| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
| |
Avoid a bit of repeating code by merging checking fd for correctness
and fd to FILE lookup into a single routine and converting the rest of
the code to use it.
Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
| |
Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
| |
Drop needless OOM check since xzalloc() will never return NULL.
Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
open_and_lseek() increases the file size when the file is opened in
write mode and scrolled past the files end. This fails badly on /dev/mem
because loff_t which we use for the file size is signed variable, which
is used as an unsigned variable in /dev/mem. To catch this case do not
try to truncate FILE_SIZE_STREAM sized files.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
| |
| |
| |
| |
| |
| | |
ftruncate needs to set errno correctly on error.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
All these checks are really testing is that resulting position is
within [0; f->size] interval. Convert all of the custom checks into a
signle one done after the switch statement to simplify the code.
Note this change also disables the validity check for f->size ==
FILE_SIZE_STREAM and whence == SEEK_END, but lseek(stream_fd, offset,
SEEK_END) wasn't a meaningful operation to begin with, so this
shouldn't be a problem.
Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
On 32-bit systems, checking for IS_ERR_VALUE(pos) is not
correct. Expanding that code we get (loff_t cast is added for clarity):
(loff_t)pos >= (unsigned long)-MAX_ERRNO
given that loff_t is a 64-bit signed value, any perfectly valid seek
offset that is greater than 0xffffc000 will result in false positive.
Moreso, as a part of fix introduced in e10efc5080 ("fs: fix memory
access via /dev/mem for MIPS64") it doesn't really solve the problem
completely on on 64-bit platforms, becuase it still leaves out a
number of perfectly valid offsets (e.g. "md 0xffffffffffffff00"
doesn't work)
Undo the original change and convert the check to simply test if
offset is negative.
Changes neccessary to alllow access to end of 64-bit address space
will be implemented in the follow-up patch.
Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Returning requested offset from .lseek() callback doesn't really give
us any new information while bringing unnecessary
complications. Change all .lseek() types (both in struct struct
cdev_operations and in struct fs_driver_d) to return 'int' and adjust
the rest of the codebase accordingly.
Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
There are no FS drivers that do not implement .lseek callback in the
codebase, so there doesn't seem to exist a use-case where lseek()
would return -ENOSYS due to fsdrv->lseek being NULL. At the same time
a large number of FS drivers implement only the most basic "always
succeeds" custom .lseek() hook.
Change the code of lseek() to treat absense of .lseek() to mean that
no special actions needs to be taken by FS driver and seek is always
successful and drop all of the trivial .lseek() implementations.
Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Instead on relying on driver callbacks to update 'pos' in FILE, do it
as a part of lseek() code. This allows us to drop a bit of repeating
code as well as making lseek() implementation consistent with write()
and read().
Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|/
|
|
|
| |
Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
| |
we use base before checking if it is NULL
Signed-off-by: Oleksij Rempel <linux@rempel-privat.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
| |
Signed-off-by: Oleksij Rempel <linux@rempel-privat.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| | |
Introduce dev_set_name() in order to hide implementation details of
setting device's name so it'd be easier to change it.
Signed-off-by: Andrey Smirnov <andrew.smirnov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|\ \ |
|
| |/
| |
| |
| | |
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|/
|
|
|
|
|
| |
d_revalidate is useful when filesystems change under the hood of the
fs layer. This can happen with network filesystems or with devfs.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
| |
Implement clear_nlink and set_nlink and remove the private versions
from UBIFS.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
| |
Implement in fs core rather than using a private version in UBIFS.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
| |
To ease code porting from Linux
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
|
| |
stat() needs to set errno correctly when returning with an error.
Fixes: b3fbfad7ae ("fs: dentry cache implementation")
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
|
|
| |
When the initial lookup fails in open we have to go to the error path
which sets errno correctly rather than returning directly.
Fixes: b3fbfad7ae ("fs: dentry cache implementation")
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This adds the Linux dentry cache implementation to barebox.
Until now every filesystem driver resolves the full path to a file for
itself. This leads to code duplication and is error prone since
resolving paths is a complicated task. Also it can narrow down the
lookup performance since barebox only knows ASCII paths and has no way
of caching lookups.
With this patch we get the Linux dcache implementation. The path
resolving code from fs/namei.c is nearly taken as-is, minus the RCU
and locking code. Dcaching is made simple as of now: We simply cache
everything and never release any dentries. Although we do reference
counting for inodes and dentries it is effectively not used yet.
We never free anything until a fs is unmounted in which case we free
everything no matter if references are taken or not.
This patch also contains a wrapper in fs/legacy.c to support
filesystems with the old API.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
|
| |
lseek checks for non-negative in-memory offsets (addresses), failing otherwise.
However negative address 0xffffffffXXXXXXXX is a valid MIPS64 virtual address.
Signed-off-by: Peter Mamonov <pmamonov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
| |
When there are links in the path mkdir/rmdir are called with then
canonicalize_path is needed which resolves the links.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
| |
fs.c has some whitespaces where there should be tabs. Fix it.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
| |
fs.c is already heavily filled with code. Move mem_write/mem_read
to devfs-core where it fits better.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In __canonicalize_path() we dereference the pointer returned by
get_fsdevice_by_path() without checking if the pointer is NULL or not.
When the pointer is NULL it leads to an Ooops.
Ooops, address error on load or ifetch!
$ 0 : 00000000 00000001 a0000026 a0811c10
$ 4 : a0402e60 a0402e48 a0811c00 a0402e58
$ 8 : 00000001 00000000 0000005a 00000023
$12 : 00000000 00000002 00601021 00000000
$16 : a0402e60 a0402e50 a0402e39 a0810000
$20 : a0402e38 a0811420 a0811424 00000000
$24 : 00000000 a080de10
$28 : 87f87d40 a03ffa68 a0810000 a080ce30
Hi : 00000002
Lo : 00000000
epc : a080ce34
ra : a080ce30
Status: 00000006
Cause : 40008010
Config: 80040483
### ERROR ### Please RESET the board ###
Fixes: d79a81736 fs: Don't bother filesystems without link support with additional stat() calls
Signed-off-by: Gaël PORTAY <gael.portay@savoirfairelinux.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Some commands need files in which they can lseek backwards which
is particularly not possible on TFTP. Instead of hiding this
behind can_lseek_backward() create a function for it which tests
if the file is on TFTP directly rather than using different
lseek operations.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
| |
| |
| |
| |
| |
| |
| |
| | |
When loop mounting, allow to specify an offset into the file, similarly
to the losetup offset option. Multiplicative suffixes are supported.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|/
|
|
|
|
|
|
| |
"path" is allocated by normalise_path() and thus must be
freed. This was done in the error path, but not in the success
path.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|\ |
|
| |
| |
| |
| |
| | |
Signed-off-by: Antony Pavlov <antonynpavlov@gmail.com>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|/
|
|
| |
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
In __canonicalize_path() we only call stat() to know if the path
is a link or not. When the filesystem doesn't support links we
already know that it's not a link, so we do not need to call stat().
This helps the tftp filesystem since the parent directories of
a file to be opened won't be stat()ed anymore, something tftp
does not support.
Fixes: a602bebc fs: Implement links to directories
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
|
|
| |
In __canonicalize_path pathes beginning with '//' can occur. This
is normally not a problem since normalize_path() will clean this
up, but it means we cannot call get_fsdevice_by_path() on these
pathes in this function, as needed in the next patch.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
| |
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Allow to create a loopback cdev from a file.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Fixed up with:
fs: Makefile: Add parseopt to all builds
parseopt.h was included to fs.c with commit 9248b, but parseopt.o has a
dependency to CONFIG_FS_NFS.
Moved parseopt.o to the default build to eliminate build failures.
Signed-off-by: Daniel Schultz <d.schultz@phytec.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
In barebox the default mount path for a cdev is /mnt/<cdevname>
which can be mounted with "mount <cdevname>" without specifying
a target path explicitly. Simplify this further by creating automount
entries for the default mount pathes which makes a manual mount
completely unnecessary.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
| | |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
So far links can only point to files. Implement links to
directories. With this all kinds of links are supported:
- relative links
- absolute links
- links including ".."
- link loops (are detected, return -EMLINK)
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
path_check_prereq() has some bugs:
- It's called with S_IFLNK, but this is effectively a no-op because this
case is not implemented in the function
- It uses lstat, but if stat or lstat should be used depends on the
usecase: For opendir() stat() must be used, since we are not
interested in the link, but in the target. For readlink() lstat() must
be used since we are indeed interested in the link, not the target.
It's easier and better understandable to open code the functionality
where it's needed, so drop path_check_prereq() and do the right thing
where called.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
| |
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
|
| |
All error paths before calling the driver's lseek callback return -1 and
set errno. Do the same if the callback returns an error.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
|
| |
Don't check for the file type here. Devices may have truncate
callbacks. Otherwise the caller should check if O_TRUNC is possible.
Signed-off-by: Teresa Remmet <t.remmet@phytec.de>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|
|
|
|
|
|
|
| |
Check that readonly openen files is not written to and writeonly
files is not read from.
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
|