The locally defined versions of both sync_file_range and fallocate are broken
on 32bit systems. On these systems two 32bit registers are needed for each
64bit parameter. Also, sync_file_range on MIPS32 needs a dummy parameters
after the fd parameter. Just leave all these subtleties to the C library.
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Declare struct_io_manager at the end of unix_io.c, undo_io.c, and
test_io.c files so that there isn't a need to forward declare every
member of this structure. That avoids a lot of redundant code
at the start of every one of these files.
Move the test_flush() function above test_abort() to avoid the need
for a forward declaration.
Fix a few instances of space before tab in these files.
Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The "mkswap" program is not available on MacOS, so just use the
existing swap0.img.bz2 and swap1.img.bz2 files directly.
Because MacOS HFS+ doesn't support sparse files (welcome to the 80's)
the m_bigjournal test takes forever to zero out the whole 42GB test
filesystem. Skip this test for Darwin kernels for now.
Unfortunately, neither "df -T" nor "stat -f -c %T" is available on
MacOS to directly determine the filesystem type, and I'm too lazy
to parse the output of "mount" and match it to the path of the test
directory in shell, so it just checks the kernel type and assumes
the filesystem type is HFS and skips the test.
Since this test runs on Linux the majority of the time, the loss of
test coverage is minimal. If MacOS should ever get a real filesystem,
this can be revisited.
Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Fix a number of non-literal string format warnings from LLVM due
to the use of _() that were not fixed in commit 45ff69ffeb.
Fix mismatched int vs. __u64 format warnings in blkmap64_rb.c.
There were also some comparisons of __u64 start or count <= 0.
Change them to be comparisons == 0, or start + count overflow.
Fix operator precedence warning for (value & (value - 1) != 0)
introduced in 11d1116a7c. It seems "&" is lower precedence
than "!=", so the above didn't fail for power-of-two values,
but only odd values. Fortunately, either s_desc_size nor
s_inode_size is valid if odd.
Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Ext4 file system also supports to set/clear 'j' attribute, but it just
say that this option is only useful for ext3 in manpage. This commit
fixes it.
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com>
In C++, "private" is a reserved keyword, so don't use it in the header
file as a function parameter name.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If a client asks us to remap a block in the middle of an extent, we
potentially have to allocate a fair number of blocks to handle extent
tree splits. A failure in either of the ext2fs_extent_insert calls
leaves us with an extent tree that no longer maps the logical block in
question and everything that came after it! Therefore, try to roll
back the extent tree changes before returning an error code.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
If we're doing a BMAP_ALLOC allocation and the extent tree update
fails, there's no point in hanging on to the newly allocated block.
So, free it to make fsck happy.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When modifying/removing an extent during punch, don't forget to update
the extent's parents.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When we're iterating extents during a punch operation, the loop exits
if the punch region is entirely to the right of the extent we're
looking at. This can happen if the punch region starts in the middle
of a hole and covers mapped extents. When this happens, we want to
skip to the next extent, because it might be punchable.
Also, if we've totally passed the punch range, stop.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Refactor the running kernel version checks to hide the details of
version code checking, etc.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When meta_bg feature is enabled, group descriptor block is allocated
every 128 block group (or every 64 block group if 64bit feature is
enabled).
In such situation, files in block group more than #128 will be removed
if sparse_super feature is enabled with tune2fs and afterwards
necessary e2fsck running.
Because tune2fs does not reallocate group descriptor blocks but just
set sparse_super feature. If ext4 has sparse_super,
ext2fs_descriptor_block_loc2() called by e2fsck thinks the block group
(e.g. #128) that it has group descriptor block at the head offset. But
that offset is used as backup super block before. So e2fsck fixes
ext4 based on invalid group descriptor blocks and this cause data
lost.
The patch avoids this problem simply by disallow tune2fs enabling
sparse_super if meta_bg is enabled.
Steps to reproduce:
1. Create ext4 which has meta_bg, ^sparse_super and 129+ block groups.
# mke2fs -t ext4 -O meta_bg,^resize_inode,^sparse_super DEV 17G
# mount DEV /MP
2. Create direcotry and files which use block group #128's metadata.
# echo $((8192*128+1)) > /sys/fs/ext4/DEV/inode_goal
# mkdir /MP/DIR
# for i in $(seq 1 100); do dd if=/dev/urandom of=/MP/DIR/file$i bs=1024 count=10; done
3. Enable sparse_super with tune2fs then execute e2fsck.
Data in block group #128 will be lost!!
# umount DEV
# tune2fs -O sparse_super DEV
# e2fsck/e2fsck -yf DEV
Signed-off-by: Akira Fujita <a-fujita@rs.jp.ne.cocm>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Setting SKIP_UDEBS=yes in rules.custom will prevent the debian/rules
makefile from building the udeb files for the debian installer.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Interpret "zero_hugefiles" relation in mke2fs.conf as a boolean value,
as documented in the man page.
If the hugefile is larger than 2GB, set the large_file file system
feature so e2fsck doesn't complain.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Commit becb01ce84 breaks building e2fsprogs with dpkg 1.15.8 which is
used in Debian 6.0 (Squeeze), since it doesn't support package
specifications qualified with an architecture (i.e., "dpkg-query -W
libblkid1:amd64").
Debian only needs to use its own version of libblkid and libuuid for
versions of Debian 5.0 (Lenny) or before. So default to using
util-linux-ng, instead of trying to test the version number of
libblkid1.
Lenny was released in February, 2009, and the current stable Debian
release is 7.x, so it is two stable releases back as of February 2014.
In the unlikely case someone needs to build a modern version of
e2fsprogs on a version of Debian which is five years old or older, can
create the file Debian/rules.custom with the line:
UTIL_LINUX_NG = no
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Also use angle brackets for the #include of dirpaths.h to avoid the
need to manually massage the Makefile.in for the util directory. This
is needed because we have to create a fake dirpaths.h file in the util
directory. The fake dirpaths.h file is rquired to break the circular
dependency caused by util/subst creating dirpaths.h, while
util/subst.c is including config.h, which includes dirpaths.h.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The getopt() function will never let optarg be NULL (at least without
using the GNU double-colon extension, which we don't use because it's
not portable), so don't bother checking for that case. It's harmless,
but it triggers a Coverity warning elsewhere, since it thinks optarg
could in fact be NULL.
Addresses-Coverity-Id: #1049156
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Initialize the on-disk structure before we fill it in, to avoid the
following valgrind warning:
Conditional jump or move depends on uninitialised value(s)
at 0x4323A8: qtree_entry_unused (quotaio_tree.c:40)
by 0x431218: v2r1_mem2diskdqblk (quotaio_v2.c:85)
by 0x432409: qtree_write_dquot (quotaio_tree.c:336)
by 0x431136: v2_commit_dquot (quotaio_v2.c:264)
by 0x42FB63: quota_write_inode (mkquota.c:126)
by 0x408BE6: create_quota_inodes (mke2fs.c:2466)
by 0x409A2D: main (mke2fs.c:2850)
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Add the extended options packed_meta_blocks and journal_location_front
which causes mke2fs to place the metadata blocks at the beginning of
the file system.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
In practice, it is **extremely** rare for users to try to use more
than the first backup superblock located at the beginning of block
group #1. (i.e., at block number 32768 for file systems with a 4k
block size). This new compat feature restricts the backup superblock
to block group #1 and the last block group in the file system.
Aside from reducing the overhead of the file system by a small number
of blocks, by eliminating the rest of the backup superblocks, it
allows us to have a much more flexible metadata layout. For example,
we can force all of the allocation bitmaps and inode table blocks to
the beginning of the disk, which allows most of the disk to be
exclusively used for contiguous data blocks.
This simplifies taking advantage of certain HDD specific features,
such as Shingled Magnetic Recording (aka Shingled Drives), and the
TCG's OPAL Storage Specification where having a simple mapping between
LBA block ranges and the data blocks used by the file system can make
life much simpler.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Instead of iterating over the allocation bitmap using
ext2fs_test_block_bitmap2(), bit by bit, use
ext2fs_find_first_set_block_bitmap2() instead.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If there are hundreds of thousands of blocks which are in use before
the first free block, it is much, MUCH faster to use
ext2fs_find_first_zero_block_bitmap2() instead of searching the
allocation bitmap bit by bit.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
By using ext2fs_mark_block_bitmap_range2 and/or
ext2fs_block_alloc_stats_range(), we can significantly speed up the
time needed by mke2fs to allocate the inode table.
For example, the CPU time needed to run the command "mke2fs -t ext4
/tmp/foo.img 32T" (where tmpfs was mounted on /tmp) was decreased from
21.7 CPU seconds down to under 1.7 seconds.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This function is more efficient than using ext2fs_block_alloc_stats2()
for each block in a range. The efficiencies come from being able to
set a block range in the block bitmap at once, and from being update
the block group descriptors once per block group. Especially now that
we are checksuming the block group descriptors, and we are using red
black trees for the allocation bitmaps, these changes can make a huge
difference in the CPU time used by mke2fs when creating very large
file systems.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Commit 8e44eb64bb (libext2fs: mark group data blocks when loading
block bitmap) simplified check_block_uninit since we are now
initializing the bitmap when it is loaded from disk. It left some
variables which were being set but never used, however. In addition,
since we only need check_block_uninit() to clear the block bitmap's
uninit flag, rename it to clear_block_uninit(), and only call it once
we have found a free block in ext2fs_new_blocks2().
This cleans up the code some and optimizes things if we need to search
multiple block groups trying to find a free block.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: Darrick J. Wong <darrick.wong@oracle.com>
When building tst_bitmaps, enable #define DEBUG_RB, so we are
always testing the sanity of the in-memory representation of the
bitmap when using red-black trees as part of a "make check" run.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Move the error checking into the the generic bitmap code, and add
support for bitmaps with cluster_bits set.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When inserting the first extent into an empty inode, the
ext2fs_extent_insert() leaves path->left set to 1 instead of 0. Since
path->curr is pointing at the last (only) extent in the file,
path->left should be 0.
This is mostly harmless, and gets corrected fairly quickly if the
calling applicaton jumps to a different part of the extent tree ---
for example, by calling ext2fs_extent_goto(), or calling
ext2fs_extent_get with the flags argument set to EXT2_EXTENT_ROOT.
Which is why we hadn't noticed this problem until now.
However, if you insert four extents using ext2fs_extent_insert, the
fourth insert will end up copying too many bytes in the i_block[]
array, since path->left is one larger than it should be. This results
in the inode fields i_generation, i_file_acl, and i_size_high getting
zeroed out.
This problem can be replicated as follows:
% cp /dev/null /tmp/foo.img
% mke2fs -F -t ext4 /tmp/foo.img 100
% debugfs -w /tmp/foo.img
debugfs: write /dev/null foo
debugfs: set_inode_field foo i_size_hi 1
debugfs: stat foo
<----- note that the inode's size is 4294967296
debugfs: extent_open foo
debugfs (extent ino 12): insert --after 0 1 100
debugfs (extent ino 12): insert --after 1 1 101
debugfs (extent ino 12): insert --after 2 1 102
debugfs (extent ino 12): insert --after 3 1 103
debugfs (extent ino 12): extent_close
debugfs: stat foo
<----- note that the inode's size is now 0
debugfs: quit
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Add functions which try to find the first set block or inode in a
bitmap. This is useful when trying to allocate a range of blocks
efficiently.
Like the find_first_zero family of functions, provide a generic O(N)
search function which will be used if there is no optimized version
provided by the red-black tree or bitarray functions.
Also, expand the test cases for ext2fs_find_first_zero_*() functions.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Now that libext2fs marks group metadata in the fs block bitmap, adjust
the expected test output to reflect expanded use of block_uninit and
the fact debugfs no longer prints block bitmap data that fails to
account for group data blocks.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Since the beginning of the uninit_bg feature, the kernel[1] and
e2fsck[2] have always been careful to detect the presence of the
BLOCK_UNINIT flag, and compute a block bitmap with any group metadata
blocks marked in that bitmap. With that in mind, I think it's safe to
say that this is a design feature of uninit_bg.
Now that we've trained libext2fs to have this same behavior whenever
it's loading a block bitmap, we no longer need to unset BLOCK_UNINIT
for a group that contains only its own group metadata -- kernel,
e2fsck, and e2fsprogs will handle this correctly.
[1] kernel git 717d50e4971b81b96c0199c91cdf0039a8cb181a
"Ext4: Uninitialized Block Groups"
[2] e2fsprogs git f5fa20078b
"Add support for EXT2_FEATURE_COMPAT_LAZY_BG"
Reported-by: Akira Fujita <a-fujita@rs.jp.nec.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Since libext2fs now detects a BLOCK_UNINIT group and calculates the
group's block bitmap, we no longer need to emulate this behavior in
e2fsck. We can simply compare the found block map against the
filesystem's, and proceed from there.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The kernel[1] and e2fsck[2] both react to a BLOCK_UNINIT group by
calculating the block bitmap that's needed to show all the group
blocks for that group (if any) and using that. However, when reading
bitmaps from disk, libext2fs simply imports a block of zeroes into the
bitmap, without bothering to check for group blocks. This erroneous
behavior results in the filesystem having a block bitmap that does not
accurately reflect disk contents, and worse yet makes it seem as
though superblocks, group descriptors, bitmaps, and inode tables are
"free" space on disk.
So, fix the block bitmap loading routines to calculate the correct
block bitmap for all groups and load it into the main fs block bitmap.
This also fixes bogus debugfs output such as:
Group 1: (Blocks 8193-16384) [INODE_UNINIT, BLOCK_UNINIT]
Checksum 0x1310, unused inodes 512
Backup superblock at 8193, Group descriptors at 8194-8217
Reserved GDT blocks at 8218-8473
Block bitmap at 283 (bg #0 + 282), Inode bitmap at 299 (bg #0 + 298)
Inode table at 442-569 (bg #0 + 441)
7911 free blocks, 512 free inodes, 0 directories, 512 unused inodes
Free blocks: 8193-16384
Free inodes: 513-1024
Notice how the "free blocks" range includes the backup sb & GDT area
and doesn't match the free block count.
Worse yet, debugfs' testb command will report those group descriptor
blocks as not being in use unless the user also instructs debugfs to
find a free block first. That is a rather surprising result:
debugfs: testb 8194
Block 8194 not in use
debugfs: ffb 1 16380
Free blocks found: 16380
debugfs: testb 8194
Block 8194 marked in use
Also, remove the part of check_block_uninit() that "fixes" the bitmap
since we're doing that at bitmap load time now.
[1] kernel git 717d50e4971b81b96c0199c91cdf0039a8cb181a
"Ext4: Uninitialized Block Groups"
[2] e2fsprogs git f5fa20078b
"Add support for EXT2_FEATURE_COMPAT_LAZY_BG"
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
On a filesystem with 1K blocks and meta_bg enabled, opening a
filesystem with automatic superblock detection tries to compensate for
the fact that the superblock lives in block 1. However, the method by
which this is done is later misinterpreted to mean "read the backup
group descriptors", which is not what we want in this case.
Therefore, in ext2fs_open3() separate the 'group zero' adjustment into
its own variable so that we don't get fed backup group descriptors
when we try to load meta_bg group descriptors.
Furthermore, enhance ext2fs_descriptor_block_loc2() to perform its own
group zero correction. The other caller of this function neglects to
do any group-zero correction of their own, so this fixes them too.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
If ext2fs_descriptor_block_loc2() is called with a meta_bg filesystem
and group_block is not the normal value, the function will return the
location of the backup group descriptor block in the next block group.
Unfortunately, it fails to account for the possibility that the backup
group contains a backup superblock but the regular superblock does
not. This is the case with block groups 48-49 on a meta_bg fs with 1k
blocks; in this case, libext2fs will fail to open the filesystem.
Therefore, teach the function to adjust for superblocks in the backup
group, if necessary.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Baruch Siach