The code for handing file overwrite incorrectly assumed that the file on
disk is always contiguous. This resulted in corrupting disk structure
every time when write to existing fragmented file happened. Fix this
by adding proper check for cluster discontinuity and adjust chunk size
on each partial write.
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
This patch partially fixes the issue revealed by the following test
script:
--->8-fat_test1.sh---
#!/bin/bash
make sandbox_defconfig
make
dd if=/dev/zero of=/tmp/10M.img bs=1024 count=10k
mkfs.vfat -v /tmp/10M.img
cat >/tmp/cmds <<EOF
x
host bind 0 /tmp/10M.img
fatls host 0
mw 0x1000000 0x0a434241 0x1000 # "ABC\n"
mw 0x1100000 0x0a464544 0x8000 # "DEF\n"
fatwrite host 0 0x1000000 file0001.raw 0x1000
fatwrite host 0 0x1000000 file0002.raw 0x1000
fatwrite host 0 0x1000000 file0003.raw 0x1000
fatwrite host 0 0x1000000 file0004.raw 0x1000
fatwrite host 0 0x1000000 file0005.raw 0x1000
fatrm host 0 file0002.raw
fatrm host 0 file0004.raw
fatls host 0
fatwrite host 0 0x1100000 file0007.raw 0x4000
fatwrite host 0 0x1100000 file0007.raw 0x4000
reset
EOF
./u-boot </tmp/cmds
#verify
rm -r /tmp/result /tmp/model
mkdir /tmp/result
mkdir /tmp/model
yes ABC | head -c 4096 >/tmp/model/file0001.raw
yes ABC | head -c 4096 >/tmp/model/file0003.raw
yes ABC | head -c 4096 >/tmp/model/file0005.raw
yes DEF | head -c 16384 >/tmp/model/file0007.raw
mcopy -n -i /tmp/10M.img ::file0001.raw /tmp/result
mcopy -n -i /tmp/10M.img ::file0003.raw /tmp/result
mcopy -n -i /tmp/10M.img ::file0005.raw /tmp/result
mcopy -n -i /tmp/10M.img ::file0007.raw /tmp/result
hd /tmp/10M.img
if diff -urq /tmp/model /tmp/result
then
echo Test okay
else
echo Test fail
fi
--->8---
Overwritting a discontiguous test file (file0007.raw) no longer causes
corruption to file0003.raw, which's data lies between the chunks of the
test file. The amount of data written to disk is still incorrect, what
causes damage to the file (file0005.raw), which's data lies next to the
test file. This will be fixed by the next patch.
Feel free to prepare a proper sandbox/py_test based tests based on the
provided test scripts.
Unlink test for FAT file system seems to fail at test_unlink2.
(When I added this test, I haven't seen any errors though.)
for example,
===8<===
fs_obj_unlink = ['fat', '/home/akashi/tmp/uboot_sandbox_test/128MB.fat32.img']
def test_unlink2(self, u_boot_console, fs_obj_unlink):
"""
Test Case 2 - delete many files
"""
fs_type,fs_img = fs_obj_unlink
with u_boot_console.log.section('Test Case 2 - unlink (many)'):
output = u_boot_console.run_command('host bind 0 %s' % fs_img)
for i in range(0, 20):
output = u_boot_console.run_command_list([
'%srm host 0:0 dir2/0123456789abcdef%02x' % (fs_type, i),
'%sls host 0:0 dir2/0123456789abcdef%02x' % (fs_type, i)])
assert('' == ''.join(output))
output = u_boot_console.run_command(
'%sls host 0:0 dir2' % fs_type)
> assert('0 file(s), 2 dir(s)' in output)
E AssertionError: assert '0 file(s), 2 dir(s)' in ' ./\r\r\n ../\r\r\n 0 0123456789abcdef11\r\r\n\r\r\n1 file(s), 2 dir(s)'
test/py/tests/test_fs/test_unlink.py:52: AssertionError
===>8===
This can happen when fat_itr_next() wrongly detects an already-
deleted directory entry.
File deletion, which was added in the commit f8240ce95d ("fs: fat:
support unlink"), is implemented by marking its entry for a short name
with DELETED_FLAG, but related entry slots for a long file name are kept
unmodified. (So entries will never be actually deleted from media.)
To handle this case correctly, an additional check for a directory slot
will be needed in fat_itr_next().
In addition, I added extra comments about long file name and short file
name format in FAT file system. Although they are not directly related
to the issue, I hope it will be helpful for better understandings
in general.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
When hitting an invalid FAT cluster while reading a file always print an
error message and return an error code.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
File was found on specified location. Info about file was read,
but then immediately destroyed using 'free' call. As a result
file size was set to 0, hence fat process didn't read any data.
Premature 'free' call removed. Resources are freed right before
function return. File is read correctly.
Signed-off-by: Martin Vystrcil <martin.vystrcil@m-linux.cz>
Contrary to fat12/16, fat32 can have root directory at any location
and its size can be expanded.
Without this patch, root directory won't grow properly and so we will
eventually fail to add files under root directory. Please note that this
can happen even if you delete many files as deleted directory entries
are not reclaimed but just marked as "deleted" under the current
implementation.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Tested-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
When a long name directory entry is created, multiple directory entries
may be occupied across a directory cluster boundary. Since only one
directory cluster is cached in a directory iterator, a first cluster must
be written back to device before switching over a second cluster.
Without this patch, some added files may be lost even if you don't see
any failures on write operation.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Tested-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
With the commit below, fat now correctly handles a file read under
a non-cluster-aligned root directory of fat12/16.
Write operation should be fixed in the same manner.
Fixes: commit 9b18358dc0 ("fs: fat: fix reading non-cluster-aligned
root directory")
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Cc: Anssi Hannula <anssi.hannula@bitwise.fi>
Tested-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
fat_itr_root() allocates fatbuf so we free it on the exit path, if
the function fails we should not free it, check the return value
and skip freeing if the function fails.
Signed-off-by: Andrew F. Davis <afd@ti.com>
A FAT12/FAT16 root directory location is specified by a sector offset and
it might not start at a cluster boundary. It also resides before the
data area (before cluster 2).
However, the current code assumes that the root directory is located at
a beginning of a cluster, causing no files to be found if that is not
the case.
Since the FAT12/FAT16 root directory is located before the data area
and is not aligned to clusters, using unsigned cluster numbers to refer
to the root directory does not work well (the "cluster number" may be
negative, and even allowing it be signed would not make it properly
aligned).
Modify the code to not use the normal cluster numbering when referring to
the root directory of FAT12/FAT16 and instead use a cluster-sized
offsets counted from the root directory start sector.
This is a relatively common case as at least the filesystem formatter on
Win7 seems to create such filesystems by default on 2GB USB sticks when
"FAT" is selected (cluster size 64 sectors, rootdir size 32 sectors,
rootdir starts at half a cluster before cluster 2).
dosfstools mkfs.vfat does not seem to create affected filesystems.
Signed-off-by: Anssi Hannula <anssi.hannula@bitwise.fi>
Reviewed-by: Bernhard Messerklinger <bernhard.messerklinger@br-automation.com>
Tested-by: Bernhard Messerklinger <bernhard.messerklinger@br-automation.com>
When compiling with DEBUG=1 an error
fs/fat/fat_write.c:831: undefined reference to `__aeabi_ldivmod'
occurred.
We should use do_div() instead of the modulus operator.
filesize and cur_pos cannot be negative. So let's use u64 to avoid
warnings.
Fixes: cb8af8af5b ("fs: fat: support write with non-zero offset")
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Release cluster block immediately when no longer use would help to reduce
64KiB memory allocated to the memory pool.
Signed-off-by: Tien Fong Chee <tien.fong.chee@intel.com>
Drop the statically allocated get_contents_vfatname_block and
dynamically allocate a buffer only if required. This saves
64KiB of memory.
Signed-off-by: Stefan Agner <stefan.ag...@toradex.com>
Signed-off-by: Tien Fong Chee <tien.fong.chee@intel.com>
Most of the time SPL only needs very simple FAT reading, so having
CONFIG_IS_ENABLED(FAT_WRITE) to exclude it from SPL build would help
to save 64KiB default max clustersize from memory.
Signed-off-by: Tien Fong Chee <tien.fong.chee@intel.com>
Reviewed-by: Simon Goldschmidt <simon.k.r.goldschmidt@gmail.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
This particular commit is causing a regression on stih410-b2260 and
other platforms when reading from FAT16. Noting that I had rebased the
original fix from Thomas onto then-current master, there is also
question from Akashi-san if the change is still needed after other FAT
fixes that have gone in.
This reverts commit a68b0e11ea.
Reported-by: Patrice Chotard <patrice.chotard@st.com>
Cc: AKASHI Takahiro <takahiro.akashi@linaro.org>
Cc: Thomas RIENOESSL <thomas.rienoessl@bachmann.info>
Signed-off-by: Tom Rini <trini@konsulko.com>
The long name apparently can be accumulated using multiple
13-byte slots. Unfortunately we never checked how many we
can actually fit in the buffer we are reading to.
Signed-off-by: Patrick Wildt <patrick@blueri.se>
The cluster size specifies how many sectors make up a cluster. A
cluster size of zero makes no sense, as it would mean that the
cluster is made up of no sectors. This will later lead into a
division by zero in sect_to_clust(), so better take care of that
early.
The MAX_CLUSTSIZE define can reduced using a define to make some
room in low-memory system. Unfortunately if the code reads a
filesystem with a bigger cluster size it will overflow the buffer.
Signed-off-by: Patrick Wildt <patrick@blueri.se>
This fixes problems accessing drives formated under
Windows as FAT16.
Signed-off-by: Thomas RIENOESSL <thomas.rienoessl@bachmann.info>
[trini: Rebase on top of f528c140c8]
Signed-off-by: Tom Rini <trini@konsulko.com>
As observed with clang:
fs/fat/fat_write.c:1024:13: warning: comparison of constant 128
with expression of type 'char' is always false
[-Wtautological-constant-out-of-range-compare]
if ((0x80 <= c) && (c <= 0xff))
~~~~ ^ ~
fs/fat/fat_write.c:1024:25: warning: comparison of constant 255
with expression of type 'char' is always true
[-Wtautological-constant-out-of-range-compare]
if ((0x80 <= c) && (c <= 0xff))
~ ^ ~~~~
Fixes: 25bb9dab14 ("fs: fat: check and normalize file name")
Signed-off-by: Tom Rini <trini@konsulko.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
Avoid CoverityScan warning SIGN_EXTENSION by changing the type of
parameter size of set_cluster().
Avoid leaking stack content when writing an incomplete last sector.
Reported-by: Coverity (CID: 184096)
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Do not leak filename_copy in case of error.
Catch out of memory when calling strdup.
Reported-by: Coverity (CID: 184086)
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
The FAT driver supports unaligned reads and writes and EFI applications
will make use of these. So a misaligned buffer is only worth a debug
message.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Signed-off-by: Alexander Graf <agraf@suse.de>
In this patch, unlink support is added to FAT file system.
A directory can be deleted only if it is empty.
In this implementation, only a directory entry for a short file name
will be removed. So entries for a long file name can and should be
reclaimed with fsck.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
In this patch, mkdir support is added to FAT file system.
A newly created directory contains only "." and ".." entries.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
The starting cluster number of directory is needed to initialize ".."
(parent directory) entry when creating a new directory.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
In this patch, all the necessary code for allowing for a file offset
at write is implemented. What plays a major roll here is get_set_cluster(),
which, in contrast to its counterpart, set_cluster(), only operates on
already-allocated clusters, overwriting with data.
So, with a file offset specified, set_contents() seeks and writes data
with set_get_cluster() until the end of a file, and, once it reaches
there, continues writing with set_cluster() for the rest.
Please note that a file will be trimmed as a result of write operation if
write ends before reaching file's end. This is an intended behavior
in order to maintain compatibility with the current interface.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
The current write implementation is quite simple: remove existing clusters
and then allocating new ones and filling them with data. This, inevitably,
enforces always writing from the beginning of a file.
As the first step to lift this restriction, fat_file_write() and
set_contents() are modified to accept an additional parameter, file offset
and further re-factored so that, in the next patch, all the necessary code
will be put into set_contents().
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
In this patch, write implementation is overhauled and rewritten by
making full use of directory iterator. The obvious bonus is that we are
now able to write to a file with a directory path, like /A/B/C/FILE.
Please note that, as there is no notion of "current directory" on u-boot,
a file name specified must contain an absolute directory path. Otherwise,
"/" (root directory) is assumed.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
It would be good that FAT write function return error code instead of
just returning -1 as fat_read_file() does.
This patch attempts to address this issue although it is 'best effort
(or estimate)' for now.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
FAT file system's long file name support is a bit complicated and has some
restrictions on its naming. We should be careful about it especially for
write as it may easily end up with wrong file system.
normalize_longname() check for the rules and normalize a file name
if necessary. Please note, however, that this function is yet to be
extended to fully comply with the standard.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
This reverts commit 0dc1bfb730.
The succeeding patch series will supersede it.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
In my attempt to re-work write operation, it was revealed that iterator's
"clust" does not always point to a cluster to which a current directory
entry ("dent") belongs.
This patch assures that it is always true by adding "next_clust" which is
used solely for dereferencing a cluster chain.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
FAT's root directory does not have "." nor ".."
So care must be taken when scanning root directory with fat_itr_resolve().
Without this patch, any file path starting with "." or ".." will not be
resolved at all.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
get_fs_info() was introduced in major re-work of read operation by Rob.
We want to reuse this function in write operation by extending it with
additional members in fsdata structure.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
In order to make the debug print in file_fat_read_at() a tad more useful,
show the offset the file is being read at alongside the filename.
Suggested-by: Tero Kristo <t-kristo@ti.com>
Signed-off-by: Andreas Dannenberg <dannenberg@ti.com>
Reviewed-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
fs_fat_write() is not able to write to subdirectories.
Currently if a filepath with a leading slash is passed, the slash is
treated as part of the filename to be created in the root directory.
Strip leading (back-)slashes.
Check that the remaining filename does not contain any illegal characters
(<>:"/\|?*). This way we will throw an error when trying to write to a
subdirectory.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Signed-off-by: Alexander Graf <agraf@suse.de>
When U-Boot started using SPDX tags we were among the early adopters and
there weren't a lot of other examples to borrow from. So we picked the
area of the file that usually had a full license text and replaced it
with an appropriate SPDX-License-Identifier: entry. Since then, the
Linux Kernel has adopted SPDX tags and they place it as the very first
line in a file (except where shebangs are used, then it's second line)
and with slightly different comment styles than us.
In part due to community overlap, in part due to better tag visibility
and in part for other minor reasons, switch over to that style.
This commit changes all instances where we have a single declared
license in the tag as both the before and after are identical in tag
contents. There's also a few places where I found we did not have a tag
and have introduced one.
Signed-off-by: Tom Rini <trini@konsulko.com>
fat.h unconditionally defines CONFIG_SUPPORT_VFAT (and has done since
2003), so as a result VFAT support is always enabled regardless of
whether a board config defines it or not. Drop this unnecessary option.
Signed-off-by: Tuomas Tynkkynen <tuomas@tuxera.com>
The message "reading %s\n" may be interesting when
debugging but otherwise it is superfluous.
Only output the message when debugging.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
commit 21a24c3bf3 ("fs/fat: fix case for FAT shortnames") made it
possible that get_name() returns file names with some upper cases.
find_directory_entry() must be updated to take this account, and use
case-insensitive functions to compare file names.
Signed-off-by: Jean-Jacques Hiblot <jjhiblot@ti.com>
Before this patch one could receive following errors when executing "fatls"
command on machine with cache enabled (ex i.MX6Q) :
=> fatls mmc 0:1
CACHE: Misaligned operation at range [4f59dfc8, 4f59e7c8]
CACHE: Misaligned operation at range [4f59dfc8, 4f59e7c8]
ERROR: v7_outer_cache_inval_range - start address is not aligned - 0x4f59dfc8
ERROR: v7_outer_cache_inval_range - stop address is not aligned - 0x4f59e7c8
CACHE: Misaligned operation at range [4f59dfc8, 4f59e7c8]
CACHE: Misaligned operation at range [4f59dfc8, 4f59e7c8]
ERROR: v7_outer_cache_inval_range - start address is not aligned - 0x4f59dfc8
ERROR: v7_outer_cache_inval_range - stop address is not aligned - 0x4f59e7c8
To alleviate this problem - the calloc()s have been replaced with
malloc_cache_aligned() and memset().
After those changes the buffers are properly aligned (with both start
address and size) to SoC cache line.
Fixes: 09fa964bba ("fs/fat: Fix 'CACHE: Misaligned operation at range' warnings")
Suggested-by: Lukasz Majewski <lukma@denx.de>
Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
Reviewed-by: Lukasz Majewski <lukma@denx.de>
Reviewed-by: Fabio Estevam <fabio.estevam@nxp.com>
Check malloc() return values and properly unwind on errors so
memory allocated for fat_itr structures get freed properly.
Also fixes a leak of fsdata.fatbuf in fat_size().
Fixes: 2460098cff ("fs/fat: Reduce stack usage")
Reported-by: Coverity (CID: 167225, 167233, 167234)
Signed-off-by: Tuomas Tynkkynen <tuomas.tynkkynen@iki.fi>
Reviewed-by: Tom Rini <trini@konsulko.com>
The 'block' field of fat_itr needs to be properly aligned for DMA and
while it does have '__aligned(ARCH_DMA_MINALIGN)', the fat_itr structure
itself needs to be properly aligned as well.
While at it use malloc_cache_aligned() for the other aligned allocations
in the file as well.
Fixes: 2460098cff ("fs/fat: Reduce stack usage")
Signed-off-by: Tuomas Tynkkynen <tuomas.tynkkynen@iki.fi>
Reviewed-by: Tom Rini <trini@konsulko.com>
If we end up back in the root directory via a '..' directory entry, set
itr->is_root accordingly. Failing to do that gives spews like
"Invalid FAT entry" and being unable to access directory entries located
past the first cluster of the root directory.
Fixes: 8eafae209c ("fat/fs: convert to directory iterators")
Reviewed-by: Tom Rini <trini@konsulko.com>
Signed-off-by: Tuomas Tynkkynen <tuomas.tynkkynen@iki.fi>
We have limited stack in SPL builds. Drop itrblock and move to
malloc/free of itr to move this off of the stack. As part of this fix a
double-free issue in fat_size().
Signed-off-by: Tom Rini <trini@konsulko.com>
---
Rework to use malloc/free as moving this to a global overflows some SH
targets.
A new fatbuf was allocated by get_fs_info() (called by fat_itr_root()),
but not freed, resulting in eventually running out of memory. Spotted
by running 'ls -r' in a large FAT filesystem from Shell.efi.
fatbuf is mainly used to cache FAT entry lookups (get_fatent())..
possibly once fat_write.c it can move into the iterator to simplify
this.
Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Łukasz Majewski <lukma@denx.de>
Use the clust_to_sect() helper that was introduced earlier, and add an
inverse sect_to_clust(), plus update the various spots that open-coded
this conversion previously.
Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
