Whamcloud - gitweb
tools/e2fsprogs.git
10 years agoe2fsck: fix printf conversion specs in ea_refcount.c
Eric Whitney [Tue, 24 Dec 2013 06:58:09 +0000 (01:58 -0500)]
e2fsck: fix printf conversion specs in ea_refcount.c

Commit 130e961a6fb9d38997c83c6397cf9ddb5ed2491f changed the type
used to represent block numbers in ea_refcount.c from blk_t to blk64_t
to add support for 64 bit extended attribute refcounting.  We also
need to adjust printf conversion specs that now don't match their new
blk64_t arguments.  This will silence compiler warnings seen when
"make check" is run and will avoid truncation of printed values.

Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agomke2fs: further updates for mke2fs(8) man page
Andreas Dilger [Mon, 23 Dec 2013 21:09:49 +0000 (16:09 -0500)]
mke2fs: further updates for mke2fs(8) man page

The mke2fs(8) man page was updated in 4727c67dc2, but needs some
more clear descriptions for extra_isize and metadta_csum features.
The uninit_bg feature is supported by all ext4-capable kernels,
and does have a slow e2fsck pass for newly-formatted filesystems,
so remove the caveat.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agodebugfs, e2fsck: fix s_desc_size handling
Andreas Dilger [Mon, 23 Dec 2013 21:04:46 +0000 (16:04 -0500)]
debugfs, e2fsck: fix s_desc_size handling

The s_desc_size in the superblock specifies the group descriptor
size in bytes, but in various places the EXT4_FEATURE_INCOMPAT_64BIT
flag implies that the descriptor size is EXT2_MIN_DESC_SIZE_64BIT
(64 bytes) instead of checking the actual size.  In other places,
the s_desc_size field is used without checking for INCOMPAT_64BIT.

In the case of ext2fs_group_desc() the s_desc_size was being ignored,
and assumed to be sizeof(struct ext4_group_desc), which would result
in garbage for any but the first group descriptor.  Similarly, in
ext2fs_group_desc_csum() and print_csum() they assumed that the
maximum group descriptor size was sizeof(struct ext4_group_desc).
Fix these functions to use the actual superblock s_desc_size if
INCOMPAT_64BIT.

Conversely, in ext2fs_swap_group_desc2() s_desc_size was used
without checking for INCOMPAT_64BIT being set.

The e2fsprogs behaviour is different than that of the kernel,
which always checks INCOMPAT_64BIT, and only uses s_desc_size to
determine the offset of group descriptors and what range of bytes
to checksum.

Allow specifying the s_desc_size field at mke2fs time with the
"-E desc_size=NNN" option.  Allow a power-of-two s_desc_size
value up to s_blocksize if INCOMPAT_64BIT is specified.  This
is not expected to be used by regular users at this time, so it
is not currently documented in the mke2fs usage or man page.

Add m_desc_size_128, f_desc_size_128, and f_desc_bad test cases to
verify mke2fs and e2fsck handling of larger group descriptor sizes.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: verify s_desc_size is power-of-two value
Andreas Dilger [Mon, 23 Dec 2013 21:03:41 +0000 (16:03 -0500)]
e2fsck: verify s_desc_size is power-of-two value

Add a LOG2_CHECK mode for check_super_value() so that it is easy
to verify values that are supposed to be power-of-two values
(s_desc_size and s_inode_size so far).  In ext2fs_check_desc()
also check for a power-of-two s_desc_size.

Print out s_desc_size in debugfs "stats" and dumpe2fs output, if
it is non-zero.

It turns out that the s_desc_size validation in check_super_block()
is not currently used by e2fsck, because the group descriptors are
verified earlier by ext2fs_check_desc(), and even without an
explicit check of s_desc_size the group descriptors fail to align
correctly on disk.  It makes sense to keep the check_super_block()
regardless, in case the code changes at some point in the future.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agotests: delete temporary resize test files
Andreas Dilger [Mon, 23 Dec 2013 21:01:51 +0000 (16:01 -0500)]
tests: delete temporary resize test files

Since commit 5ad07acad if $TMP cannot hold large test filesystems
for resize testing the resize_test creates temporary test files
in the local working directory.  Since it overrides TMPFILE locally
the calling program does not delete the generated file correctly.

Delete the large $TMPFILE within resize_test if it passes, but leave
it for debugging if the test failed.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: read only parameter incorrectly compared
Johan Erlandsson [Thu, 19 Dec 2013 18:55:51 +0000 (13:55 -0500)]
e2fsck: read only parameter incorrectly compared

Don't check for lost+found in read only mode.

[Note: this patch was originally made against 1.41.14 version of
e2fsprogs found as part of the AOSP (Android Open Source Program)
tree.  My Signed-off-by relies on the fact that the original patch
author would have had to have filed a contribution agreement with Open
Handset Alliance before this commit before this commit was allowed
into the AOSP tree. -- tytso]

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoClean up sparse warnings
Theodore Ts'o [Mon, 16 Dec 2013 23:56:36 +0000 (18:56 -0500)]
Clean up sparse warnings

Mostly by adding static and removing excess extern qualifiers.  Also
convert a few remaining non-ANSI function declarations to ANSI.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsprogs: mke2fs manpage updates
Uri Corin [Mon, 16 Dec 2013 05:53:35 +0000 (00:53 -0500)]
e2fsprogs: mke2fs manpage updates

I've added documentation for all available filesystem features
settable by mke2fs.
* Features now include the kernel constant attached to the function.
* Some pre-existing feature explanations have been expanded based on
ext4.wiki.kernel.org.

[ Edited by tytso for stylistic and correctness. ]

Addresses-Debian-Bug: #586218

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
10 years agomke2fs: disallow creating FS on a loop mounted file with no option
Kazuya Mio [Mon, 16 Dec 2013 05:48:12 +0000 (00:48 -0500)]
mke2fs: disallow creating FS on a loop mounted file with no option

When /etc/mtab is a symlink of /proc/mounts, mke2fs without -FF option
can create a filesystem on the image file that is mounted.
According to mke2fs man page, we should specify -FF option in this case.

This patch protects filesystem from unintended mke2fs caused by human error.

How to reproduce:
  # mke2fs -t ext4 -Fq fs.img
  # mount -o loop fs.img /mnt/mp1
  # mke2fs -t ext4 -Fq fs.img && echo "mke2fs success"
  mke2fs success

Signed-off-by: Kazuya Mio <k-mio@sx.jp.nec.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agomke2fs: fix force option to create filesystem on mounted device
Kazuya Mio [Mon, 16 Dec 2013 05:42:22 +0000 (00:42 -0500)]
mke2fs: fix force option to create filesystem on mounted device

According to mke2fs man, we can create a filesystem on the mounted
device when -FF option is specified.

However, currently we have to specify -F option third to force mke2fs.
This patch fixes the problem.

Signed-off-by: Kazuya Mio <k-mio@sx.jp.nec.com>
Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: try implied cluster allocation when expanding a dir
Darrick J. Wong [Mon, 16 Dec 2013 04:54:07 +0000 (23:54 -0500)]
e2fsck: try implied cluster allocation when expanding a dir

When we're expanding a directory, check to see if we're doing an
implied cluster allocation; if so, we don't need to allocate a new
block, and we certainly don't need to update the summary counts.

Reported-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: print cluster ranges when encountering bitmap errors
Darrick J. Wong [Mon, 16 Dec 2013 04:53:32 +0000 (23:53 -0500)]
e2fsck: print cluster ranges when encountering bitmap errors

If pass5 finds bitmap errors in a range of clusters, don't print each
cluster number individually when we could print only the start and end
cluster number.  e2fsck already does this for the non-bigalloc case.

Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: only release clusters when shortening a directory during a rehash
Darrick J. Wong [Wed, 11 Dec 2013 01:23:12 +0000 (17:23 -0800)]
e2fsck: only release clusters when shortening a directory during a rehash

When the rehash process is running on a bigalloc filesystem, it
compresses all the directory entries and hash structures into the
beginning of the directory file and then uses block_iterate3() to free
the blocks off the end of the file.  It seems to call
ext2fs_block_alloc_stats2() for every block in a cluster, which is
unfortunate because this function allocates and frees entire clusters
(and updates the summary counts accordingly).  In this case e2fsck
writes out incorrect summary counts.

Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: don't update the summary counts when doing implied cluster allocation
Darrick J. Wong [Mon, 16 Dec 2013 04:50:52 +0000 (23:50 -0500)]
libext2fs: don't update the summary counts when doing implied cluster allocation

When we're appending a block to a directory file or the journal file,
and the new block is part of a cluster that has already been allocated
to the file (implied cluster allocation), don't update the bitmap or
the summary counts because that was performed when the cluster was
allocated.

Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: only punch complete clusters
Darrick J. Wong [Mon, 16 Dec 2013 04:50:03 +0000 (23:50 -0500)]
libext2fs: only punch complete clusters

When bigalloc is enabled, using ext2fs_block_alloc_stats2() to free
any block in a cluster has the effect of freeing the entire cluster.
This is problematic if a caller instructs us to punch, say, blocks
12-15 of a 16-block cluster, because blocks 0-11 now point to a "free"
cluster.

The naive way to solve this problem is to see if any of the other
blocks in this logical cluster map to a physical cluster.  If so, then
we know that the cluster is still in use and it mustn't be freed.
Otherwise, we are punching the last mapped block in this cluster, so
we can free the cluster.

The implementation given only does the rigorous checks for the partial
clusters at the beginning and end of the punching range.

Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoext2fs,blkid: delete unused LIST_HEAD() macro
Andreas Dilger [Mon, 16 Dec 2013 04:26:38 +0000 (23:26 -0500)]
ext2fs,blkid: delete unused LIST_HEAD() macro

The LIST_HEAD() macro conflicts with the <sys/queue.h> declaration
of the same name.  Delete the unused LIST_HEAD() macro from the
libext2fs and libblkid headers to avoid compiler warnings.  It can
be replaced by INIT_LIST_HEAD() or LIST_HEAD_INIT() if needed.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoext2fs: quiet compiler warnings
Andreas Dilger [Mon, 16 Dec 2013 04:26:25 +0000 (23:26 -0500)]
ext2fs: quiet compiler warnings

Include ext2fsP.h in fileio.c for ext2fs_file_block_offset_too_big()
declaration.  Fix up the declaration to mark it extern in the header.

Include <strings.h> header for strcasecmp() in tune2fs.c if available,
as described in the strcasecmp(3) man page, instead of doing this
indirectly by declaring _BSD_SOURCE and getting it from <string.h>.

If CONFIG_QUOTA is undefined, parse_quota_opts() is unused in
tune2fs.c so #ifdef it out.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agodebugfs: clean up debugfs man page and help text
Andreas Dilger [Mon, 16 Dec 2013 03:12:43 +0000 (22:12 -0500)]
debugfs: clean up debugfs man page and help text

Clean up the formatting of the debugfs(8) man page:
- add command aliases to the main descriptions
- use bold for sub-commands and italics for their arguments
- remove trailing spaces
- add "dirty" sub-command
- clarify "block_dump" and "zap_block" sub-commands

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agobuild: quiet LLVM non-literal string format warning
Andreas Dilger [Mon, 16 Dec 2013 03:11:40 +0000 (22:11 -0500)]
build: quiet LLVM non-literal string format warning

Compiling with LLVM generates a large number of warnings due
to the use of _() for wrapping strings for i18n:

    warning: format string is not a string literal
          (potentially insecure) [-Wformat-security]
    ./nls-enable.h:4:14: note: expanded from macro '_'
    #define _(a) (gettext (a))
                 ^~~~~~~~~~~~

These warnings are fixed by using "%s" as the format string,
and then _() is used as the string argument.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agobuild: use long long for __u64 by default
Andreas Dilger [Mon, 16 Dec 2013 03:09:24 +0000 (22:09 -0500)]
build: use long long for __u64 by default

Don't print a verbose configure error in parse-types.h if
<asm/types.h> missing and __[SU]*_TYPEDEF is unset.  This is
always the case for non-Linux builds.

The printf formatting strings all use "%llu" for printing 64-bit
values and this it produces a large number of warnings if __u64
is defined as "unsigned long".  If __U64_TYPEDEF isn't set use
"unsigned long long" for __u64 in ext2-types.h and blkid-types.h
by default instead of using "unsigned long".

Fix a few places where "%d" or "%u" or "%Lu" were used to print a
64-bit value, by converting them to use "%lld" or "%llu" instead.

Fix a few places where "%lu" was used to print .tv_usec, by casting
the variable to "(long)" since .tv_usec is "int" on some systems.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agotune2fs: forbid changing uuid on an uninit_bg filesystem
Darrick J. Wong [Sun, 15 Dec 2013 01:51:04 +0000 (20:51 -0500)]
tune2fs: forbid changing uuid on an uninit_bg filesystem

The old uninit_bg checksums depend on the UUID, so prohibit changes to
the UUID if a checksumming filesystem is mounted, because this
introduces a nasty race where the kernel and tune2fs are both trying
to rewrite group descriptors at the same time, with different ideas
about what the UUID is.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: use ext2fs_punch() to truncate quota file
Darrick J. Wong [Sun, 15 Dec 2013 00:52:40 +0000 (19:52 -0500)]
libext2fs: use ext2fs_punch() to truncate quota file

Use the new ext2fs_punch() call to truncate the quota file.  This also
eliminates the need to fix it to work with bigalloc.

Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: clamp block-map punch range end to 2^32 blocks
Darrick J. Wong [Sun, 15 Dec 2013 00:46:50 +0000 (19:46 -0500)]
libext2fs: clamp block-map punch range end to 2^32 blocks

In the ^extent case, passing ~0ULL as the 'end' parameter to
ext2fs_punch() causes the (end - start + 1) calculation to overflow to
zero.  Since the old-style mapped block files cannot have more than
2^32 blocks, just clamp it to ~0U.

This fixes a regression in t_quota_2off with the patch "libext2fs: use
ext2fs_punch() to truncate quota file" applied.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: zero block contents past EOF when setting size
Darrick J. Wong [Thu, 12 Dec 2013 18:31:35 +0000 (13:31 -0500)]
libext2fs: zero block contents past EOF when setting size

When we set the file size, find the block containing EOF, and zero
everything in that block past EOF so that we can't return stale data
if we ever use fallocate or truncate to lengthen the file.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: in rehash, mark newly allocated extent blocks as found
Darrick J. Wong [Wed, 11 Dec 2013 01:22:07 +0000 (17:22 -0800)]
e2fsck: in rehash, mark newly allocated extent blocks as found

When we're rehashing directories, it's possible that an extent block
(or a map block) could be (silently) allocated by the underlying
libext2fs when expanding the directory.  This silent allocation is not
captured in block_found_map, which is disastrous if later the rehash
process expands another directory and uses that same block from
before without realizing that it's now in use.

Therefore, if we notice that the free block count has dropped by more
than what e2fsck allocated itself during the expansion, we iterate the
directory's blocks a second time to ensure that these silent
allocations are marked in the found blocks bitmap.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: don't error out when punching a totally sparse file
Darrick J. Wong [Thu, 12 Dec 2013 18:25:40 +0000 (13:25 -0500)]
libext2fs: don't error out when punching a totally sparse file

If we're asked to punch a file with no data blocks mapped to it and a
non-zero length, we don't need to do any work in ext2fs_punch_extent()
and can return success.  Unfortunately, the extent_get() function
returns "no current node" because it (correctly) failed to find any
extents, which is bubbled up to callers.  Since no extents being found
is not an error in this corner case, fix up ext2fs_punch_extent() to
return 0 to callers.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: fix punching extents when there are no left extents
Darrick J. Wong [Thu, 12 Dec 2013 18:23:51 +0000 (13:23 -0500)]
libext2fs: fix punching extents when there are no left extents

When deleting an entire extent, we cannot always slip to the previous
leaf extent because there might not /be/ a previous extent.
Attempting to correct for that error by asking for the 'current' leaf
extent also doesn't work, because the failed attempt to change to the
previous extent leaves us with no current extent.

Fix this problem by recording the lblk of the next extent before
deleting the current extent and _goto()ing to the next extent after
the deletion.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: fail fileio write if we can't allocate a block
Darrick J. Wong [Thu, 12 Dec 2013 18:19:46 +0000 (13:19 -0500)]
libext2fs: fail fileio write if we can't allocate a block

If we're using ext2fs_file_write() to write to a hole in a file,
ensure that we can actually allocate the block before updating i_size.
In other words, don't update i_size and don't return success if we hit
an error while allocating space.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: fix a broken close() test
Darrick J. Wong [Thu, 12 Dec 2013 18:17:58 +0000 (13:17 -0500)]
libext2fs: fix a broken close() test

Zero is a valid file descriptor, so close it.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: fix memory leaks (on error paths)
Darrick J. Wong [Wed, 11 Dec 2013 01:21:34 +0000 (17:21 -0800)]
libext2fs: fix memory leaks (on error paths)

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: check return values
Darrick J. Wong [Thu, 12 Dec 2013 18:14:50 +0000 (13:14 -0500)]
libext2fs: check return values

Fix up a few places where we ignore return values.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibquota: fix memory leak (on error path)
Darrick J. Wong [Thu, 12 Dec 2013 18:13:51 +0000 (13:13 -0500)]
libquota: fix memory leak (on error path)

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibss: fix memory handling errors
Darrick J. Wong [Thu, 12 Dec 2013 18:12:09 +0000 (13:12 -0500)]
libss: fix memory handling errors

Fix memory allocation calculations and check for NULL pointer returns.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibss: fix fd error handling
Darrick J. Wong [Thu, 12 Dec 2013 18:11:06 +0000 (13:11 -0500)]
libss: fix fd error handling

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agodumpe2fs: check return values
Darrick J. Wong [Thu, 12 Dec 2013 18:10:29 +0000 (13:10 -0500)]
dumpe2fs: check return values

Check the return values from ext2fs_get_block_bitmap_range2(); if an
error happened, print that and don't print garbage bitmap.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2image: check return value from check_if_mounted()
Darrick J. Wong [Thu, 12 Dec 2013 18:09:22 +0000 (13:09 -0500)]
e2image: check return value from check_if_mounted()

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe4defrag: defensively check results of sysconf(_SC_PAGESIZE)
Darrick J. Wong [Thu, 12 Dec 2013 18:08:45 +0000 (13:08 -0500)]
e4defrag: defensively check results of sysconf(_SC_PAGESIZE)

sysconf(_SC_PAGESIZE) will probably never return an error, but just in
case it does, we shouldn't pass what looks like a huge number to
sync_file_range() and posix_fadvise().

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe4defrag: don't crash if umounts the filesystem races with us
Darrick J. Wong [Thu, 12 Dec 2013 18:07:33 +0000 (13:07 -0500)]
e4defrag: don't crash if umounts the filesystem races with us

If someone umounts the filesystem between statfs64() and the getmntent()
iteration, we can exit the loop having never set mnt_type, and strcmp
can crash.  Fix the potential NULL dereference.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agomke2fs: don't leak memory (on error path)
Darrick J. Wong [Thu, 12 Dec 2013 18:06:44 +0000 (13:06 -0500)]
mke2fs: don't leak memory (on error path)

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agomisc: don't leak file descriptors (on error paths)
Darrick J. Wong [Thu, 12 Dec 2013 18:05:50 +0000 (13:05 -0500)]
misc: don't leak file descriptors (on error paths)

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: fix memory leaks (on error path)
Darrick J. Wong [Thu, 12 Dec 2013 17:57:48 +0000 (12:57 -0500)]
e2fsck: fix memory leaks (on error path)

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agodebugfs: don't leak mmp_s memory (on error path)
Darrick J. Wong [Thu, 12 Dec 2013 17:52:00 +0000 (12:52 -0500)]
debugfs: don't leak mmp_s memory (on error path)

ext2fs_free_mem() takes a pointer to a pointer, similar to
ext2fs_get_mem().  Improve the documentation, and fix debugfs.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agodebugfs: don't leak fd when calling dump_file
Darrick J. Wong [Thu, 12 Dec 2013 17:50:58 +0000 (12:50 -0500)]
debugfs: don't leak fd when calling dump_file

The caller of dump_file provides a fd to write to, so the caller
should also dispose of the fd.  Also, the fd never gets closed if
preserve=1.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: fileio should use 64bit io routines
Darrick J. Wong [Thu, 12 Dec 2013 17:50:04 +0000 (12:50 -0500)]
libext2fs: fileio should use 64bit io routines

When reading or writing file blocks, use the IO manager routines that
can handle 64bit block numbers.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agodebugfs: handle 64bit block numbers
Darrick J. Wong [Thu, 12 Dec 2013 17:49:14 +0000 (12:49 -0500)]
debugfs: handle 64bit block numbers

debugfs should use strtoull wrappers for reading block numbers from
the command line.  "unsigned long" isn't wide enough to handle block
numbers on 32bit platforms.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: make symlinks safe for 64bit blocks and extents
Darrick J. Wong [Thu, 12 Dec 2013 17:48:33 +0000 (12:48 -0500)]
libext2fs: make symlinks safe for 64bit blocks and extents

If we have to create a big symlink (i.e. one that doesn't fit into
i_block[]), we are not 64bit block safe and the namei code does not
handle extents at all.  Fix both.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agodebugfs: fix various minor bogosity
Darrick J. Wong [Thu, 12 Dec 2013 17:43:36 +0000 (12:43 -0500)]
debugfs: fix various minor bogosity

We should really use the ext2fs memory allocator functions in
copy_file(), and we really should return a value if there's allocation
problems.

Also fix up a minor bogosity in an error message.

Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Cc: Robert Yang <liezhi.yang@windriver.com>
Cc: Darren Hart <dvhart@linux.intel.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: fix "a" vs "an" in the error catalog
Darrick J. Wong [Thu, 12 Dec 2013 17:41:58 +0000 (12:41 -0500)]
libext2fs: fix "a" vs "an" in the error catalog

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: don't allow ridiculously large logical block numbers
Darrick J. Wong [Thu, 12 Dec 2013 17:40:31 +0000 (12:40 -0500)]
libext2fs: don't allow ridiculously large logical block numbers

Forbid clients from trying to map logical block numbers that are
larger than the lblk->pblk data structures are capable of handling.
While we're at it, don't let clients set the file size to a number
that's beyond what can be mapped.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agodebugfs: fix init_filesys help text
Darrick J. Wong [Thu, 12 Dec 2013 17:29:07 +0000 (12:29 -0500)]
debugfs: fix init_filesys help text

The help text for debugfs' init_filesys command is incorrect; the
second parameter is the size of the filesystem in blocks, not the size
of an individual filesystem block.  There is in fact no way to set
that parameter.

Reported-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agomke2fs: load configfile blocksize setting before 64bit checks
Darrick J. Wong [Wed, 11 Dec 2013 01:18:37 +0000 (17:18 -0800)]
mke2fs: load configfile blocksize setting before 64bit checks

mke2fs has a series of checks to ensure that we don't create a
filesystem too big for its blocksize -- if auto-64bit is on, then it
turns on 64bit; otherwise it complains.  Unfortunately, it performs
these checks before looking in mke2fs.conf for a blocksize, which
means that the checks are incorrect if the user specifies a non-4096
blocksize in the config file and says nothing on the command line.
The bug also has the effect of mandating a 4k block size on any block
device larger than 4T in that situation.  Therefore, read the block
size from the config file before performing the 64bit checks.

Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: fix tests that set LARGE_FILE
Darrick J. Wong [Wed, 11 Dec 2013 01:18:27 +0000 (17:18 -0800)]
libext2fs: fix tests that set LARGE_FILE

For each site where we test for a large file (> 2GB) and set the
LARGE_FILE feature, use a helper function to make the size test
consistent with the test that's in e2fsck.  This fixes the fsck
complaints when we try to create a 2GB journal (not so hard with 64k
block size) and fixes the incorrect test in fileio.c.

Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: don't overflow when punching indirect blocks with large blocks
Darrick J. Wong [Thu, 12 Dec 2013 17:06:03 +0000 (12:06 -0500)]
libext2fs: don't overflow when punching indirect blocks with large blocks

On a FS with a rather large blockize (> 4K), the old block map
structure can construct a fat enough "tree" (or whatever we call that
lopsided thing) that (at least in theory) one could create mappings
for logical blocks higher than 32 bits.  In practice this doesn't
happen, but the 'max' and 'iter' variables that the punch helpers use
will overflow because the BLOCK_SIZE_BITS shifts are too large to fit
a 32-bit variable.  The current variable declarations also cause punch
to fail on TIND-mapped blocks even if the file is < 16T.  So enlarge
the fields to fit.

Yes, this is an obscure corner case, but it seems a little silly if we
can't punch a file's block 300,000,000 on a 64k-block filesystem.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: fix problem comments to match actual message
Andreas Dilger [Thu, 12 Dec 2013 01:10:39 +0000 (20:10 -0500)]
e2fsck: fix problem comments to match actual message

Fix the e2fsck problem comments to match the actual message printed,
so that it is possible to find the problem code when searching by
the message.

Signed-off-by: Andreas Dilger <andreas.dilger@intel.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoUpdate config.{guess,sub} to the latest version
Theodore Ts'o [Wed, 11 Dec 2013 04:12:38 +0000 (23:12 -0500)]
Update config.{guess,sub} to the latest version

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agodebugfs: update usage message for open and filefrag
Zheng Liu [Mon, 9 Dec 2013 02:01:09 +0000 (21:01 -0500)]
debugfs: update usage message for open and filefrag

The ussage message for the open and filefrag commands were missing
options; this commit fixes this.

Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agodebugfs: fix range mode of freei
Lev Solomonov [Thu, 5 Dec 2013 02:55:22 +0000 (21:55 -0500)]
debugfs: fix range mode of freei

commit 2ae5d1fdb82b31bf270c4699c59a087c846db2f0 was supposed to teach
'seti' and 'freei' to act on a range of consecutive inodes. apparently
only 'seti' has learned: 'freei' doesn't advance the ino, repeatedly
acting on the same one instead.

Signed-off-by: Lev Solomonov <solo@primarydata.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agommp: fix 64-bit handling of s_mmp_block
Andreas Dilger [Wed, 4 Dec 2013 01:02:12 +0000 (20:02 -0500)]
mmp: fix 64-bit handling of s_mmp_block

Fix the checking of s_mmp_block in e2fsck_pass1() and
ext2fs_mmp_read() to handle the high 32 bits of s_blocks_count.
Remove redundant check of s_mmp_block in do_dump_mmp() right before
ext2fs_mmp_read() is called.

Also fix s_blocks_count_hi in check_backup_super_block(), since it
cannot use the ext2fs_blocks_count() helper easily.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: fix function declarations to match definition
Darrick J. Wong [Tue, 3 Dec 2013 19:24:53 +0000 (14:24 -0500)]
libext2fs: fix function declarations to match definition

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agotests: check backing filesystem can handle large file
Andreas Dilger [Tue, 3 Dec 2013 05:11:55 +0000 (00:11 -0500)]
tests: check backing filesystem can handle large file

The new resize tests create 2TB test files, but tmpfs in kernels
before 3.1 have a max file size of 256GB.  Ext3 may also have
a size limit for smaller blocksize filesystems.

Fix the resize_test script to verify that $TMPFILE can be resized
to the final test size, and if that fails try creating the file on
the local filesystem instead of in $TMPDIR.  If that cannot hold
the large filesystem, skip the test.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Tested-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
10 years agolibext2fs: fix printf conversion spec in tst_iscan.c
Eric Whitney [Tue, 3 Dec 2013 04:37:02 +0000 (23:37 -0500)]
libext2fs: fix printf conversion spec in tst_iscan.c

A recent patch to fix blk_t to blk64_t assignment mismatches in
e2fsprogs (commit 4dbfd79d1458ce1259b951377e341aeb6197f8c1) created
a printf conversion spec / argument type mismatch in tst_iscan.c.
Fix this to avoid truncation of the printed value and to silence
a compiler warning seen when "make check" is run.

Signed-off-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: use dgrp_t for block group numbers
Theodore Ts'o [Tue, 3 Dec 2013 04:21:31 +0000 (23:21 -0500)]
e2fsck: use dgrp_t for block group numbers

Make e2fsck consistently use dgrp_t for bloc group numbers to avoid
-Wconveresion noise.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: add support for 64-bit extended attribute block refcounting
Theodore Ts'o [Tue, 3 Dec 2013 04:07:32 +0000 (23:07 -0500)]
e2fsck: add support for 64-bit extended attribute block refcounting

If we have a 64-bit file system with extended attribute blocks, e2fsck
would not correctly handle EA blocks that were located beyond the
32-bit block number boundary.  Fix this by teaching
e2fsck/ea_refcount.c to use 64-bit block numbers.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: fix j_maxlen if the file system is exactly 1 << 32 blocks
Theodore Ts'o [Tue, 3 Dec 2013 03:26:58 +0000 (22:26 -0500)]
e2fsck: fix j_maxlen if the file system is exactly 1 << 32 blocks

If the external journal device has exactly 1 << 32 blocks,
journal->j_maxlen would get set to zero, which would cause e2fsck to
declare the journal to be invalid.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: add explicit casts to bitops.h
Theodore Ts'o [Tue, 3 Dec 2013 03:24:12 +0000 (22:24 -0500)]
libext2fs: add explicit casts to bitops.h

Add some explicit casts to silence some -Wconversion noise.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: add explicit casts to ext2fs.h
Theodore Ts'o [Tue, 3 Dec 2013 02:53:56 +0000 (21:53 -0500)]
libext2fs: add explicit casts to ext2fs.h

Add some explicit casts to silence some -Wconversion noise.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agodumpe2fs: fix printing of block offsets for 64-bit file systems
Theodore Ts'o [Tue, 3 Dec 2013 02:51:24 +0000 (21:51 -0500)]
dumpe2fs: fix printing of block offsets for 64-bit file systems

Use ext2fs_group_first_block2() instead of ext2fs_group_first_block()
to avoid dumpe2fs from printing crazy block offsets when we have block
numbers which are larger than 32 bits.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: use blk_t instead of blk64_t in check_resize_inode()
Theodore Ts'o [Tue, 3 Dec 2013 02:49:24 +0000 (21:49 -0500)]
e2fsck: use blk_t instead of blk64_t in check_resize_inode()

The resize inode only works on 32-bit block numbers, so use blk_t
instead of blk64_t.  This avoids some -Wconversion noise, and slims
the compiled code slightly, especially on 32-bit platforms.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: use errcode_t to suppress some -Wconversion warnings
Theodore Ts'o [Tue, 3 Dec 2013 02:37:10 +0000 (21:37 -0500)]
e2fsck: use errcode_t to suppress some -Wconversion warnings

We need to store some error codes using an int to keep recovery.c as
close as possible to the recovery.c source file in the kernel.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: use problem_t to suppress some -Wconversion warnings
Theodore Ts'o [Tue, 3 Dec 2013 01:52:43 +0000 (20:52 -0500)]
e2fsck: use problem_t to suppress some -Wconversion warnings

All code which stores a problem code should use the problem_t type.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agoe2fsck: use ext2fs_write_dir_block3() instead of ext2fs_write_dir_block()
Kit Westneat [Tue, 3 Dec 2013 00:11:52 +0000 (19:11 -0500)]
e2fsck: use ext2fs_write_dir_block3() instead of ext2fs_write_dir_block()

The use of ext2fs_write_dir_block() meant that attempts to fix
deleted/unused inodes in a directory would not be fixed for file
systems with 64-bit block numbers.  (And some random block with the
high 32-bits cleared would get corrupted.)

Fix a similar problem when expanding directories and when creating the
lost+found dirctory.

Signed-off-by: Kit Westneat <kwestneat@ddn.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
10 years agolibext2fs: fix some memory leaks with image file handling
Theodore Ts'o [Mon, 2 Dec 2013 19:52:31 +0000 (14:52 -0500)]
libext2fs: fix some memory leaks with image file handling

These memory leaks were discovered by using "valgrind
--leak-check=full" while running "e2image -I bar.img foo.e2i"

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
10 years agolibext2fs: set the fs block size to new_io in ext2fs_rewrite_to_io()
Kit Westneat [Mon, 2 Dec 2013 19:47:08 +0000 (14:47 -0500)]
libext2fs: set the fs block size to new_io in ext2fs_rewrite_to_io()

e2image manually opens a new IO channel, and then sets the file system
to use this new IO channel via ext2fs_rewrite+to_io().  We need to
make sure the IO channel is set to the file system's block size to
avoid some nasty buffer overruns.

[ Modified by tytso to use io_channel_set_blksize() ]

Signed-off-by: Kit Westneat <kwestneat@ddn.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>
10 years agoe2image: pass the correct size write_header
Theodore Ts'o [Mon, 2 Dec 2013 18:14:43 +0000 (13:14 -0500)]
e2image: pass the correct size write_header

Commit bf0449b1a654, which added the ability to write qcow2 files,
generalized the write_header() file to take the size of the header
structure which it writes out.  Unfortunately, it changed the call
which supported original e2image format to pass in fs->blocksize,
instead of the actual size of the e2image header structure (which is
substantially smaller than fs->blocksize).  This meant that we copied
in stack garbage into the e2image file, and it made valgrind quite
unhappy.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agoFix optional static code analysis using sparse
Theodore Ts'o [Thu, 24 Oct 2013 00:09:12 +0000 (20:09 -0400)]
Fix optional static code analysis using sparse

Commit 832cb612: "e2fsprogs: add (optional) sparse checking to the
build" breaks systems that are not using GNU make.  In addition, it
breaks if the developer tries to build in a subdirectory (i.e., if he
or she tries running "make" in the misc or e2fsck or lib/ext2fs
directory), since CHECK_CMD is not set.

Fix this by moving the sparse setup to MCONFIG.

Cc: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agolibext2fs: reject 64bit badblocks numbers
Darrick J. Wong [Wed, 23 Oct 2013 23:43:32 +0000 (19:43 -0400)]
libext2fs: reject 64bit badblocks numbers

Don't accept block numbers larger than 2^32 for the badblocks list,
and don't run badblocks on them either.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
11 years agomke2fs: don't let resize= turn on resize_inode when meta_bg is set
Theodore Ts'o [Fri, 18 Oct 2013 04:49:16 +0000 (21:49 -0700)]
mke2fs: don't let resize= turn on resize_inode when meta_bg is set

Passing the "-E resize=NNN" option to mke2fs sets the resize_inode
feature.  However, resize_inode and meta_bg are mutually exclusive;
unfortunately, we check this constraint before we parse the extended
options.  Fix this by moving this check after the calls
parse_extended_opts().

Reported-by: "Darrick J. Wong" <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agolibext2fs: stop iterating dirents when done linking
Darrick J. Wong [Fri, 18 Oct 2013 04:49:01 +0000 (21:49 -0700)]
libext2fs: stop iterating dirents when done linking

When we've succesfully linked an inode into a directory, we can stop
iterating the directory.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
11 years agoFix noquota build
Theodore Ts'o [Wed, 23 Oct 2013 16:51:32 +0000 (12:51 -0400)]
Fix noquota build

We need to build libquota even if the quota code is disabled.  This
fixes a build regression introduced by commit 43075b42bd: 'quota: fix
disabling quota, add quota tests".

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agomke2fs: clarify inode ratio in man page
Phillip Susi [Wed, 16 Oct 2013 03:37:51 +0000 (23:37 -0400)]
mke2fs: clarify inode ratio in man page

The man page still said it was not possible to change the number of
inodes on a filesystem after creating it.  You actually can by
resizing the fs, so clarify this language a bit.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
11 years agoquota: fix disabling quota, add quota tests
Niu Yawei [Mon, 14 Oct 2013 13:49:27 +0000 (09:49 -0400)]
quota: fix disabling quota, add quota tests

Update all superblock copies when disabling the quota feature.

Added basic tests for the quota feature.

Signed-off-by: Niu Yawei <niu@whamcloud.com>
Signed-off-by: Andreas Dilger <adilger@whamcloud.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agotune2fs: more respect for quota config option
Eric Sandeen [Mon, 14 Oct 2013 13:03:06 +0000 (09:03 -0400)]
tune2fs: more respect for quota config option

Commit 44a2cca3 disabled tune2fs -O quota when the build
didn't have --enable-quota at configure time, but that
wasn't quite enough.

We need to exclude the "-Q" option as well from tune2fs
when --enable-quota isn't specified at configure time.

Otherwise, tune2fs -Q can set the quota feature, but no other
utility will touch the filesystem due to the unknown flag,
if buitl w/o --enable-quota.

So put everything related to "-Q" under #ifdef CONFIG_QUOTA;
usage output (was missing before) and option parsing.

Addresses-Red-Hat-Bugzilla: #1010709
Reported-by: Bert DeKnuydt <Bert.Deknuydt@esat.kuleuven.be>
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agomke2fs: remove quota docs if not configured in
Eric Sandeen [Mon, 14 Oct 2013 12:54:15 +0000 (08:54 -0400)]
mke2fs: remove quota docs if not configured in

If quota isn't turned on with --enable-quota, then comment
quota documentation out of the mke2fs manpage.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agocontrib/populate-extfs.sh: use debugfs to populate extX fs
Robert Yang [Mon, 14 Oct 2013 02:09:12 +0000 (22:09 -0400)]
contrib/populate-extfs.sh: use debugfs to populate extX fs

This script uses debugfs command to populate the ext2/3/4 filesystem
from a given directory, it is a little similar to genext2fs, but this
one fully supports ext3 and ext4.

Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Acked-by: Darren Hart <dvhart@linux.intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
11 years agodebugfs.c: do sparse copy when src is a sparse file
Robert Yang [Mon, 14 Oct 2013 02:08:40 +0000 (22:08 -0400)]
debugfs.c: do sparse copy when src is a sparse file

Let debugfs do sparse copy when src is a sparse file, just like
"cp --sparse=auto"

* For the:
  #define IO_BUFSIZE 64*1024
  this is a suggested value from gnu coreutils:
  http://git.savannah.gnu.org/gitweb/?p=coreutils.git;a=blob;f=src/ioblksize.h;h=1ae93255e7d0ccf0855208c7ae5888209997bf16;hb=HEAD

* Use malloc() to allocate memory for the buffer since put 64K (or
  more) on the stack seems not a good idea.

Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Acked-by: Darren Hart <dvhart@linux.intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
11 years agodebugfs.c: the max length of debugfs argument is too short
Robert Yang [Mon, 14 Oct 2013 02:07:57 +0000 (22:07 -0400)]
debugfs.c: the max length of debugfs argument is too short

The max length of debugfs argument is 256 which is too short, the
arguments are two paths, the PATH_MAX is 4096 according to
/usr/include/linux/limits.h, so use BUFSIZ (which is 8192 on Linux
systems), that's also what the ss library uses.

Signed-off-by: Robert Yang <liezhi.yang@windriver.com>
Acked-by: Darren Hart <dvhart@linux.intel.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
11 years agotests: remove version number dependency in r_min_itable
Theodore Ts'o [Sun, 13 Oct 2013 02:25:29 +0000 (22:25 -0400)]
tests: remove version number dependency in r_min_itable

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agoe2fsprogs: add (optional) sparse checking to the build
Darrick J. Wong [Sat, 12 Oct 2013 03:02:12 +0000 (23:02 -0400)]
e2fsprogs: add (optional) sparse checking to the build

Run sparse against source files when building e2fsprogs with 'make C=1'.  If
instead C=2, it configures basic ext2 types for bitwise checking with sparse,
which can help find the (many many) spots where conversion errors are
(possibly) happening.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agoe2image: complain if running e2image -r or -Q on a mounted filesystem
Carlos Maiolino [Sat, 12 Oct 2013 01:38:53 +0000 (21:38 -0400)]
e2image: complain if running e2image -r or -Q on a mounted filesystem

Several users have used e2image on a mounted RW filesystem, resulting in
inconsistent, useless e2images for debugging purposes.

This commit will forbid this and print an error message, although the
user can override this using a new force option.

Signed-off-by: Carlos Maiolino <cmaiolino@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
11 years agolibext2fs: openfs() musn't allow bigalloc without EXT2_FLAGS_64BITS
Darrick J. Wong [Sat, 12 Oct 2013 01:20:54 +0000 (21:20 -0400)]
libext2fs: openfs() musn't allow bigalloc without EXT2_FLAGS_64BITS

Currently, only the new 64-bit bitmap implementation supports the
block<->cluster conversions that bigalloc requires.  Therefore, if we
have a bigalloc filesystem, require EXT2_FLAGS_64BITS be passed in to
ext2fs_open().  This does not mean that bigalloc file systems have to
be 64-bits; just that the userspace utilities have to be able to use
the new 64-bit capable library functions.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Lukas Czerner <lczerner@redhat.com>
11 years agoe2fsck: enable extents on all 64bit filesystems
Darrick J. Wong [Sat, 12 Oct 2013 01:20:36 +0000 (21:20 -0400)]
e2fsck: enable extents on all 64bit filesystems

Since it's impossible to address all blocks of a 64bit filesystem
without extents, have e2fsck turn on the feature if it finds (64bit &&
!extents).

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agomke2fs: complain about creating 64bit filesystems without extents
Darrick J. Wong [Sat, 12 Oct 2013 01:20:22 +0000 (21:20 -0400)]
mke2fs: complain about creating 64bit filesystems without extents

A 64bit filesystem without extents is not terribly useful, because the
old block map format does not support pointing to high block numbers.
Warn the user who tries to create such an animal.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agolibext2fs: search all possible blocks in implied_cluster_alloc()
Darrick J. Wong [Tue, 8 Oct 2013 16:08:46 +0000 (12:08 -0400)]
libext2fs: search all possible blocks in implied_cluster_alloc()

implied_cluster_alloc() is written such that if the the user passes in
a logical block that is the zeroth block in a logical cluster (lblk %
cluster_ratio == 0), then it will assume that there is no physical
cluster mapped to any other part of the logical cluster.

This is not true if we happen to be allocating logical blocks in
reverse order.  Therefore, search the whole cluster, except for the
lblk that we passed in.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agolibext2fs: fix off-by-one error in file truncation
Darrick J. Wong [Tue, 8 Oct 2013 15:51:23 +0000 (11:51 -0400)]
libext2fs: fix off-by-one error in file truncation

When told to truncate a file, ext2fs_file_set_size2() should start with
the first block past the end of the file.  The current calculation
jumps one more block ahead, with the result that it fails to hack off
the last block.  Adding blocksize-1 and dividing is sufficient to find
the last block.

Reviewed-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agolibext2fs: update i_size in ext2fs_file_write()
Niu Yawei [Tue, 8 Oct 2013 03:30:10 +0000 (11:30 +0800)]
libext2fs: update i_size in ext2fs_file_write()

ext2fs_file_write() needs to update i_size on successful write,
otherwise, ext2fs_file_read() in same open/close cycle will not
be able to read the just written data.

This fixes a bug which results in the the problem of quotacheck
triggered on 'tune2fs -O quota' failed to write back multiple
users/groups accounting information.

Signed-off-by: Niu Yawei <yawei.niu@intel.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agoe2fsprogs: fix inode and block relocation functions to use blk64_t
Darrick J. Wong [Mon, 7 Oct 2013 13:57:36 +0000 (09:57 -0400)]
e2fsprogs: fix inode and block relocation functions to use blk64_t

The inode and block relocation functions aren't currently compiled in
(so we don't need to worry about breaking ABI compatibility).  They
were originally intended for use by resize2fs, but we never ended up
using them, so (wisely) they weren't ever included in libext2fs as an
exported interface (they're not even compiled by the Makefile).

Fix them so that in case we ever use them, so that in places where raw
data types (int, long, etc.) stood in for blk_t and blk64_t.  Also fix
some sites where we should probably be using blk64_t.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agoe2fsprogs: fix blk_t <- blk64_t assignment mismatches
Darrick J. Wong [Mon, 7 Oct 2013 13:51:48 +0000 (09:51 -0400)]
e2fsprogs: fix blk_t <- blk64_t assignment mismatches

Fix all the places where we should be using a blk64_t instead of a
blk_t.  These fixes are more severe because 64bit values could be
truncated silently.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agolibext2fs: ind_punch() must not stop examining blocks prematurely
Darrick J. Wong [Mon, 7 Oct 2013 13:51:35 +0000 (09:51 -0400)]
libext2fs: ind_punch() must not stop examining blocks prematurely

When we're iterating the main loop in ind_punch(), "offset" tracks how
far we've progressed into the block map, "start" tells us where to
start punching, and "count" tells us how many blocks we are to punch
after "start".  Therefore, we would like to break out of the loop once
the "offset" that we're looking at has progressed past the end of the
punch range.  Unfortunately, if start !=0, the if-break clause in the
loop causes us to break out of the loop early.

Therefore, change the breakout test to terminate the loop at the
correct time.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agolibext2fs: allow callers to punch a single block
Darrick J. Wong [Mon, 7 Oct 2013 13:51:20 +0000 (09:51 -0400)]
libext2fs: allow callers to punch a single block

The range of blocks to punch is treated as an inclusive range on both
ends, i.e. if start=1 and end=2, both blocks 1 and 2 are punched out.
Thus, start == end means that the caller wishes to punch a single
block.  Remove the check that prevents us from punching a single
block.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agolibext2fs: rewind extent pointer when totally deleting an extent
Darrick J. Wong [Mon, 7 Oct 2013 13:35:22 +0000 (09:35 -0400)]
libext2fs: rewind extent pointer when totally deleting an extent

During a punch operation, if we decide to delete an extent out of the
extent tree, the subsequent extents are moved on top of the current
extent (that is to say, they're memmmove'd down one slot).  Therefore
it is not correct to advance to the next leaf because that means we
miss half the extents in the range!  Rereading the current pointer
should be fine.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>