/usr/share/doc/dosfstools
Edit: /usr/share/doc/dosfstools/ChangeLog (106302B)
commit cad2c17
Author: Andreas Bombe
Date: Mon Jan 23 03:10:56 2017 +0100
fatlabel: Make sure printf style patterns in label are not expanded
Writing a label to an unlabeled filesystem involves creating a root
directory entry to contain the label. For this, the
alloc_rootdir_entry() function used for creating new directory nodes for
recovered orphan cluster chains is reused. However, this expects the
filename to be a printf style pattern to generate a unique name.
If the filesystem is unlabeled and the new label contains a printf
specifier other than %d or similar, the attempted expansion of the
pattern might crash fatlabel.
Change alloc_rootdir_entry() to take an additional argument gen_name and
only expand the name if its value is true.
Signed-off-by: Andreas Bombe
commit 77371e2
Author: Andreas Bombe
Date: Fri Nov 4 13:10:58 2016 +0100
testsuite: Add missing fsck test files to dist_check_DATA
Signed-off-by: Andreas Bombe
commit d6df1d4
Author: Andreas Bombe
Date: Thu Nov 3 20:03:25 2016 +0100
testsuite: Test for bad characters in file names
The FAT16 test image contains four files with three errors. One has a
space in the name, one a space in the extension and one contains '>' in
the name.
Signed-off-by: Andreas Bombe
commit 307966c
Author: Andreas Bombe
Date: Sun Sep 18 00:11:43 2016 +0200
fsck: Fix reporting of length to truncate with shared clusters
When files share clusters and one of the two files has been selected to
truncate fsck reports which file has been selected and the length to
which it will be truncated. All code paths erronously printed the length
of the file that was not going to be truncated, leading to possible user
confusion.
Signed-off-by: Andreas Bombe
commit b33424e
Author: Andreas Bombe
Date: Sat Sep 17 23:34:07 2016 +0200
testsuite: Test for files sharing clusters
The FAT32 test image contains three files. One shares a cluster with the
root directory and the other two share clusters among themselves.
Signed-off-by: Andreas Bombe
commit db5ec04
Author: Andreas Bombe
Date: Sat Sep 17 22:41:07 2016 +0200
testsuite: Run fsck in test-mkfs
Now the test-mkfs script also runs fsck on the created image to test its
basic operation. The test fails if fsck reports seeing an error.
Signed-off-by: Andreas Bombe
commit 4ac17ae
Author: Andreas Bombe
Date: Sat Sep 17 22:25:06 2016 +0200
Fix FAT32 FSInfo signature checking
In commit 7444694 the definition of struct info_sector was corrected to
match the FSInfo sector specification. The initialization was adapted,
but other places were missed. All runs of fsck would then incorrectly
report a signature error on the FSInfo sector.
Signed-off-by: Andreas Bombe
commit e883d07
Author: Nico Roeser
Date: Fri Sep 16 20:05:08 2016 +0200
mkfs: Bring back ability to set sector size.
This had accidently been broken when the code was overhauled, resulting
in unconditionally setting the sector size when devinfo detected it.
Signed-off-by: Andreas Bombe
commit 4ef83db
Author: Andreas Bombe
Date: Fri Sep 16 03:32:12 2016 +0200
Request minimum automake version 1.11 and add workaround for <1.13
Before automake 1.13, parallel-tests wasn't the default set it
explicitly in the automake options list. The parallel test harness was
introduced in 1.11 therefore that is the minimum now.
However, the AM_TESTS_ENVIRONMENT variable was only starting to work in
1.13, so there is now a version check and a conditional assignment to
TESTS_ENVIRONMENT (which should be for user setting only) in case
automake is 1.11 or 1.12.
Without it, the fact of xxd being found is not passed to the test
scripts which then skip all tests due to perceived lack of xxd.
Signed-off-by: Andreas Bombe
commit 6a8c8b7
Author: Andreas Bombe
Date: Fri Sep 16 01:45:17 2016 +0200
Add missing autoreconf step to Travis CI configuration
Signed-off-by: Andreas Bombe
commit 54f7aef
Author: Andreas Bombe
Date: Fri Sep 16 01:39:08 2016 +0200
Add configuration for the Travis CI build service
Signed-off-by: Andreas Bombe
commit 5421752
Author: Andreas Bombe
Date: Wed Sep 14 20:37:57 2016 +0200
Check and fix printf style arguments to die() and pdie()
Add the "format" gcc style attribute to the declarations of die() and
pdie() so that the compiler knows how to check their arguments for
correctness.
Fix all problematic calls detected by this.
Signed-off-by: Andreas Bombe
commit 83e5ec5
Author: Andreas Bombe
Date: Wed Sep 14 03:38:08 2016 +0200
Make some implicit integer conversions safer
Coverity reports a few places where signed or short integers are
promoted to int and then extended to long which might cause unintended
sign extension. Given the context (being sector sizes and similar) the
values should never actually reach unsafe values, but this patch cleans
it up anyway.
Signed-off-by: Andreas Bombe
commit ff5e718
Author: Andreas Bombe
Date: Tue Sep 13 20:25:34 2016 +0200
mkfs: Remove redundant unreachable warning message
At some point the check for too many clusters for user requested FAT16
was moved a few lines above. The original location can only be reached
when the cluster count is too low for FAT16 which means the too many
clusters message can not be triggered there.
Found through the dead code detection of Coverity Scan.
Signed-off-by: Andreas Bombe
commit e359a33
Author: Andreas Bombe
Date: Tue Sep 13 20:12:21 2016 +0200
Fix octal character parsing in file_cvt()
Apparently the octal code parsing got rewritten into a loop at some
point while the rest of the code was not removed. It was impossible to
specify an octal character code without the function returning an error.
This affected the -u and -d options for fsck.
Found through the dead code detection of Coverity Scan.
Signed-off-by: Andreas Bombe
commit ccd0a9d
Author: Andreas Bombe
Date: Mon Sep 12 20:21:53 2016 +0200
testsuite: Test for circular cluster chain
The test image contains a file four clusters in length where the third
cluster has been modified to point back to the second cluster.
Signed-off-by: Andreas Bombe
commit fab40f8
Author: Andreas Bombe
Date: Mon Sep 12 19:51:10 2016 +0200
Fix truncation logic in circular chain check
When scanning for circular chains, the first FAT entry that appeared a
second time in a chain should cause the previous cluster to be the
termination of the chain.
To that end, test_file() keeps variables to point to the previous
cluster and the current number of clusters. When the -t option is given,
all clusters are checked to be readable and skipped (variables not
updates) in test_file() when they are not.
The problem was that the variables were updated only when the read check
succeeded. If the -t option is not given, that check does not happen and
the variables keep their initial value. The circular chain fix would
then always truncate to zero length and fail to update the free cluster
count, requiring a second fsck run.
Fix by updating these variables unconditionally when -t option is not
given.
Signed-off-by: Andreas Bombe
commit 8394ceb
Author: Andreas Bombe
Date: Mon Sep 12 14:39:11 2016 +0200
testsuite: Test for too long cluster chain
The test image contains a file marked as 7 bytes in length (just a small
part of 1 cluster) in the directory entry that has 2 clusters assigned
to it in the FAT.
Signed-off-by: Andreas Bombe
commit 7444694
Author: Andreas Bombe
Date: Wed Sep 7 21:09:41 2016 +0200
Fix definition and initialization of fsinfo sector
Change the definition of the fsinfo struct to be more in line with the
official definition, in particular make the trailing signature field 32
bit instead of 16 bit. During initialization in init_fsinfo(), first
clear the whole struct to zero before setting fields, to make sure
reserved fields are zeroed.
Fixes #33.
Signed-off-by: Andreas Bombe
commit f78cb68
Author: Andreas Bombe
Date: Wed Sep 7 20:45:29 2016 +0200
Add fsck testsuite
Add the framework for a fsck testsuite and a few tests, including a test
for the yet unimplemented check for misordered dot/dotdot directory
entries.
The test driver runs fsck twice on a given image and fails the test if
the first run does not detect an error or the second run still detects
an error.
Signed-off-by: Andreas Bombe
commit 7667295
Author: Andreas Bombe
Date: Tue May 31 03:32:42 2016 +0200
Add testsuite foundations
Add testsuite on the basis of the automake framework. As a start, three
tests for basic mkfs functionality compare mkfs generated images against
references images.
Signed-off-by: Andreas Bombe
commit f881195
Author: Andreas Bombe
Date: Wed Sep 7 03:48:16 2016 +0200
Append +git to version number in configure.ac
Don't let dosfstools compiled from git with additional commits show the
exact release version number.
Signed-off-by: Andreas Bombe
commit 8788615
Author: Álvaro Fernández Rojas
Date: Thu Jun 30 08:43:50 2016 +0200
Add config.guess and config.sub to gitignore
Signed-off-by: Álvaro Fernández Rojas
Signed-off-by: Andreas Bombe
commit 39ab842
Author: Álvaro Fernández Rojas
Date: Thu Jun 30 08:41:51 2016 +0200
Use autoconf to look for OS X endian include.
Signed-off-by: Álvaro Fernández Rojas
Signed-off-by: Andreas Bombe
commit 69ac2b0
Author: Alessio Sergi
Date: Fri May 6 18:18:18 2016 +0200
src/device_info.c: Fix undefined PATH_MAX under musl
In musl libc, PATH_MAX is defined in .
Signed-off-by: Andreas Bombe
commit 744db84
Author: Lars Wendler
Date: Tue May 10 09:47:57 2016 +0200
Don't link to udev via LDFLAGS. There's LDADD for such a task.
See also:
https://www.gnu.org/software/automake/manual/html_node/Linking.html
Signed-off-by: Lars Wendler
Signed-off-by: Andreas Bombe
commit 1e76e57
Author: Andreas Bombe
Date: Wed May 11 03:44:58 2016 +0200
mkfs: Default to 64/32 heads/sectors for targets smaller than 512 MB
This may put defaults in certain use cases a little bit more in line
with the old defaults in versions up to 3.0.28. It has mostly aesthetic
value in most cases.
Signed-off-by: Andreas Bombe
commit 786e66e (tag: v4.0)
Author: Andreas Bombe
Date: Fri May 6 03:19:07 2016 +0200
Releasing version 4.0.
Signed-off-by: Andreas Bombe
commit a79ff90
Author: Andreas Bombe
Date: Fri May 6 02:34:42 2016 +0200
src/Makefile.am: Fix CPPFLAGS for VPATH builds
The -I flag to add the blkdev subdir to the include search path is now
relative to $(srcdir) to allow VPATH builds to work. Additionally move
the -I flag from the mkfs_fat and testdevinfo CFLAGS to CPPFLAGS where
it actually belongs.
Signed-off-by: Andreas Bombe
commit e8eff14
Author: Andreas Bombe
Date: Wed Apr 27 21:38:37 2016 +0200
read_boot(): Handle excessive FAT size specifications
The variable used for storing the FAT size (in bytes) was an unsigned
int. Since the size in sectors read from the BPB was not sufficiently
checked, this could end up being zero after multiplying it with the
sector size while some offsets still stayed excessive. Ultimately it
would cause segfaults when accessing FAT entries for which no memory
was allocated.
Make it more robust by changing the types used to store FAT size to
off_t and abort if there is no room for data clusters. Additionally
check that FAT size is not specified as zero.
Fixes #25 and fixes #26.
Reported-by: Hanno Böck
Signed-off-by: Andreas Bombe
commit 016800e
Author: Andreas Bombe
Date: Wed Apr 27 14:16:53 2016 +0200
Use variable total_fat_entries in read_boot() for readability
Signed-off-by: Andreas Bombe
commit a6478d8
Author: Álvaro Fernández Rojas
Date: Fri Apr 8 12:20:46 2016 +0200
Add missing iconv library for OS X
Signed-off-by: Álvaro Fernández Rojas
Signed-off-by: Andreas Bombe
commit ce67dc6
Author: Álvaro Fernández Rojas
Date: Fri Apr 8 12:20:27 2016 +0200
Add endian support for OS X
Signed-off-by: Álvaro Fernández Rojas
Signed-off-by: Andreas Bombe
commit 08f3869
Author: Joel Holdsworth
Date: Thu Mar 10 00:53:07 2016 +0000
Configure option to disable building with libudev
Signed-off-by: Andreas Bombe
commit ed4e47b
Author: Andreas Bombe
Date: Mon Feb 22 03:47:14 2016 +0100
Remove use of PATH_MAX in path_name()
The length of a file path on the checked filesystem has no relation to
the maximum path length of the system fsck is running on. So replace it
with a constant of our own.
As a bonus this will not fail compilation on a system without PATH_MAX.
Signed-off-by: Andreas Bombe
commit b1a38ab
Author: Andreas Bombe
Date: Wed Feb 17 21:04:35 2016 +0100
Add preliminary entry for release 4.0 to NEWS
Signed-off-by: Andreas Bombe
commit 4ad3e9e
Author: Andreas Bombe
Date: Wed Feb 17 21:02:06 2016 +0100
Adjust ridiculous source indentation in io.c
Signed-off-by: Andreas Bombe
commit b96acb2
Author: Andreas Bombe
Date: Wed Feb 17 20:51:53 2016 +0100
Document ./configure --enable-compat-symlinks in README
Signed-off-by: Andreas Bombe
commit de39c5c
Author: Andreas Bombe
Date: Wed Feb 17 15:16:27 2016 +0100
Add include paths.h in the HAVE_DECL_GETMNTENT case
_PATH_MOUNTED is now used for getmntent() in place of MOUNTED because
the latter was marked as a deprecated alias in glibc's mntent.h. The
mntent.h of musl libc does not include the _PATH_MOUNTED however. Fix
this by including paths.h alongside mntent.h
Signed-off-by: Andreas Bombe
commit 86c7acd
Author: Andreas Bombe
Date: Wed Feb 17 15:06:56 2016 +0100
man fsck: Document the -c option
Signed-off-by: Andreas Bombe
commit ef9a73c
Author: Andreas Bombe
Date: Mon Feb 15 02:10:57 2016 +0100
Add NEWS file with changes of the last two releases
Signed-off-by: Andreas Bombe
commit bda6551
Author: Andreas Bombe
Date: Fri Feb 12 03:56:16 2016 +0100
Make filesystem mounted check portable
A new function is_device_mounted() in device_info.c is now used by
check_mount() in mkfs.fat.c. It contains the getmntent() using code
used before in check_mount() and now an alternative using getmntinfo()
as found on the BSDs.
In case neither function is available, is_device_mounted() defaults to
reporting that the device isn't mounted.
Signed-off-by: Andreas Bombe
commit 4b8c9cc
Author: Andreas Bombe
Date: Fri Feb 12 01:57:33 2016 +0100
Make use of endian.h portable to BSD
The endian.h found on Linux and the BSDs appear to be compatible, but
they are found in different locations. Add tests in configure.ac and a
new endian_compat.h file that has the logic to include the correct
files.
Signed-off-by: Andreas Bombe
commit d7665f2
Author: Andreas Bombe
Date: Wed Feb 10 21:30:06 2016 +0100
Fix format string in check_file() (%lu → %llu)
The cluster chain length printing needs a 64 bit calculation, so we can
just use unsigned long long instead of uint64_t and use the format
string %llu.
Signed-off-by: Andreas Bombe
commit 6225e59
Author: Andreas Bombe
Date: Fri Feb 5 14:39:00 2016 +0100
blkdev.c: Prevent unused parameter warnings in fallback code
Signed-off-by: Andreas Bombe
commit 34cdded
Author: Andreas Bombe
Date: Fri Feb 5 14:36:14 2016 +0100
blkdev_get_size(): Remove unused variable ch
Signed-off-by: Andreas Bombe
commit 5571d29
Author: Andreas Bombe
Date: Wed Feb 10 03:22:19 2016 +0100
Reinstate alignment of FAT32 structures to cluster size
This reverts commits 17c956cb9 and d63e0d627 where the alignment was
removed because it created problems with a device that refused to read
the aligned filesystem. The option -a is already provided to disable
alignment in order to handle such cases.
This change brings it back in line with FAT12/16 where alignment wasn't
disabled and brings consistency with the current command line options,
where only the option to disable alignment exists but no opposite option
to enable it.
Signed-off-by: Andreas Bombe
commit 2c71ace
Author: Andreas Bombe
Date: Wed Feb 3 03:38:33 2016 +0100
Makefile.am: Add historic documentation to distribution
Signed-off-by: Andreas Bombe
commit bcbae63
Author: Andreas Bombe
Date: Wed Feb 3 02:38:24 2016 +0100
src/Makefile.am: Add forgotten msdos_fs.h to mkfs_fat_SOURCES
Signed-off-by: Andreas Bombe
commit 19d1a13
Author: Andreas Bombe
Date: Wed Feb 3 02:34:51 2016 +0100
Remove sys/ioctl.h and linux/fd.h include from io.c
These weren't used anymore and the linux/fd.h include would
gratuitously cause compilation to fail on non-Linux environments.
Signed-off-by: Andreas Bombe
commit f691660
Author: Andreas Bombe
Date: Wed Feb 3 02:31:00 2016 +0100
Reinstate some #include
These shouldn't have been removed in commit 245d0cce5. Put them back for
correctness even though the definitions were pulled in implicitly.
Signed-off-by: Andreas Bombe
commit 51afd41
Author: Andreas Bombe
Date: Wed Jan 27 21:38:29 2016 +0100
src/Makefile.am: Put all header files in appropriate _SOURCES variables
With the headers missing the dist targets of the automake generated
Makefiles would not include them and make the resulting dist
unbuildable.
Also combine sources collections into common variables for
deduplication.
Signed-off-by: Andreas Bombe
commit bf6f142
Author: Andreas Bombe
Date: Wed Jan 27 15:28:42 2016 +0100
mkfs man: Note that sector sizes > 4096 are non-standard
Signed-off-by: Andreas Bombe
commit ea96c32
Author: Andreas Bombe
Date: Mon Jan 25 21:30:23 2016 +0100
mkfs: Improve parsing of bad blocks file
The bad blocks file that can be given to mkfs via the -l option had a
very simplistic design. It failed to notice it was parsing an empty
line and would report errors for that.
Replace it with a more robust version that ignores empty lines as well
as leading and trailing white space. Additionally it produces meaningful
error messages.
GitHub: Fixes #17
Signed-off-by: Andreas Bombe
commit dfb5bea
Author: Andreas Bombe
Date: Mon Jan 25 02:47:12 2016 +0100
mkfs: Limit filesystem size on targets that are too large
For FAT filesystems, the number of sectors has to fit into a 32 bit
variable. Previously this was not checked possibly causing invalid
filesystems to be generated.
Now there is a check for that case which will limit the number of
sectors if needed and print a warning that disk space will be left
unused in that case.
Signed-off-by: Andreas Bombe
commit fc0343f
Author: Andreas Bombe
Date: Fri Jan 22 21:29:20 2016 +0100
mkfs: Improved bounds checking in mark_FAT_sector()/mark_FAT_cluster()
In mark_FAT_sector() the sector number itself is now checked against
limits instead of the computed cluster number. Even with sector number
before the start of the data area, the cluster number may be valid for
the first cluster due to dividing by the cluster size.
Both functions now check for upper limits and should prevent writing
past the valid end of the FAT.
Signed-off-by: Andreas Bombe
commit 9211c8a
Author: Andreas Bombe
Date: Fri Jan 15 02:17:16 2016 +0100
mkfs: Fix offset error in FAT12/16 bad cluster marking
The root directory wasn't factored in to the calculation of the data
area start sector. On FAT32 the root directory is in the data area, but
for FAT12 and FAT16 it is a reserved space before the start of the data
area.
On FAT12 and FAT16, this resulted in the wrong clusters being marked
during bad blocks mapping, whether from check_blocks() or from reading
the user supplied bad blocks file.
Signed-off-by: Andreas Bombe
commit 0627a62
Author: Andreas Bombe
Date: Fri Jan 15 01:59:59 2016 +0100
mkfs: Fix off-by-2 error in bad cluster marking
mark_FAT_sector(), which has the mark_sector_bad() macro as its sole
user, computed the cluster number corresponding to the sector by taking
its offset from the first data sector and dividing by sectors per
cluster.
What it missed was that the first data cluster is number 2 and not 0.
This meant all marks were off by 2 and when the first two clusters are
supposed to be marked, it would overwrite the reserved cluster values
and create an invalid filesystem.
Signed-off-by: Andreas Bombe
commit 2dca9aa
Author: Andreas Bombe
Date: Thu Jan 14 14:43:00 2016 +0100
.gitignore: Add .dirstamp
Signed-off-by: Andreas Bombe
commit 6a966fb
Author: Andreas Bombe
Date: Thu Jan 14 14:38:53 2016 +0100
mkfs: Reword non-standard sector size warning
Signed-off-by: Andreas Bombe
commit bebc9ac
Author: Andreas Bombe
Date: Wed Dec 30 15:10:35 2015 +0100
Clean up includes in mkfs.fat.c
Moving the device probing out into device_info.c removed the need for a
number of includes in mkfs.fat.c. Remove them and add a define for
BLOCK_SIZE, which was the only thing used from linux/fs.h.
Signed-off-by: Andreas Bombe
commit 12a1d46
Author: Andreas Bombe
Date: Sun Nov 29 01:59:10 2015 +0100
Don't use pointer to first member when more of the struct gets copied
Where a fs_write() of "first 13 bytes of directory entry" is intended,
actually use pointer to directory entry structure instead of the 11
byte name field at the beginning.
This does not change how the code works, it is just a clean up.
Signed-off-by: Andreas Bombe
commit d38bd2d
Author: Andreas Bombe
Date: Sun Nov 29 00:44:48 2015 +0100
Remove name/extension split in directory entry structures
Both the DIR_ENT structure in fsck.fat.h and the msdos_dir_entry in
msdos_fs.h - these represent the on disk format of directory entries -
had the name field split into name[8] followed by ext[3].
By far the most operations on name are on the full name including
extension and they treated the name field as an 11 byte array. This is
an array overflow that worked because the structs have the attribute
packed and the extension field is following right after.
Nevertheless, this is not clean C and the merging of both fields
actually simplified the code in a few places.
Signed-off-by: Andreas Bombe
commit 0847e4c
Author: Andreas Bombe
Date: Fri Nov 27 21:29:49 2015 +0100
Free allocated strings after use
There are multiple calls to cnv_unicode() in lfn.c which returns an
allocated string. Most had the appropriate free() calls after printing
the strings. Add the missing two calls where memory was leaked.
Found by Coverity.
Signed-off-by: Andreas Bombe
commit 1b7d91e
Author: Andreas Bombe
Date: Fri Nov 27 03:34:51 2015 +0100
Add test for and include linux/hdreg.h in blkdev.c
Before, blkdev did not include it and depended on its own fallback
definition of the ioctl and struct hd_geometry.
Signed-off-by: Andreas Bombe
commit 5024372
Author: Andreas Bombe
Date: Fri Nov 27 03:28:14 2015 +0100
Add FDGETPRM attempt to blkdev_get_geometry()
If HDIO_GETGEO isn't available or has failed, try FDGETPRM. This should
get the geometry from floppy drivers where HDIO_GETGEO isn't supported.
Signed-off-by: Andreas Bombe
commit 4a146d7
Author: Andreas Bombe
Date: Fri Nov 27 03:25:55 2015 +0100
Add blkdev_get_start() for getting partition start offset
Signed-off-by: Andreas Bombe
commit 254f8ab
Author: Andreas Bombe
Date: Thu Nov 26 23:33:02 2015 +0100
Remove use of libblkid again
It appears libblkid is not as widely available as presumed, since some
platforms only have the original libblkid included in e2fsprogs which
lacks the needed functionality. This commit removes the requirement and
use of libblkid.
As a replacement, blkdev.c from util-linux is included, which offers the
required basic functionality in a portable way.
Signed-off-by: Andreas Bombe
commit c9fb33c
Author: Andreas Bombe
Date: Wed Oct 21 21:32:30 2015 +0200
Use just device size not major number in Atari mode
In Atari mode, read_boot() in boot.c used the device major number to
determine whether to use FAT12. It would always use FAT12 for a floppy,
otherwise only if it is a RAM disk or loopback device and has a size
corresponding to standard floppy formats.
Since this check was already broken for a long time (another place that
assumed 8 bit major numbers) and there is no real point to make the
distinction based on device, this commit reduces the check to just
compare against standard floppy sizes.
Signed-off-by: Andreas Bombe
commit 64486ad
Author: Andreas Bombe
Date: Wed Oct 21 00:18:22 2015 +0200
Remove loff_t and llseek()
There appear to have been multiple conversions to 64 bit file offsets on
32 bit architectures in dosfstools over the years, but today with the
proper setup off_t is 64 bits and simple lseek() can be used. The
AC_SYS_LARGEFILE macro in configure.ac does what is required to make
that happen.
Given this, convert all uses of loff_t to off_t, remove llseek()
definitions and change llseek() calls to plain lseek().
Signed-off-by: Andreas Bombe
commit e03a5f4
Author: Andreas Bombe
Date: Fri Oct 16 21:47:04 2015 +0200
Remove DJGPP support
These macros have been added back when dosfstools was around version 2.
It is difficult to say whether these are still working correctly or
whether they are in use at all. FreeDOS appears to still show version
2.11 of dosfstools in their software directory.
Supporting actual MS-DOS or compatible may need more work in the
current state of things and this DJGPP support can be removed until
then.
Signed-off-by: Andreas Bombe
commit 5b9a88d
Author: Andreas Bombe
Date: Thu Oct 8 16:17:22 2015 +0200
mkfs.fat: Complete overhaul of device probing
The device probing in mkfs.fat is used to get device parameters where
needed and also to decide whether to refuse overwriting a device due to
possible user error. This code has suffered severe bitrot and is highly
Linux specific. Highlights include using hardcoded major/minor device
numbers to classify a device, and using 8 bits major/minor numbers that
have become obsolete a long time ago and thus often misidentifying a
device.
The overhauled implementation is now in src/device_info.c and makes use
of libudev (optional, recommended) and libblkid (required) to probe the
device and where Linux ioctls are required it provides fallbacks and
does not attempt to call these on non-Linux systems. The FAT parameter
selection has been unified and simplified in the process.
A new executable testdevinfo has been added that gets built but not
automatically installed. It takes one file name, uses the same probing
as mkfs.fat would with verbose messages enabled and displays the
results.
Signed-off-by: Andreas Bombe
commit 53eddfc
Author: Andreas Bombe
Date: Sat Sep 12 02:54:33 2015 +0200
Die on out of range cluster values in set_fat()/get_fat()
To prevent bugs caused by FAT corruption inside fsck to go unnoticed,
add a check against out of range requested cluster values in get_fat()
and against out of range cluster to change and new cluster value in
set_fat().
When an invalid cluster value is detected, these functions now die()
with an "internal error" message.
Signed-off-by: Andreas Bombe
commit 456767b
Author: Andreas Bombe
Date: Mon Sep 14 00:33:49 2015 +0200
configure.ac: Use AS_HELP_STRING to format option help
Signed-off-by: Andreas Bombe
commit 3cfb479
Author: Andreas Bombe
Date: Mon Sep 14 00:30:40 2015 +0200
version.h: Use @configure_input@ autoconf variable in boilerplate
Signed-off-by: Andreas Bombe
commit 41ef834
Author: Andreas Bombe
Date: Mon Sep 14 00:29:27 2015 +0200
.gitignore: Add TAGS in addition to tags
Signed-off-by: Andreas Bombe
commit bdc3d2a
Author: Andreas Bombe
Date: Fri Sep 11 20:28:12 2015 +0200
Rename clusters field in DOS_FS struct
Rename it to data_clusters to prevent mistaking the clusters field of
the DOS_FS struct as the total number of FAT entries instead of the
number of data clusters (two less than the number of entries).
Signed-off-by: Andreas Bombe
commit 0790812
Author: Andreas Bombe
Date: Fri Sep 11 19:47:29 2015 +0200
set_fat(): Fix off-by-2 error leading to corruption in FAT12
In FAT12 two 12 bit entries are combined to a 24 bit value (three
bytes). Therefore, when an even numbered FAT entry is set in FAT12, it
must be be combined with the following entry. To prevent accessing
beyond the end of the FAT array, it must be checked that the cluster is
not the last one.
Previously, the check tested that the requested cluster was equal to
fs->clusters - 1. However, fs->clusters is the number of data clusters
not including the two reserved FAT entries at the start so the test
triggered two clusters early.
If the third to last entry was written on a FAT12 filesystem with an
odd number of clusters, the second to last entry would be corrupted.
This corruption may also lead to invalid memory accesses when the
corrupted entry becomes out of bounds and is used later.
Change the test to fs->clusters + 1 to fix.
Reported-by: Hanno Böck
Signed-off-by: Andreas Bombe
commit 39ce90f
Author: Andreas Bombe
Date: Fri Sep 11 19:34:10 2015 +0200
set_fat(): Move FAT12 next cluster check up
In FAT12 two 12 bit entries are combined to a 24 bit value (three
bytes). Therefore, when an even numbered FAT entry is set in FAT12, it
must be be combined with the following entry. To prevent accessing
beyond the end of the FAT array, it must be checked that the cluster is
not the last one.
This check was broken in ff1b24e9 (first included in 3.0.3) as the
lookup was done unconditionally and the check influenced only using the
looked up value.
Move the check up to fix.
Signed-off-by: Andreas Bombe
commit 2aad1c8
Author: Andreas Bombe
Date: Tue Sep 8 03:58:29 2015 +0200
Prevent out of bound array read in date_dos2unix()
The function date_dos2unix() is called during fsck while showing
information about duplicate file names. In case the date field of a
directory entry contains the invalid value 0 for the month,
date_dos2unix would read index -1 of the day_n array.
Add a check to prevent that and also make the day_n array const on this
occasion.
Reported-by: Hanno Böck
Signed-off-by: Andreas Bombe
commit 3b95786
Author: Yann E. MORIN
Date: Sun Aug 16 15:55:43 2015 +0200
mkfs.fat: fix incorrect int type
u_int32_t is not a stanard type, while uint32_t is. This fixes builds
with the musl C library, which only defines so-called "clean" headers;
build failures are like (back-quotes and elision manually added for
readability):
http://autobuild.buildroot.org/results/a09/a0923d7f6d4dbae02eba4c5024bbdae3a52aa85a/build-end.log
/home/peko/autobuild/instance-1/output/host/usr/bin/x86_64-linux-gcc -D_LARGEFILE_SOURCE \
-D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -Os -D_GNU_SOURCE -D_LARGEFILE_SOURCE \
-D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -c -o mkfs.fat.o src/mkfs.fat.c
src/mkfs.fat.c: In function 'main':
src/mkfs.fat.c:1415:18: error: 'u_int32_t' undeclared (first use in this function)
volume_id = (u_int32_t) ((create_timeval.tv_sec << 20) | create_timeval.tv_usec); [...]
^
src/mkfs.fat.c:1415:18: note: each undeclared identifier is reported only once for each
function it appears in
Signed-off-by: "Yann E. MORIN"
Signed-off-by: Andreas Bombe
commit 2b1c4d1
Author: Andreas Bombe
Date: Wed Jun 3 03:33:10 2015 +0200
Add README.md, remove Markdown formatting from README
Signed-off-by: Andreas Bombe
commit b720acc
Author: Andreas Bombe
Date: Wed Jun 3 03:27:24 2015 +0200
Add simple README in Markdown format
Signed-off-by: Andreas Bombe
commit 2b255e6
Author: Andreas Bombe
Date: Tue Jun 2 18:25:06 2015 +0200
Configure option for legacy names symlinks
The symlinks from the old names (mkdosfs, dosfsck, etc.) are now only
created on "make install" when the --enable-compat-symlinks option was
given to configure.
Signed-off-by: Andreas Bombe
commit 0643db7
Author: Andreas Bombe
Date: Sun May 31 02:40:19 2015 +0200
Convert build system to autoconf/automake
In preparation for fixing the horribly outdated and broken device
checking - which will likely involve using additional libraries like
libblkid - as well as making this package portable to other operating
systems, the build system is now the tried and true autoconf/automake
combination which should make both goals a little more straightforward.
The release version number and date are now in configure.ac and
substituted by configure where they are needed. Now it is no longer
necessary to change the number in multiple places for a release and the
man pages get the number substituted directly into them, making the
update-version.sh script and the VERSION file obsolete.
The English man pages are moved back up one directory to mark their
status as the master copy for all translations. At the moment the po4a
translation infrastructure is defunct since it isn't integrated into
the automake environment yet. So far it hasn't been used, so that is
not an actual regression.
The date in the man pages is not automatically updated anymore. This is
as it should be, since the date is supposed to signify the time of the
last nontrivial change and not the release date of the software.
Signed-off-by: Andreas Bombe
commit 85022fe (tag: v3.0.28)
Author: Andreas Bombe
Date: Sat May 16 02:56:17 2015 +0200
Releasing version 3.0.28.
Signed-off-by: Andreas Bombe
commit ad1342e
Author: Andreas Bombe
Date: Sat May 16 02:10:18 2015 +0200
manpages: Mark MT and ME tags as untranslated for po4a
Signed-off-by: Andreas Bombe
commit 3ed9ec2
Author: Andreas Bombe
Date: Sun May 10 19:59:02 2015 +0200
mkfs: Small changes to FAT32 cluster size selection
Put the cluster size selection back in line with the table used in
Microsoft's fatgen103.pdf and fix the comment. This only involved
changing some comparison operators, all values stayed the same.
Signed-off-by: Andreas Bombe
commit 9984552
Author: Andreas Bombe
Date: Mon Apr 20 23:41:34 2015 +0200
fsck: Mention -r is default in usage message
Signed-off-by: Andreas Bombe
commit 3eaca68
Author: Andreas Bombe
Date: Mon Apr 20 23:30:56 2015 +0200
manpages: Remove obsolete information about Linux FAT support
Signed-off-by: Andreas Bombe
commit a5e34de
Author: Andreas Bombe
Date: Thu Apr 16 23:16:08 2015 +0200
manpages: Convert the rest of argument placeholders to upper case
Signed-off-by: Andreas Bombe
commit 2303765
Author: Andreas Bombe
Date: Thu Apr 16 22:51:20 2015 +0200
manpages: Fix formatting
Make the manpages conform to the rules for groff sources. First,
eliminate empty lines because these create vertical spaces where it
isn't desired. Man page sources should not contain empty lines. Second,
put a line break between sentences in a paragraph. A period is
recognized and formatted as a full stop period only when a line break
comes right after it.
Signed-off-by: Andreas Bombe
commit d3969b8
Author: Andreas Bombe
Date: Fri Apr 10 20:32:50 2015 +0200
fsck: Verify first cluster of a file is not 1
Previously the FAT was checked for any out of bounds entries which were
cleared. However the first cluster of a file, as specified in its
directory entry, was not verified to not be 1.
In addition to missing this filesystem error, code that correctly
assumes the FAT contained no bad entries anymore could still look up
invalid table indices depending on the value stored in entry 1. With
the right values and FAT size this can lead to a segfault by accessing
unallocated memory.
Now test_file() will ignore files where the first cluster equals 1 and
an additional check in check_file() will truncate them.
This bug was reported in http://bugs.debian.org/773885 by Jakub Wilk.
Signed-off-by: Andreas Bombe
commit 2a56575
Author: Andreas Bombe
Date: Thu Apr 2 23:35:00 2015 +0200
Simplify synopses of man pages
Since there is only one mode of invocation for the tools, it is
somewhat pointless to list every possible option in the synopsis.
Fix a few style inconsistencies in addition, specifically that
placeholders for option arguments are in capital letters and should be
shown where the option is described.
Signed-off-by: Andreas Bombe
commit 40f1436
Author: Andreas Bombe
Date: Wed Apr 1 20:55:58 2015 +0200
Update homepage and maintainer sections of man pages
Signed-off-by: Andreas Bombe
commit 0169db8
Author: Andreas Bombe
Date: Wed Apr 1 20:05:01 2015 +0200
mkfs: Show a clearer message that file already exists with -C
Signed-off-by: Andreas Bombe
commit 2967e27
Author: Andreas Bombe
Date: Mon Mar 30 21:12:51 2015 +0200
mkfs: Remove O_TRUNC from file creation call
open() is called with O_EXCL | O_CREAT, so it will definitely create a
new, empty file or error out. O_TRUNC is superfluous and may be harmless
in practice but its effect under these circumstances is unspecified.
Signed-off-by: Andreas Bombe
commit 90612df
Author: Andreas Bombe
Date: Mon Mar 30 21:09:51 2015 +0200
mkfs: Use ftruncate() for expanding created file
Signed-off-by: Andreas Bombe
commit 58e8e09
Author: Andreas Bombe
Date: Tue Mar 24 21:32:40 2015 +0100
mkfs: Add --invariant option
The --invariant option is designed to prevent random or time based
differences to end up in generated filesystems so that multiple runs
generate on the same file or device create completely identical results.
This is intended for debugging or regression testing of mkfs.fat.
Signed-off-by: Andreas Bombe
commit 354feaf
Author: Andreas Bombe
Date: Thu Mar 19 21:35:21 2015 +0100
mkfs: Add long option parsing
Add long option parsing with getopt_long() to mkfs and define --help as
the first long option. The usage() function now takes an exit code
parameter so that the --help option handling can exit the program with
a success exit code.
Signed-off-by: Andreas Bombe
commit b611549
Author: Andreas Bombe
Date: Mon Mar 16 21:08:35 2015 +0100
Do not expect EOF as return value of getopt()
Signed-off-by: Andreas Bombe
commit a41fc32
Author: Andreas Bombe
Date: Wed Mar 11 21:45:04 2015 +0100
fsck.fat: Fix read beyond end of array on FAT12
When a FAT12 filesystem contains an odd number of clusters, setting the
last cluster with set_fat() will trigger a read of the next entry,
which does not exist in the fat array allocated for this.
Round up the allocation to an even number of FAT entries for FAT12 so
that this is fixed without introducing special casing in get_fat().
Signed-off-by: Andreas Bombe
commit e18d72f
Author: Álvaro Fernández Rojas
Date: Sat Mar 7 13:23:23 2015 +0100
Remove linux/msdos_fs.h includes
Signed-off-by: Álvaro Fernández Rojas
Signed-off-by: Andreas Bombe
commit 245d0cc
Author: Álvaro Fernández Rojas
Date: Thu Feb 26 19:22:54 2015 +0100
Remove non standard int types
Signed-off-by: Álvaro Fernández Rojas
Signed-off-by: Andreas Bombe
commit 628b3ec
Author: Álvaro Fernández Rojas
Date: Thu Feb 26 19:41:19 2015 +0100
Improve .gitignore
Signed-off-by: Álvaro Fernández Rojas
Signed-off-by: Andreas Bombe
commit 02b5a6d
Author: Álvaro Fernández Rojas
Date: Thu Feb 26 12:51:23 2015 +0100
Makefile: avoid using install -D
OS X and FreeBSD are not compatible with this option
Signed-off-by: Álvaro Fernández Rojas
Signed-off-by: Andreas Bombe
commit fb6aea6
Author: Álvaro Fernández Rojas
Date: Thu Feb 26 12:50:57 2015 +0100
Makefile: fix typo in uninstall-man
Signed-off-by: Álvaro Fernández Rojas
Signed-off-by: Andreas Bombe
commit 357ab07
Author: Andreas Bombe
Date: Thu Feb 19 21:22:54 2015 +0100
fsck.fat: Make -r option default
Default mode is now to interactively repair with the option to write
the changes back at the end (like the -r option) instead of the previous
default mode of interactively correcting but never writing back the
changes.
The -r option continues to be recognized by fsck.fat.
Signed-off-by: Andreas Bombe
commit 266a5fc
Author: Andreas Bombe
Date: Fri Feb 20 03:19:28 2015 +0100
mkfs.fat: Allow 0xF0 to be specified as media byte
Let the -M option accept 0xF0, which should be the proper descriptor
byte for 3.5" 1.44 MB and 2.88 MB floppies.
Also split the error reporting for -M between badly formatted and
invalid numbers.
Signed-off-by: Andreas Bombe
commit 16e97e9
Author: Andreas Bombe
Date: Wed Jan 28 15:07:18 2015 +0100
Make all char* that may take literals const
Every char* variable or function argument that may be given a literal
string is now made const. Additionally add -Wwrite-strings to CFLAGS to
enable a warning where const would be missing.
Signed-off-by: Andreas Bombe
commit 8b8948c
Author: Andreas Bombe
Date: Mon Dec 29 18:24:54 2014 +0100
mkfs.fat.c: Use unsigned char for binary data
Simple char technically works for the dummy_boot_jump variables, but
some compiler warning settings would give a warning over a signed char
overflowing with the values given as initializers.
Signed-off-by: Andreas Bombe
commit 45aeed0
Author: Andreas Bombe
Date: Sun Nov 16 19:49:22 2014 +0100
Fix indentation of "fix power loss damage" commit
I fixed the (sometimes misleading) indentation of the code introduced in
commit 6893c45 to be in line with the reset of the code and also removed
the /* PATCH ED+DL */ comments which are quite meaningless to the wider
world.
Signed-off-by: Andreas Bombe
commit 75c5446
Author: Andreas Bombe
Date: Sat Nov 15 16:48:48 2014 +0100
Add tags and editor backup files to .gitignore
Signed-off-by: Andreas Bombe
commit cb98ae2 (tag: v3.0.27)
Author: Andreas Bombe
Date: Wed Nov 12 01:10:23 2014 +0100
Releasing version 3.0.27.
Signed-off-by: Andreas Bombe
commit e2c8f06
Author: Andreas Bombe
Date: Wed Nov 12 00:22:17 2014 +0100
fsck.fat: Don't print version string every time -v is encountered
Remove the printing of the version string every time -v is seen during
command line parsing in fsck.fat. The version string is printed anyway
before opening the filesystem device/image.
Signed-off-by: Andreas Bombe
commit 82076b6
Author: Andreas Bombe
Date: Tue Nov 11 23:25:30 2014 +0100
Fix attempt to rename root dir in fsck due to uninitialized fields
When add_file() is called with offset 0, it will construct a DIR_ENT for
the root directory instead of reading the contents from the filesystem.
It did not initialize the whole DIR_ENT on the stack, just select
values.
In particular, the lcase field was left with an undefined value. If
that value happened to include the FAT_NO_83NAME bit, the "neither long
nor short file name" check in bad_name() added in 3.0.26 would trigger
and cause an attempt to rename the entry (which is not possible).
Example run:
$ /sbin/fsck.fat -y bad.img
fsck.fat 3.0.26 (2014-03-07)
/
Bad short file name ().
Auto-renaming it.
Renamed to
bad.img: 14 files, 19388/403266 clusters
This commit changes the initialization zeroize the whole struct before
setting individual fields. Thanks to AlexisM, who found the cause and
posted a patch on the Debian bug http://bugs.debian.org/764992 .
Signed-off-by: Andreas Bombe
commit c24ecb6
Author: Andreas Bombe
Date: Tue Nov 11 22:49:50 2014 +0100
Support long file names in volume labeling code
The code to find the volume label directory entry in find_volume_de()
did not consider long file names so far. Directory entries that make up
long file names have four attribute bits set, including the "volume"
bit.
This caused the code to mistake a directory entry that is part of a
long file name as the volume name entry. If such an entry is found
first, fatlabel would print garbage when asked to display the label and
mangle the long file name when asked to set it. The latter would lead
to the loss of the long file name and require a fsck to clean up.
Change so that the set of attributes equal that of LFN entries will no
langer match as a volume label.
Signed-off-by: Andreas Bombe
commit 1646f6e (tag: v3.0.26)
Author: Daniel Baumann
Date: Fri Mar 7 18:40:13 2014 +0100
Releasing version 3.0.26.
Signed-off-by: Daniel Baumann
commit 6893c45
Author: Dir Lotter
Date: Fri Mar 7 18:25:39 2014 +0100
Fix "odd" files created by frequent power-loss.
After running many power losses the filesystem can degrate, containing "odd"
files making the filesystem corrupt that could not be solved by fsck:
* file was not visible in a ls -l
* ls -i reported a "invalid file name" on the console
* a test program with diropen/dirread showed the file, a stat on this file
failed
* file was not accessible and could not be deleted
After digging into the code we found why fsck didn't repair the file system:
One thing was we don't have short filenames. Another issue was that the LFN
pointer was set to NULL and so it looked like we didn't have short and long
filenames.
Our patch of check.c includes:
* returns 1 from function bad_name() in case no short and no long filename
exist
* auto_rename() and rename_file() got a special handling for the case no short
file name exist:
- it enables the short file name (we think here was a weakness of the old
code: it changed the short filename but didn't enabled it in the
file->dir_ent.lcase entry)
- it reset all attributes except ATTR_DIR and ATTR_VOLUME
This solved our problem pretty well.
Signed-off-by: Daniel Baumann
commit 621e11f
Author: Natanael Copa
Date: Sat Feb 8 18:53:30 2014 +0100
Build fixes for musl libc.
Signed-off-by: Daniel Baumann