doc/RelNotes/v1.43.0.txt

E2fsprogs 1.43 (May 17, 2016)
=============================

Add support for the ext4 metadata checksum, checksum seed, inline
data, encryption, project quota, and read-only features.

Support for the very old, experimental, and never-added-to-mainline
compression feature has been removed.

Mke2fs will now create file systems with the metadata_csum and 64bit
features enabled by default.

The mke2fs program will now support multiple -O options (instead of
just ignoring all but the last -O option).

Mke2fs will now correctly determine the size of a file system and
handle the discard operation correctly when the "-E offset=NNN" is
used.  (Addresses Debian Bug: #803629)

The tune2fs program will ask the user for confirmation before starting
dangerous operations if the terminal is available, and it will replay
the journal if necessary.

Add an ext2/3/4 FUSE server

Debugfs can now modify extended attributes and journal transactions.

Debugfs now can properly display and set extended timestamps.

E2fsck will offer to fix extended timestamps that were incorrectly
encoded by older kernels.

Fix miscellaneous MMP bugs in debugfs.

Fix sparse_super2 bugs in mke2fs and resize2fs.

Fix or improve offset support in mke2fs, e2undo, and libext2fs.

The resize2fs command can now convert file systems between 64-bit and
32-bit mode.

Fix potential out-of-bounds memory access in resize2fs when doing a
large off-line file system grow operation.

We now use a new e2undo file format which is much more efficient and
faster than the old tdb-based scheme.  Since it so much faster,
e2fsck, tune2fs, debugfs, and resize2fs now also can support using
creating an undo file.

The mke2fs command can now set the error behavior when initializing
the file system (so the administrator doesn't have to issue a separate
tune2fs -e command).

Teach mke2fs to parse a human-readable argument for -i option.

Teach mke2fs to automatically handle creating file systems > 500T by
automatically enable the meta_bg feature.

Mke2fs will now prompt for user verification before rewriting a file
system's superblock using the -S option.

Mke2fs no longer complains if the user tries to create a file system
using the entire block device (e.g., without using a partition).  The
minor number convention is not used by all block devices, and it's
quite normal in some circumstances to create a file system on /dev/sdc
instead of /dev/sdc1.

Fix an "mke2fs -d" bug which could create inodes with invalid
extended attribute blocks

E2fsck is now much more paranoid about not freeing or corrupting
critical metadata blocks, such as inode table blocks, even if
corrupted indirect blocks or extent trees point at these blocks.

E2fsck now prints block ranges in pass1b instead of listing all of the
blocks exhaustively.

E2fsck will try to expand the root directory if the lost+found can't
be linked to the root directory.  Also, offer to use the root
directory if lost+found can't be created.

E2fsck is now more paranoid handling corrupted extent trees as well as
corrupted journals.

E2fsck can now rebuild extent trees, either (a) to optimize them, (b)
to recover from a corrupted extent tree, or (c) to convert
block-mapped inodes to use extents.

E2fsck now has a readahead mechanism which can significantly speed its
performance, especially on RAID arrays.

E2fsck now has a "yes to all" option which the user can give if she is
tired of answering 'y' to a very large number of questions.

E2fsck will now ignore the badblocks inode if the contents of the
badblocks inode indicate that the portion inode table containing the
badblocks inode is bad.  (We must go deeper...)

E2fsck can now correctly fix directory with holes on bigalloc file
systems.

E2fsck will now check for extent mapped inodes with no extent header
and clear them if found.

E2fsck now checks to make sure the extended attribute header doesn't
result in the xattr extending beyond the bounds of the inode

Fixed a bug in e2fsck to avoid overrunning a buffer containing jbd2
revoke records if the journal is corrupted.

Fixed a bug in e2fsck which could cause it loop forever if a special
inode has too many invalid block mappings.

Fixed a bug in e2fsck which could cause pass1b/c/d processing to get
confused if an attempt to allocate a block can't find any free space
in the file system.

E2fsck will no longer try to force rewrite blocks located beyond the
file system.

The e2fsck program will no longer update the quota inodes when it is
interrupted.

Fix a bug which in rare cases would cause e2fsck -fD to corrupt an
extent-mapped directory.

Fixed a bug in resize2fs which could lead to resize2fs crashing or a
corrupted file system if the file system is almost completely full
when trying grow a file system and we need to allocate blocks to grow
the block group descriptors.

Fixed a bug in resize2fs which could cause it to get fooled trying to
determine the RAID array's stride when flex_bg is enabled.

Fix resize2fs bug that could leave block allocation bitmaps
uninitialized.

The dumpe2fs output has been improved so it is cleaner and always fits
within 80 columns.  Also added a more easily machine-parsable output
of dumpe2fs.

The mke2fs program can now pre-populate a file system from a directory
hierarchy using the -d option.

The mke2fs program now skips zeroing inode table blocks if they were
already zeroed using the discard feature.

Check to make sure file system features which can not be supported by
HURD are not enabled if the file system is created to be
HURD-compatible.

Added a new e2fuzz command that will fuzz an ext4 image for testing
purposes.

The debugfs logdump command can now deal with 64-bit revoke tables
correctly.  Also, "logdump -O" will print the old log contents (before
the journal was replayed).

The debugfs bmap command can now be used to set or allocate a physical
block.

Fixed a bug so "filefrag -B -e -v" does not return a separate entry
for each block.

The filefrag program now accounts for holes in sparse files created by
the file punch operation as an expected/optimal mapping.

The file I/O functions now correctly handle inodes containing
uninitialized blocks.

Fix a bug in tune2fs so that removing uninit_bg feature on a bigalloc
file system won't result in corrupted block bitmaps.

Updated/fixed various man pages.  (Addresses Debian Bug: #798425)


Programmer's Notes
------------------

Fixed coverity, sparse, gcc -Wall, and clang warnings/nits.

Added Android build files so that e2fsprogs can be built in the
Android source tree.

Reduce the use of libc functions in libext2fs that may not be present
in the boot loader environment, at least for those functions that are
needed by boot loaders such as yaboot.

Developers can now override the debugging and optimization flags by
redefining the CFLAGS makefile macro.

The libext2fs library now has support for fallocate.

The mke2fs command will now ask the user for confirmation if block
device or image file contains an existing file system image, and
stdout and stdin are connected to a tty.

The libext2fs library now picks a more intelligent goal block when
doing block allocations.

The libext2fs library will now automatically set the BLOCK_UNINT flag
if all of the blocks in a block group are free, to speed up future
e2fsck and dumpe2fs operations on the file system.

Add two new functions ext2fs_new_range() and ext2fs_alloc_range() to
libext2fs.

The ext2fs_zero_blocks() command will use FALLOC_FL_ZERO_RANGE for
file-based images.

The ext2fs_bmap() function supports new flags BMAP_UNINIT and
BMAP_ZERO.

The ext2fs_new_block2() function will now call the alloc_block hook
before checking fs->block_map.

Support for the MMP feature can now be disabled at compile time.

Added support to manipulate extended attributes to libext2fs.

Added a lot of new regression tests.

Added endian annotations so it's possible to scan e2fsprogs for endian
problems using a static code analyzer.

Fixed memory leaks in libext2fs.

The e2fsck jbd2 handling code has been resynced with the 3.16 kernel.
There is now a script in the contrib directory which automates most of
the resync process.

The build system will now run cppcheck (a static code analysis tool)
via "make C=1"