The e2fsprogs and kernel implementation of directory hash tree has a
bug which causes the implementation to be dependent on whether
characters are signed or unsigned. Platforms such as the PowerPC,
Arm, and S/390 have signed characters by default, which means that
hash directories on those systems are incompatible with hash
directories on other systems, such as the x86.
To fix this we add a new flags field to the superblock, and define two
new bits in that field to indicate whether or not the directory should
be signed or unsigned. If the bits are not set, e2fsck and fixed
kernels will set them to the signed/unsigned value of the currently
running platform, and then respect those bits when calculating the
directory hash. This allows compatibility with current filesystems,
as well as allowing cross-architectural compatibility.
Addresses Debian Bug: #389772
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
- EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE (0x0040?) - add s_min_extra_isize and
s_want_extra_isize fields to superblock, which allow specifying
the minimum and desired i_extra_isize fields in large inodes
(for nsec+epoch timestamps, potential other uses). Needs RO_COMPAT
flag handling, needs e2fsck support, patch complete, little testing.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
- EXT4_FEATURE_INCOMPAT_64BIT (0x0080) - support for 64-bit block count
fields in the superblock (s_blocks_count_hi, s_free_blocks_count_hi),
large group descriptors (s_desc_size), extents with high 16 bits
(ee_start_hi, ei_leaf_hi), inode ACL (i_file_acl_hi). May also grow
to encompass the previously proposed BIG_BG.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
- EXT4_FEATURE_RO_COMPAT_DIR_NLINK (0x0020?) - allow directories to have
> 65000 subdirectories (i_nlinks) by setting i_nlinks = 1 for such
directories. RO_COMPAT protects old filesystems from unlinking such
directories incorrectly and losing all files therein.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
- EXT4_FEATURE_RO_COMPAT_GDT_CSUM (0x0010?) - store a crc16 checksum in
the group descriptor (s_uuid[16] | __u32 group | ext3_group_desc
(excluding gd_checksum itself)). This allows the kernel to more safely
manage UNINIT groups.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
- EXT4_FEATURE_RO_COMPAT_HUGE_FILE (0x0008) - change i_blocks to be
in units of s_blocksize units instead of 512-byte sectors, use
l_i_frag and l_i_fsize as i_blocks_hi (could also be part of 64BIT).
E2fsck and debugfs changed to support i_blocks_hi instead of l_i_frag and
l_i_fsize.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Add in randomness based on Linux's thread id (gettid) to avoid race
conditions when two threads try to generate uuid's at the same time.
This shouldn't be an issue if /dev/urandom has proper locking and is
present, so this is just a failsafe.
Addresses SourceForge Bug: #1529672
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Check for potential overflow for filesystems contained in regular files
where the filesystem image size is returned by stat64().
Signed-off-by: Eric Sandeen <esandeen@sandeen.net>
Create new ext2fs library inline functions in order to calculate
the starting and ending blocks in a block group.
Signed-off-by: Eric Sandeen <esandeen@redhat.com>
There were still some %d's lurking when we print blocks & inodes; also
many of the counters in the e2fsck_struct were signed, and probably
need to be unsigned to avoid overflows.
Signed-off-by: Eric Sandeen <esandeen@redhat.com>
For loops iterating over all group descriptors, consistently define
first_block and last_block in a way that they are inclusive of the
range, and do not overflow.
Previously on the last block group we did a test of <= first +
dec_blocks; this would actually wrap back to 0 for a total block count
of 2^32-1
Also add handling of last block group which may be smaller.
Signed-off-by: Eric Sandeen <esandeen@redhat.com>
Add a new functiom, e2p_percent(), which correct calculates the percentage
of a number based on a given percentage, without worrying about overflow
issues. This is used where we calculate the number of reserved blocks using
a percentage of the total number of blocks in a filesystem.
Based on patches from Eric Sandeen, but generalized to use this new function.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Eric Sandeen <esandeen@redhat.com>
For loops such as:
for (i=1; i <= fs->super->s_blocks_count; i++) {
<do_stuff>
}
if i is an int and s_blocks_count is (2^32-1), the condition is never false.
Change these loops to:
for (i=1; i <= fs->super->s_blocks_count && i > 0; i++) {
<do_stuff>
}
to stop the loop when we overflow i
Signed-off-by: Eric Sandeen <esandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Add a new function, ext2fs_div_ceil(), which correctly calculates a division
of two unsigned integer where the result is always rounded up the next
largest integer. This is used everywhere where we might have
previously caused an overflow when the number of blocks
or inodes is too close to 2**32-1.
Based on patches from Eric Sandeen, but generalized to use this new function
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Eric Sandeen <esandeen@redhat.com>
This patch fixes the blkid.8.in description of the "-l" option. The man
page gives the impression that the first match is the one that is returned.
However, the blkid_find_dev_with_tag() function returns the device with
the highest priority (which is good, because that is what people really want).
Signed-off-by: Andreas Dilger <adilger@clusterfs.com>
all lib/blkid/tst* files to be removed with "make clean", in particular
tst_types.c. That was causing a failure of "make check" in an RPM source
tree. Fix is to explicitly list the test binaries, as lib/ext2fs/Makefile.in
does.
As "make check" was only calling test_probe and tst_types (and none
of the other tst_* tests) it was not clear what was going on, and an
"hg update" would always return the old tst_types.c file back so the
problem was only being seen intermittently... It isn't clear whether
you want the other tst_* programs to be run as part of "make check".
Signed-off-by: Andreas Dilger <adilger@clusterfs.com>
This patch allows "inode_size" to be specified in the mke2fs.conf file,
and always compiles in the "-I" option. In addition, it disallows
specifying the inode size on rev 0 filesystems, though I don't think
this was much of a danger anyways.
Clean up dead lines in ext2fs.h.
Signed-off-by: Andreas Dilger <adilger@clusterfs.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
SPARCs do not like unaligned halfword access and throw SIGBUS.
Read data "manually" instead.
Tested on Solaris 8/SPARC with gcc 2.95.3.
Signed-off-by: Matthias Andree <matthias.andree@gmx.de>
Try DIOCGMEDIASIZE ioctl() if defined, to obtain
the media size on FreeBSD 5.0 and newer.
The binary search fallback doesn't work, as FreeBSD
block devices are unbuffered and refuse reads below
the block size.
Signed-off-by: Matthias Andree <matthias.andree@gmx.de>
When allocating space for the RAID filesystems with the stride parameter,
place each portion of the group's inode table right up after the superblock
(if present) in order to minimize fragmentation of the freespace.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This feature is initially intended for testing purposes; it allows an
ext2/ext3 developer to create very large filesystems using sparse files
where most of the block groups are not initialized and so do not require
much disk space. Eventually it could be used as a way of speeding up
mke2fs and e2fsck for large filesystem, but that would be best done by
adding an RO_COMPAT extension to the filesystem to allow the inode table
to be lazily initialized on a per-block basis, instead of being entirely initialized
or entirely unused on a per-blockgroup basis.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Theodore Ts'o