel, so there is now a new flag -O which allows the user to specify with which features she would like to create the filesystem; "mke2fs -O none" will create a filesystsem compatible with 2.0 kernels. The tune2fs program now has a -O option which allows the user to set and reset "safe" filesystem features. Currently, the only ones which allows to be modified are the filetype and sparse_superblock features. Note setting or clearing either feature will require running e2fsck on the filesystem afterwards. (n.b. Clearing the sparse_superblock feature requires that there is enough free space on the filesystem for the extra superblocks which will be created by e2fsck.) Debugfs can now set and print filesystem features in the superblock using the "features" command. Dumpe2fs will print out the complete set of features when listing the superblock. Dumpe2fs has new options -f (force) and -h (header-only). Fixed a bug in e2fsck which could cause the PROGRAMMING ERROR/bonehead message to come up. This could happen when decrementing or incrementing a link count could result in an overflow. Fixed a bug in e2fsck where the block count on the lost+found directory would not be properly incremented when the directory was expanded to the point where an indirect block needed to be allocated. E2fsck now makes some additional sanity checks on the superblock to avoid crashing or giving a memory allocation error if some of the values in the superblock are unresonable (but the superblock otherwise looks valid). Fixed a bug in e2fsck where a very badly corrupted filesystem might require two passes to completely fix the filesystem. This happened if an inode claimed blocks that was part of the filesystem metadata (typically, when garbage was written into an inode table or indirect block, since this kind of filesystem corruption normally doesn't happen otherwise). On the Alpha, glibc declares st_flags although it isn't actually used; the configure script was improved to detect this case so that e2fsprogs can avoid using the non-functional stat field. The manual pages were updated to use a more consistent formatting style consistent with standard Unix man pages. Mke2fs's man page added documentation for a few previously undocumented options. Fixed minor display bugs in tune2fs and mke2fs. Programmer's notes: ------------------- Improved portability of e2fsprogs to non-Unix systems (in particular, NT). Added features to parse and print feature strings into the e2p library. (e2p_feature2string, e2p_string2feature, e2p_edit_feature). ext2fs_mkdir() and ext2fs_new_dir_block() now creates directories whose directory entries contain proper filetype information if the filesystem supports it. ext2fs_link() now uses the low 3 bits of its flags parameter to pass the directory entry filetype information. This is used to set the directory entry filetype information if the filesystem supports it. Fixed a bug in ext2fs_expand_dir() where the block count in a directory's inode would not be properly incremented when the directory was expanded to the point where an indirect block needed to be allocated. E2fsprogs 1.15 (July 18, 1999) ============================== Add configuration checks so that e2fsprogs will compile cleanly on Linux 2.3 kernels that have renamed i_version to i_generation. E2fsck now prints a progress/completion bar (and not just a simple spinner) if the -C0 option is requested or if it receives a SIGUSR1 signal. Fsck will automatically manage the (potentially muliple) e2fsck processes to print completion bars if it is given a -C option, with the right thing happening if multiple filesystems are being checked in parallel. Mke2fs now has better automatic hueristics to determine the filesystem parameters to be used for a particular filesystem. Added a new option -T which allows the user to specify how the filesystem is to be used, which helps mke2fs do a better job selecting the filesystem parameters. Mke2fs now creates revision 1 filesystems by default, and with the sparse superblock feature enabled. The sparse superblock feature is not understood by Linux 2.0 kernels, so they will only allow read-only mounts of filesystems with this sparse superblocks. Fix bug where if /dev/null couldn't be opened (should never happen), e2fsck would hang in a tight loop. Make e2fsck handle the case where /lost+found isn't a directory. E2fsck now uses mallinfo if it exists to get accurate statistics about its memory usage. Fix bug in e2fsck where it wouldn't check to see if a disconnected inode had any problems before connecting it to /lost+found. Add check to e2fsck so it makes sure that total number of inodes in the filesystem is a sane number. Fix fencepost error when clearing an the end of the block bitmap which caused the last block in the bitmap not to get cleared. Cleaned up a number of messages in e2fsck: * The message "Group's #'s copy of the group descriptor..." was fixed so that the correct number would be displayed. * Added missing space in the "disk write-protected" error messsage * Cleaned up the error message printed when a non-interactive e2fsck needs to abort a check because the filesystem appears to be mounted. Added a new command-line utility, uuidgen, which will create and print a UUID. Make debugfs's icheck command more robust by checking to make sure an inode has valid blocks before interarting over the inode's blocks. UUID generation now uses a random-based scheme whenever possible to prevent potential privacy problems. Man pages for all of the UUID functions in the lirbary were added. Fixed bug in fsck so it won't coredump if a filesystem not in /etc/fstab is given to it. Fsck now understands the UUID=xxxx and LABEL=yyyy forms in /etc/fstab that most of the other mount utilities understands. Mke2fs will make a filesystem even if it appears mounted if the force option is given. Dumpe2fs has new command-line options which allow a filesystem expert to specify the superblock and blocksize when opening a filesystem. This is mainly useful when examining the remains of a toasted filesystem. The badblocks program has been updated to display correctly on disks with large block numbers. The badblocks program no longer gives spurious errors when errors occur on non-block boundaries, which is common if the blocksize is larger than 1k. Mke2fs will sync the disk device every MKE2FS_SYNC block groups if the MKE2FS_SYNC environment variable is set. This is to work around a VM bug in the 2.0 kernel. I've heard a report that a RAID user was able to trigger it even using a 2.2 kernel, but hopefully it will not be needed for most Linux 2.2 users. Fixed miscellaneous documentation and man pages. Programmer's notes: ------------------- Cleaned up functions such as pass1_get_blocks, pass1_read_inode which in e2fsck's pass1.c really should have been static. The return value of the uuid_compare() function was changed to make it match with the convetions used by strcmp, memcmp, and Paul Leach's UUID sample document. The "make depend" process has now been made more automated; it now automatically word-wraps the dependencies, and only replaces source Makefile.in if there has been a change in the dependencies. Also, a top-level "make depend" now recurses through all the subdirectories automatically. The Makefile in .../util has been changed so that subst is built using the native C compiler during a cross-compilation, since the subst program is only used during the build process. Also add an explicit rule to build util/subst by cd'ing to the correct directory and running Makefile. The man directories are defined in terms mandir, so that the configure script can override the location of the manual pages. The config files have been updated to recognize new machine types for both the i386 and alpha families. Fsck has been modified so that it will accurately create an fsck_instance even when the noexecute flag is set. This allows for accurate debugging of the fsck pass structure. Also, when the verbose flag is given twice, fsck will print debugging information about when fsck is waiting for jobs to finish. E2fsprogs 1.14 (January 9, 1999) ================================ Fix the fstab parsing code so that it can handle blank lines and comment characters. Also, missing pass numbers need to be treated as zero. Fixed a bug in e2fsck where under some circumstances (when e2fsck needs to restart processing after fixing an egregious inconsistency) it would try to access already freed memory. E2fsck now prints non-printable characters in directory entries and pathnames using '^' and 'M-' notation. Fixed chattr so that it will ignore symbolic links when doing recursive descent traversals. For both chattr and lsattr, no longer print the version string unless the -V option is given. Allow the system administrator to directly specify the number of inodes desired in the filesystem, for some special cases where this is necessary. Fix portability problems so that e2fsprogs can be compiled under Linux 1.2 systems and Solaris systems. Update the config.guess file with a more recent version that will identify newer Linux platforms. Programmer's notes ------------------ Ext2fs_read_inode and ext2fs_write_inode will now return an error if an inode number of zero is passed to them. E2fsprogs 1.13 (December 15, 1998) ================================== Fixed a bug in debugfs where an error messages weren't getting printed when the ext2 library routines to read inodes returned errors in the stat, cmri and rm commands. Fixed a bug in mke2fs so that if a ridiculous inode ratio parameter is provided, it won't create an inode table smaller than the minimum number of inodes required for a proper ext2 filesystem. Fsck now parses the /etc/fstab file directly (instead of using getmntent()), so that it can distinguish between a missing pass number field and pass number field of zero. This caused problems for diskless workstations where all of the filesystems in /etc/fstab have an explicit pass number of zero, and fsck could not distinguish this from a /etc/fstab file with missing pass numbers. E2fsck will create a /lost+found directory if there isn't one in the filesystem, since it's safer to create the lost+found directory before it's needed. Fixed e2fsck so that it would detect bogus immutable inodes which happen to be sockets and FIFO files, and offer to clear them. If a filesystem has multiple reasons why it needs to be checked, and one of the reasons is that it is uncleanly mounted, e2fsck will print that as the reason why the filesystem is being checked. Cleaned up the output routines of mke2fs so that it doesn't overflow an 80 column display when formating really big filesystems. Added a sanity check to e2fsck to make sure that file descriptors 0, 1, 2 are open before opening the hard disk. This avoids a problem where a broken program might exec e2fsck with those file descriptors closed, which would cause disastrous results if the kernel returns a file descriptor for the block device which is also used by FILE * stdout. Fixed up the e2fsck progress reporting functions so that the values reliably reach 100% at the completion of all of the e2fsck passes. Fixed minor documentation bugs in man pages and usage messages. Programmer's notes: ------------------- Fixed a number of lint warnings in the ext2fs library and potential portability problems from other OS's header files that might define CPP macros for names like "max" and "min". ext2fs_badblocks_list_add() has been made more efficient when it needs to grow the bad blocks list. Fixed a bug in e2fsck which caused it to dereference a freed pointer just before exiting. Fixed the substition process for generating the mk_cmds and compile_et scripts so that they will work outside of the build tree. Add sanity check to e2fsck so that if an internal routine (ext2fs_get_dir_info) returns NULL, avoid dereferencing the pointer and causing a core dump. This should never happen, but... E2fsprogs 1.12 (July 9, 1998) ================================== E2fsprogs now works with glibc (at least with the version shipped wtih RedHat 5.0). The ext2fs_llseek() function should now work even with i386 ELF shared libraries and if llseek() is not present. We also explicitly do a configure test to see if (a) llseek is in libc, and (b) if llseek is declared in the system header files. (See standard complaints about libc developers don't understand the concept of compatibility with previous versions of libc.) The ext2fs library now writes out the block group number in each of the superblock copies. This makes it easier to automatically determine the starting block group of the filesystem when the block group information is trashed. Added support for the EXT2_FEATURE_INCOMPAT_FILETYPE feature, which means that e2fsprogs will ignore the high 8 bits of the directory entry's name_len field, so that it can be used for other purposes. Added support for the EXT2_FEATURE_RO_COMPAT_LARGE_FILE feature. E2fsprogs will now support filesystems with 64-bit sized files. Added support for the EXT2_FEATURE_COMPAT_DIR_PREALLOC feature. Added new program "e2label", contributed by Andries Brouwer. E2label provides an easy-to-use interface to modify the filesystem label. Fixed bug so that lsattr -v works instead of producing a core dump. Fixed a minor bug in mke2fs so that all groups with bad superblock backup blocks are printed (not just the first one). Mke2fs will check the size of the device, and if the user specifies a filesystem size larger than the apparent size of the device it will print a warning message and ask if the user wants to proceed. E2fsck has a new option -C, which sends completion information to the specified file descriptor. For the most part, this is intended for programs to use, although -C 0 will print a spinning character to the stdout device, which may be useful for users who want to see something happening while e2fsck goes about its business. Fixed a bug in e2fsck which could cause a core dump when it needs to expand the /lost+found directory, and sometimes the bitmaps haven't been merged in. Also fixed a related bug where ext2fs_write_dir_block was used to write out a non-directory block. (Which would be bad on a non-Intel platform with byte swapping going on.) Fixed bug in e2fsck where it would print a "programming error" message instead of correctly identifying where a bad block was in used when the bad block was in a non-primary superblock or block group descriptor. Also fixed a related bug when sparse superblocks are in use and there is a bad block where a superblock or block group descriptor would have been in a group that doesn't include a superblock. Fixed a bug in e2fsck (really in libext2fs's dblist function) where if the block group descriptor table is corrupt, it was possible to try to allocate a huge array, fail, and then abort e2fsck. ext2fs_get_num_dirs() now sanity checks the block group descriptor, and subsitutes reasonable values if the descriptors are obviously bogus. If e2fsck finds a device file which has the immutable flag set and the i_blocks beyond the normal device number are non-zero, e2fsck will offer to remove it, since it's probably caused by garbage in the inode table. When opening a filesystem, e2fsck specially checks for the EROFS error code, and prints a specific error message to the user which is more user friendly. If the filesystem revision is too high, change the e2fsck to print that this is either because e2fsck is out of date, or because the superblock is corrupt. E2fsck now checks for directories that have duplicate '.' and '..' entries, and fixes this corruption. E2fsck no longer forces a sync of the filesystem (with attendant sleep calls) at all times. The ext2fs_flush() function now performs a sync only if it needed to write data blocks to disk. Fixed a minor bug in e2fsck's pass1b's file cloning function, where certain errors would not be properly reported. Updated and expanded a few points in the man pages which users complained wheren't explicit enough. Added special case byte-swapping code if compiling on the PowerPC, to accomodate the strange big-endian variant of the ext2 filesystem that was previously used on the PowerPC port. Programmer's notes: ------------------- Removed C++ keywords from the ext2fs libraries so that it could be compiled with C++. E2fsck's internal organization has now been massively reorganized so that pass*.c don't have any printf statements. Instead, all problems are reported through the fix_problem() abstraction interface. E2fsck has also been revamped so that it can be called as a library from a application. Added new fileio primitives in libext2fs for reading and writing files on an unmounted ext2 filesystem. This interface is now used by debugfs. Added a new libext2fs function for mapping logical block numbers of a file to a physical block number. Added a new libext2fs function, ext2fs_alloc_block(), which allocates a block, zeros it, and updates the filesystem accounting records appropriately. Added a new libext2fs function, ext2fs_set_bitmap_padding(), which sets the padding of the bitmap to be all one's. Used by e2fsck pass 5. The libext2fs functions now use a set of memory allocation wrapper functions: ext2fs_get_mem, ext2fs_free_mem, and ext2fs_resize_mem, instead of malloc, free, and resize. This makes it easier for us to be ported to strange environments where malloc, et. al. aren't necessarily available. Change the libext2fs fucntion to return ext2-specific error codes (EXT2_DIR_EXISTS and EXT2_DB_NOT_FOUND, for example) instead of using and depending on the existence of system error codes (such as EEXIST and ENOENT). Renamed io.h to ext2_io.h to avoid collision with other OS's header files. Add protection against ext2_io.h and ext2fs.h being included multiple times. The types used for memory lengths, etc. have been made more portable. In generla, the code has been made 16-bit safe. Added Mark Habersack's contributed DOS disk i/o routines. Miscellaneous portability fixes, including not depending on char's being signed. The io_channel structure has a new element, app_data, which is initialized by the ext2fs routines to contain a copy of the filesystem handle. ext2fs_check_directory()'s callback function may now return the error EXT2_ET_CALLBACK_NOTHANDLED if it wishes ext2fs_check_directory() to really do the checking, despite the presence of the callback function. E2fsprosg 1.11 (June 17, 1997) ============================== Fixed e2fsck to detect (previously ignored) conflicts between the superblock or block group descriptors and block bitmaps, inode bitmaps, and inode tables. Fixed bug in e2fsck so that when the message printed out when a block or inode bitmap conflicts with other data, it has the correct group number. Fixed bug in e2fsck and mke2fs where the blocksize wasn't being passed to badblocks. This meant that not all of the filesystem was being tested for bad blocks! Fixed an array boundary overrun case which cropped up in ext2fs_badblocks_list_test when a user tried running "mke2fs -c -b 4096". Adjusted the number of columns printed by mke2fs when displaying the superblock backups to avoid running over 80 columns when making a really big filesystem. Fixed up the man pages for e2fsck, debugfs, badblocks, chattr, dumpe2fs, fsck, mke2fs, and tune2fs (typos and other minor grammar fixes), thanks to some suggestions from Bill Hawes (whawes@star.net). Programmer's notes: ------------------- Fixed install rule in lib/ss so that ss_err.h is actually getting installed. Fixed bug in ext2fs_copy_bitmap; the destination bitmap wasn't getting bassed back to the caller. Fixed bug in ext2fs_inode_scan_goto_blockgroup; it had not been setting the current inode number (which meant this function wasn't working at all). Fixed bug in ext2fs_resize_generic_bitmap; it had not be zeroing all blocks in the bitmap when increasing the size of the bitmap. Changed the initial number of blocks allocated by ext2fs_init_dblist() to be more realistic. Added a new function ext2fs_allocate_group_table, which sets up the group descriptor information (and allocates inode and block bitmaps, and inode tables for a particular group). The function was created by factoring out code form ext2fs_allocate_tables(). Added a new function ext2fs_move_blocks which takes a bitmap of the blocks to be moved, and moves them to another location on the boardboard. Make the unix_io channel's io_channel_flush implementation calls sync() to to flush the kernel buffers to disk. Added a new function ext2fs_dblist_count returns the number of directory blocks in dblist. E2fsprogs 1.10 (April 24, 1997) =============================== Mke2fs once again defaults to creating revision #0 filesystems, since people were complaining about breaking compatibility with 1.2 kernels. Warning messages were added to the mke2fs and tune2fs man pages that the sparse superblock option isn't supported by most kernels yet (1.2 and 2.0 both don't support parse superblocks.) Added new flag to mke2fs, -R , which allows the user to tell mke2fs about the RAID configuration of the filesystem. Currently the only supported raid option is "stride" which specifies the width of the RAID stripe. Fixed bug in e2fsck where pass1b would bomb out if there were any blocks marked bad in the inode table. Fixed rare bug in mke2fs where if the user had a very unlucky number of blocks in a filesystem (probability less than .002) the resulting filesystem would be corrupt in the last block group. Fixed bug where if e2fsck tried to allocate a block to fix a filesystem corruption problem and the filesystem had no free blocks, ext2fs_new_block() would loop forever. The configure script now checks explicitly to see if "-static" works, since that can't be assumed to be true --- RedHat doesn't install libc-static by default. Fixed bug in libext2's block iterator functions where under some cirmcustances, file with holes would cause the bcount parameter to the callback function to be incorrect. This bug didn't affect any of e2fsprogs programs, but it was discovered by Paul Mackerras, the author of the PPC boot loader. Removed use of static variables to store the inode cache in libext2fs. This caused problems if more than one filesystem was accessed via libext2fs (static variables in libraries are generally a bad idea). Again, this didn't affect e2fsprogs programs, but it was discovered by Paul Mackerras. Fixed minor bugs and version code drift to assure that e2fsprogs 1.10 will compile cleanly with 1.2.13 kernels (even with a.out shared libraries!) Programmer's notes: ------------------- Added new functions to duplicate an ext2 filesystem handle, and its associated substructure. New functions: ext2fs_dup_handle(), ext2fs_copy_dblist(), ext2fs_badblocks_copy(), ext2fs_copy_bitmap(). Other structures, such as the io_channel and the inode_cache, now have a ref count so that they only get freed when they are no longer used by any filesystem handle. (These functions were added as part of the development effort for an ext2 resizer). E2fsprogs 1.09 (April 14, 1997) =============================== Fixed bug in mke2fs (really in lib/ext2fs/initialize.c) which was accidentally introduced in the 1.08 release. The overhead calculation was accidentally removed, which caused ext2fs_initialize() to not notice when the filesystem size needed to be adjusted down because there wasn't enough space in the last block group. Fixed bug in version parsing library routine; it was always parsing the library version string, instead of using the passed-in string. Clarified chattr man page. E2fsprogs 1.08 (April 10, 1997) =============================== E2fsck 1.07 was very slow when checking very large filesystems with a lot of files that had hard links (i.e., news spools). This was fixed by seriously revamping the icount abstraction. Added a formal test suite for the icount abstraction. Debugfs now has a "-l" option to the "ls" command, which lists the inode number, permissions, owner, group, size, and name of the files in the directory. Fix a bug in e2fsck where when a directory had its blocks moved to another location during the pass 1b processing, the directory block list wasn't updated, so pass 2 wouldn't check (and correct) the correct directory block. E2fsck will now treat inodes which contain blocks which are claimed by the filesystem metadata by treating them as multiply claimed blocks. This way, the data in those blocks can be copied to a new block during the pass 1b--1d processing. E2fsck will attempt to determine the correct superblock number and display it in the diagnostic and warning messages if possible. Add support for a new (incompatible) feature, "sparse_super". This feature reduces the number of blocks which contain copies of backup superblocks and block group descriptors. (It is only an incompatible feature because of a bug in ext2_free_blocks.) mke2fs and tune2fs now support a new -s option; e2fsck will recognize filesystems built with this feature turned on. E2fsck now checks the library to make sure is the correct version, using new library functions. (This helps to diagnose incorrectly installed e2fsprogs distributions.) Dumpe2fs now prints more information; its now prints the the filesystem revision number, the filesystem sparse_super feature (if present), the block ranges for each block group, and the offset from the beginning of the block group. Mke2fs now distributes the inode and block bitmap blok so that the won't be concentrated in one or two disks in RAID/striping setups. Also, if the user chooses a 2k or 4k block group, mke2fs will try to choose the largest blocks per group that be chosen. (For 2k blocks, you can have up to 16384 blocks/group; for 4k blocks, you can have up to 32768 blocks/group.) Previously mke2fs would not allow specification of more than 8192 blocks per group, even if you were using a 2k or 4k block group. Programmer's notes: ------------------- Added a new function ext2fs_create_icount2() which takes a "hint" argument. This hint argument presets the icount array with the list of inodes which actually need to be in the icount array. This really helps to speed up e2fsck. Added a new function ext2fs_icount_validate() which checks the rep invariant for the icount structure. This is used mostly for testing. The error mesasage given when a bad inode number is passed to test_generic_bitmap to reflect EXT2FS_TEST_ERROR (instead of EXT2FS_UNMARK_ERROR). Added a new function ext2fs_set_dir_block which sets the block of a dblist entry, given the directory inode and blockcnt. Added a new function ext2fs_get_library_version() which returns the current library version, and ext2fs_parse_version_string() which returns a version number based on a e2fsprogs version string. The icount functions will return EINVAL if the passed in inode number is out of bounds. E2fsprogs 1.07 (March 9, 1997) ============================== E2fsck is now uses much less memory when checking really large filesystems (or rather, filesystems with a large number of inodes). Previously a filesystem with 1 million inodes required 4 megabytes of memory to store inode count statistics; that storage requirement has now been reduced to roughly half a megabyte. E2fsck can now properly deal with bad blocks appearing inside the inode table. Instead of trying to relocate the inode table (which often failed because there wasn't enough space), the inodes in the bad block are marked as in use. E2fsck will automatically try to use the backup superblocks if the primary superblocks have a bad magic number or have missing meta-data blocks (or meta-data blocks which are out of range). E2fsck's pass 3 has been made more efficient; most noticeable on filesystems with a very large number of directories. Completely revamped e2fsck's system of printing problem reports. It is now table driven, to make them more easily customizeable and extendable. Error messages which can be printed out during preen mode are now one line long. Fixed e2fsck's filesystem swapping code so that it won't try to swap fast symbolic links or deleted files. Fixed e2fsck core dumping when fixing a filesystem which has no directories (not even a root directory). Added a check to e2fsck to make sure that the length of every directory entry is a multiple of 4 (since the kernel complains if it isn't). Added a check to e2fsck to make sure that a directory entry isn't a link to the root directory, since that isn't allowed. Added a check to e2fsk to now make sure the '.' and '..' directory entries are null terminated, since the 2.0 kernel requires it. Added check to write_bitmaps() to make sure the superblock doesn't get trashed if the inode or block bitmap is marked as being block zero. Added checking of the new feature set fields in the superblock, to avoid dealing with new filesystem features that this package wasn't set up to handle. Fixed a fencepost error in ext2fs_new_block() which would occasionally try to allocate a block beyond the end of a filesystem. When the UUID library picks a random IEEE 802 address (because it can't find one from a network card), it sets the multicast bit, to avoid conflicting with a legitimate IEEE 802 address. Mke2fs now sets the root directory's owner to be the real uid of the user running mke2fs. If the real uid is non-zero, it also sets the group ownership of the root directory to be the real group-id of the user running mke2fs. Mke2fs now has more intelligent error checking when it is given a non-existent device. When badblocks is given the -vv option, it now updates the block that it is currently testing on every block. Fixed a bug in fsck where it wouldn't modify the PATH envirnoment currently correctly if PATH wasn't already set. Shared libraries now built with dependencies. This allows the shared library files to be used with dlopen(); it also makes the transition to libc 6 easier, since ld.so can tell which libc a particular shared library expects to use. Programmer's notes: ------------------- Added new abstraction (defined in dblist.c) for maintaining a list of blocks which belongs to directories. This is used in e2fsck and other programs which need to iterate over all directories. Added new functions which test to see if a contiguous range of blocks (or inodes) are available. (ext2fs_*_bitmap_range). Added new function (ext2_inode_has_valid_blocks) which returns true if an inode has valid blocks. (moved from e2fsck code). Added new function (ext2fs_allocate_tables) which allocates the meta-data blocks as part of initializing a filesystem. (moved from mke2fs code). Added a new I/O manager for testing purposes. It will either allow a program to intercept I/O requests, or print debugging messages to trace the activity of a program using the I/O manager. The badblocks_list functions now store the bad blocks in a sorted order, and use a binary search to speed up badblocks_list_test. The inode scan function ext2fs_get_next_inode() may now return a soft error returns: MISSING_INODE_TABLE and BAD_BLOCK_IN_INODE_TABLE in those cases where part of an inode table is missing or there is a bad block in the inode table. Added a new function (ext2fs_block_iterate2) which adds new arguments to the callback function to return a pointer (block and offset) to the reference of the block. Added new function (ext2fs_inode_scan_goto_blockgroup) which allows an application to jump to a particular block group while doing an inode scan. The badblocks list functions were renamed from badblocks_* to ext2fs_badblocks_*. Backwards compatibility functions are available for now, but programs should be modified to use the new interface. Some of the library functions were reorganized into separate files to reduce the size of some programs which statically link against the ext2 library. Put in some miscellaneous fixes for the Alpha platform. E2fsprogs 1.06 (October 7, 1996) ================================ Fixed serious bug in e2fsck: if the block descriptors are bad, don't smash the backup copies in ext2fs_close(). (The problem was that when e2fsck -p discovered the problem, while it was closing the filesystem and exiting, it was also blowing away the backup superblocks on the disk, which was less than friendly.) We now make it the case that we only write out the backup superblock and the back block descriptors if the filesystem is completely free from problems. Fixed a bug in block_interate in the lib/ext2fs library which caused e2fsck to fail on GNU Hurd-created filesystems. Add support for Linux/FT's bootloader, which actually uses EXT2_BOOT_LOADER, and sets its mode bits which caused e2fsck to want to clear the inode. Add support for the "A" (no atime update) attribute. (Note: this attribute is not yet in production kernels.) The test suite is not automatically run when doing a "make all" from the top level directory. Users should manually run "make check" if they wish to run the test suite. Upon a preenhalt(), make the printed message more explicit that running e2fsck "MANAULLY" means without the -p or -a options. In e2fsck, if a disconnected inode is zero-length, offer to clear it instead of offering to connect it to lost+found. In e2fsck, if a filesystem was just unmounted uncleanly, and needs e2fsck to be run over it, change e2fsck to explicitly display this fact. For dumpe2fs and e2fsck, cause the -V option to print out which version of the ext2fs library is actually getting used. (This will help detect mismatches of using a 1.06 utility with a 1.05 library, etc.) Programmers' notes: ------------------- EXT2_SWAP_BYTES was changed to EXT2_FLAG_SWAP_BYTES, which better fits the naming convention. In ext2fs_initialize(), make sure the description for the inode bitmap is correctly initialize. Fixed minor type typo in ext2fs_allocate_generic_bitmap(); E2fsprogs 1.05 (September 7, 1996) ================================== Add support for new fields in the ext2 superblock --- volume name, volume UUID, and last mounted field. Dumpe2fs displays these fields, tune2fs and mke2fs allows you to set them. E2fsck will automatically generate a UUID for those volumes that don't have them. Put in support for e2fsck to recognize HURD specific ext2 features --- most notably, the translator block. The e2fsprogs tools will now use the creator_os field in the superblock to correctly handle different OS-specific variants of the ext2 filesystem. E2fsck now fixes inodes which have a the deletion time set, but which have a non-zero i_link_count field by offering to clear the deletion time. Previously e2fsck assumed that the inode was deleted (per 0.3c ext2 kernel behavior) and offered to unlink the file. If e2fsck sets the clean bit, but nothing else, set the exit code FSCK_NONDESTRUCT. After all, e2fsck did fix a filesystem error --- it set the filesystem valid bit when it was previously cleared. :-) This was needed to make the HURD fsck driver happy. If the user refuses to attach an unattached inode, e2fsck will no longer set the inode's link count. Otherwise, the inode would end up getting marked as unused, which might cause loss of data later. Make the message issued by e2fsck when the superblock is corrupt less confusing for users. It now mentions that another reason for the "corrupt superblock" message might be that the partition might not be an ext2 filesystem at all (it might swap, msdos filesystem, ufs, etc.) Make the libext2 library more robuest so that e2fsck won't coredump on an illegal superblock where the blocksize is zero. (f_crashdisk is the test case). By default, create filesystems where the default checkinterval is 6 months (180 days). Linux servers can be robust enough that 20 reboots can be a long, long time. Added configure flag --enable-old-bitops, which forces the bitops to use the old (native) bitmask operations. By default on the sparc platform, the standard ext2 bit ordering is now used. Added a new feature to e2fsck to byte-swap filesystems; this can be used to convert old m68k filesystems to use the standard byte-order storage for the superblock, inodes, and directory blocks. This function is invoked by using the '-s' option to e2fsck. Debugfs's "dump" command has been enhanced so that it writes out the exact size of the file so that the nulls at the end of the file are eliminated. The command also accept a new "-p" option which will attempt preserve to preserve the ownernship, permissions, and file modification/access times. Debugfs has two new options, -f and -R. The -R option allows the user to execute a single debugfs command from the command line. The -f option allows the user to specify a "command file" containing debugfs commands which will get executed. Dumpe2fs now pretty prints the check interval, instead of just printing the check interval as a number of seconds. Fix bugs in debugfs: the params command when no filesystem is opened no longer causes a core dump. It is now possible to unlink a file when a pathame containing a '/' is specified. Tune2fs has a new -C option which sets the number of times the filesystem has been mounted. Fix the chattr '-v' option so that it actually works. Chattr was being buggy about the -v option parsing. Programmers' notes: ------------------- The directory lib/uuid contains a set of library routines to generate DCE compatible UUIDs. Extended ext2fs_namei() to handle symbolic links. Added new function ext2fs_nami_follow() which will follow last symbolic link in the case where the pathname points to a sym link. The ext2fs_block_iterate function will now return the HURD translator block, if present. The new flag BLOCK_FLAG_DATA_ONLY will cause the iterator to return data blocks only. The ext2fs.h file now defines constants BLOCK_COUNT_IND, BLOCK_COUNT_DIND, BLOCK_COUNT_TIND, and BLOCK_COUNT_TRANSLATOR, which are the magic values passed in the block count field of the iterator callback function. The test script driver now takes an optional second argument, which is the test case to be run. This allows you to run a test case without needing to run the entire test suite. On Linux ELF systems, install the .so files in the correct places (/usr/lib). The .so files must be stored in the same directory as the .a files. Fixed miscellaneous HURD compilation issues with header file being included in the right order. Fixed debugfs so that it resets optind to zero, not one, since setting optind to zero is more correct. E2fsprogs 1.04 (May 16, 1996) ============================= First "official" (1.03 was a limited release only) to support building e2fsprogs under Linux 2.0 kernels (as well as late model 1.3 and 1.99 kernels). This package includes a RPM specs file, that it can be built using the RedHat Package Manager. E2fsck now prints a hint that if there are lots of incorrectly located inode bitmaps, block bitmaps, and inode table blocks, the user might want to try using e2fsck -b 8193 first, to see if that fares any better. For ext2 filesystem written with the hurd, debugfs will now print out the translator field when printing an inode structure. Lots of miscellaneous linking/installation cleanups: Libraries are now linked using a relative pathname, instead of relying on -L working correct. It doesn't, in many cases, including current versions of GNU ld. This guarantees that the build tree is linking with the right libraries, instead of the ones installed in /usr/lib. Header files, man pages, and the et/ss shell scripts are now generated using a custom substitution script, instead of relying on the configure script. This prevents needless recompilation of files; in addition, the custom substitution script is much faster. e2fsck may now be linked dynamically, by using the --enable-dynamic-e2fsck flag to configure. This is not recommended, since it increases e2fsck's dependence on other files, but some people need to save disk space, and other critical programs on their systems were being linked dynamically anyway. Programs such as fsck which didn't need to be linked against libext2fs (or mke2fs which didn't need to be linked against libe2p) only link against libraries they actually need. Otherwise, those programs would require the presense of libraries that otherwise could be removed from a rescuse diskette. The ss include files are now installed correctly so they can actually be used by another package. If the profiling libraries are built, they are now installed on a "make install-libs". E2fsprogs 1.03 (March 27, 1996) =============================== Change the m68k bit numbering for bitmasks to match the bit numbering used by all other ext2 implementations. (This change was requested by the m68k kernel development team.) Support (in-development) filesystem format revision which supports (among other things) dynamically sized inodes. Fixed a bug in the ext2 library so that an intelligent error is returned if mke2fs is run with a ridiculously small number of blocks for a partition. Fixed a bug in the ext2 library which required that the device be openable in read/write mode in order to determine its size. This caused e2fsck -n to require read/write access when it was not previously necessary. Fixed a bug in e2fsck which casued it to occasionally fail the test suite depending on which version of the floating point library it was using. Fixed a bug in e2fsck so that it now halts with a fatal error when certain superblock consistency checks fail. Previously it continued running e2fsck, with some potential confusing/damaging consequences. Added new flag to fsck which allows the root to be checked in parallel with other filesytems. This is not the safest thing in the world to do, but some system administrators really wanted it. Fixed -Wall flames in lib/ss. E2fsprogs 1.02 (January 16, 1996) ================================= Fix to allow e2fsprogs to be compiled on recent 1.3 (pl45+) kernels. Change e2fsck to print statistics of how many non-contiguous files are on the system. Note that a file which is larger than 8k blocks, it is guaranteed to be non-contiguous. In mke2fs, print a warning message if a user tries to format a whole disk (/dev/hda versus /dev/hda1). If a user really wants to format a whole disk, the -F (force) option forces mke2fs to format a whole disk as a filesytem. Fix a bug in fsck where in some cases it might start checking partitions in the next pass before it finishes checking partitions in the current pass. This still won't cause two partitions on the same disk will be checked, so it's rarely a problem in real life. Patch lsattr so that it won't hang when checking a named pipe. Minor compilation fixes: * Fix the order of libraries that were linked in debugfs. * Allow the sources to be compiled with -ansi turned on.