The quotaoff operation has a race with inode allocation that results
in a livelock. An inode allocation that occurs before the quota
status flags are updated acquires the appropriate dquots for the
inode via xfs_qm_vop_dqalloc(). It then inserts the XFS_INEW inode
into the perag radix tree, sometime later attaches the dquots to the
inode and finally clears the XFS_INEW flag. Quotaoff expects to
release the dquots from all inodes in the filesystem via
xfs_qm_dqrele_all_inodes(). This invokes the AG inode iterator,
which skips inodes in the XFS_INEW state because they are not fully
constructed. If the scan occurs after dquots have been attached to
an inode, but before XFS_INEW is cleared, the newly allocated inode
will continue to hold a reference to the applicable dquots. When
quotaoff invokes xfs_qm_dqpurge_all(), the reference count of those
dquot(s) remain elevated and the dqpurge scan spins indefinitely.
To address this problem, update the xfs_qm_dqrele_all_inodes() scan
to wait on inodes marked on the XFS_INEW state. We wait on the
inodes explicitly rather than skip and retry to avoid continuous
retry loops due to a parallel inode allocation workload. Since
quotaoff updates the quota state flags and uses a synchronous
transaction before the dqrele scan, and dquots are attached to
inodes after radix tree insertion iff quota is enabled, one INEW
waiting pass through the AG guarantees that the scan has processed
all inodes that could possibly hold dquot references.
Reported-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
The AG inode iterator currently skips new inodes as such inodes are
inserted into the inode radix tree before they are fully
constructed. Certain contexts require the ability to wait on the
construction of new inodes, however. The fs-wide dquot release from
the quotaoff sequence is an example of this.
Update the AG inode iterator to support the ability to wait on
inodes flagged with XFS_INEW upon request. Create a new
xfs_inode_ag_iterator_flags() interface and support a set of
iteration flags to modify the iteration behavior. When the
XFS_AGITER_INEW_WAIT flag is set, include XFS_INEW flags in the
radix tree inode lookup and wait on them before the callback is
executed.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Inodes that are inserted into the perag tree but still under
construction are flagged with the XFS_INEW bit. Most contexts either
skip such inodes when they are encountered or have the ability to
handle them.
The runtime quotaoff sequence introduces a context that must wait
for construction of such inodes to correctly ensure that all dquots
in the fs are released. In anticipation of this, support the ability
to wait on new inodes. Wake the appropriate bit when XFS_INEW is
cleared.
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Copy the uuid of the filesystem to struct super_block s_uuid field,
as several other filesystems already do. Copy regardless of the nouuid
mount option, because other filesystems also do not guaranty uniqueness
of the s_uuid field in super_block struct.
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
mempool_alloc() cannot fail if the gfp flags allow it to
sleep, and both GFP_FS allows for sleeping.
So these tests of the return value from mempool_alloc()
cannot be needed.
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Steve French <smfrench@gmail.com>
commit 620d8745b3 ("Introduce cifs_copy_file_range()") changes the
behaviour of the cifs ioctl call CIFS_IOC_COPYCHUNK_FILE. In case of
successful writes, it now returns the number of bytes written. This
return value is treated as an error by the xfstest cifs/001. Depending
on the errno set at that time, this may or may not result in the test
failing.
The patch fixes this by setting the return value to 0 in case of
successful writes.
Fixes: commit 620d8745b3 ("Introduce cifs_copy_file_range()")
Reported-by: Eryu Guan <eguan@redhat.com>
Signed-off-by: Sachin Prabhu <sprabhu@redhat.com>
Acked-by: Pavel Shilovsky <pshilov@microsoft.com>
Cc: stable@vger.kernel.org
Signed-off-by: Steve French <smfrench@gmail.com>
Incorrect return value for shares not using the prefix path means that
we will never match superblocks for these shares.
Fixes: commit c1d8b24d18 ("Compare prepaths when comparing superblocks")
Cc: stable@vger.kernel.org
Signed-off-by: Sachin Prabhu <sprabhu@redhat.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
Signed-off-by: Steve French <smfrench@gmail.com>
Lockdep complains about a possible deadlock between mount and unlink
(which is technically impossible), but fixing this improves possible
future multiple-backend support, and keeps locking in the right order.
The lockdep warning could be triggered by unlinking a file in the
pstore filesystem:
-> #1 (&sb->s_type->i_mutex_key#14){++++++}:
lock_acquire+0xc9/0x220
down_write+0x3f/0x70
pstore_mkfile+0x1f4/0x460
pstore_get_records+0x17a/0x320
pstore_fill_super+0xa4/0xc0
mount_single+0x89/0xb0
pstore_mount+0x13/0x20
mount_fs+0xf/0x90
vfs_kern_mount+0x66/0x170
do_mount+0x190/0xd50
SyS_mount+0x90/0xd0
entry_SYSCALL_64_fastpath+0x1c/0xb1
-> #0 (&psinfo->read_mutex){+.+.+.}:
__lock_acquire+0x1ac0/0x1bb0
lock_acquire+0xc9/0x220
__mutex_lock+0x6e/0x990
mutex_lock_nested+0x16/0x20
pstore_unlink+0x3f/0xa0
vfs_unlink+0xb5/0x190
do_unlinkat+0x24c/0x2a0
SyS_unlinkat+0x16/0x30
entry_SYSCALL_64_fastpath+0x1c/0xb1
Possible unsafe locking scenario:
CPU0 CPU1
---- ----
lock(&sb->s_type->i_mutex_key#14);
lock(&psinfo->read_mutex);
lock(&sb->s_type->i_mutex_key#14);
lock(&psinfo->read_mutex);
Reported-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reported-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
We want to use kthread_stop() in order to ensure the threads are
shut down before we tear down the nfs_callback_info in nfs_callback_down.
Tested-and-reviewed-by: Kinglong Mee <kinglongmee@gmail.com>
Reported-by: Kinglong Mee <kinglongmee@gmail.com>
Fixes: bb6aeba736 ("NFSv4.x: Switch to using svc_set_num_threads()...")
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Since the vmalloc code has been removed from write_pmsg() in the commit
"5bf6d1b pstore/pmsg: drop bounce buffer", remove the unused header
vmalloc.h.
Signed-off-by: Geliang Tang <geliangtang@gmail.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
Pull nfsd fixes from Bruce Fields:
"Thanks to Ari Kauppi and Tuomas Haanpää at Synopsis for spotting bugs
in our NFSv2/v3 xdr code that could crash the server or leak memory"
* tag 'nfsd-4.11-3' of git://linux-nfs.org/~bfields/linux:
nfsd: stricter decoding of write-like NFSv2/v3 ops
nfsd4: minor NFSv2/v3 write decoding cleanup
nfsd: check for oversized NFSv2/v3 arguments
Pull ceph fix from Ilya Dryomov:
"A fix for a kernel stack overflow bug in ceph setattr code, marked for
stable"
* tag 'ceph-for-4.11-rc9' of git://github.com/ceph/ceph-client:
ceph: fix recursion between ceph_set_acl() and __ceph_setattr()
Pull vfs fixes from Al Viro:
- fix orangefs handling of faults on write() - I'd missed that one back
when orangefs was going through review.
- readdir counterpart of "9p: cope with bogus responses from server in
p9_client_{read,write}" - server might be lying or broken, and we'd
better not overrun the kmalloc'ed buffer we are copying the results
into.
- NFS O_DIRECT read/write can leave iov_iter advanced by too much;
that's what had been causing iov_iter_pipe() warnings davej had been
seeing.
- statx_timestamp.tv_nsec type fix (s32 -> u32). That one really should
go in before 4.11.
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
uapi: change the type of struct statx_timestamp.tv_nsec to unsigned
fix nfs O_DIRECT advancing iov_iter too much
p9_client_readdir() fix
orangefs_bufmap_copy_from_iovec(): fix EFAULT handling
fstrim can take really long time on big, slow device or on file system
with a lots of allocation groups. Currently there is no way for the user
to cancell the operation. This patch makes it possible for the user to
kill fstrim pocess by adding the check for fatal_signal_pending() in
xfs_trim_extents().
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Reported-by: Zdenek Kabelac <zkabelac@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
The change in commit 1e2f82d1e9 ("statx: Kill fd-with-NULL-path
support in favour of AT_EMPTY_PATH") to error on a NULL pathname to
statx() is inconsistent.
It results in the error EINVAL for a NULL pathname. Other system calls
with similar APIs (fchownat(), fstatat(), linkat()), return EFAULT.
The solution is simply to remove the EINVAL check. As I already pointed
out in [1], user_path_at*() and filename_lookup() will handle the NULL
pathname as per the other APIs, to correctly produce the error EFAULT.
[1] https://lkml.org/lkml/2017/4/26/561
Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Eric Sandeen <sandeen@sandeen.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently we just stash anything we got into file->f_flags, and the
report it in fcntl(F_GETFD). This patch just clears out all unknown
flags so that we don't pass them to the fs or report them.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Add a central define for all valid open flags, and use it in the uniqueness
check.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
_submit_bh() allowed submitting a buffer_head for I/O using custom
bio_flags. It used to be used by jbd to set BIO_SNAP_STABLE, introduced
by commit 7136851117 ("mm: make snapshotting pages for stable writes a
per-bio operation"). However, the code and flag has since been removed
and no _submit_bh() users remain.
These days, bio_flags are mostly used internally by the block layer to
track the state of bio's. As such, it doesn't really make sense for
filesystems to use them instead of op_flags when wanting special
behavior for block requests.
Therefore, remove _submit_bh() and trim the bio_flags argument from
submit_bh_wbc().
Cc: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
simple_fill_super() is passed an array of tree_descr structures which
describe the files to create in the filesystem's root directory. Since
these arrays are never modified intentionally, they should be 'const' so
that they are placed in .rodata and benefit from memory protection.
This patch updates the function signature and all users, and also
constifies tree_descr.name.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Previous AFFS patch fixed OFS write operations but unveiled
another bug: files greater than 4KB are being created with a wrong
size resulting in errors like the following:
dd if=/dev/zero of=file bs=4097 count=1
cp file /mnt/affs/
cp: error writing '/mnt/affs/file': Bad address
Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
We called unconditionally affs_bread_ino() with create 0 resulting in
"error (device ...): get_block(): strange block request 0"
when trying to write on AFFS OFS format.
This patch adds create parameter to that function.
0 for affs_readpage_ofs()
1 for affs_write_begin_ofs()
Bug was found here:
https://bugzilla.kernel.org/show_bug.cgi?id=114961
Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
node generation has to be stored on disk.
AFAICS we won't be able to manage it on AFFS.
This patch removes relevant check in affs_nfs_get_inode()
Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Have that file in global include/linux is not needed.
Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
AFFS symbolic links were broken since kernel 2.6.29
Problem was bisected to the following
commit ebd09abbd9 ("vfs: ensure page symlinks are NUL-terminated")
commit 035146851c ("vfs: introduce helper function to safely
NUL-terminate symlinks")
AFFS wasn't setting inode size when reading symbolic link from disk or
creating a new one. Result was zero allocation in pagecache.
ln -s file symlink
ls -lrt
file
symlink ->
This patch adds inode isize information on inode get and symbolic link
addition.
Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
With the new statx() syscall, the following both allow the attributes of
the file attached to a file descriptor to be retrieved:
statx(dfd, NULL, 0, ...);
and:
statx(dfd, "", AT_EMPTY_PATH, ...);
Change the code to reject the first option, though this means copying
the path and engaging pathwalk for the fstat() equivalent. dfd can be a
non-directory provided path is "".
[ The timing of this isn't wonderful, but applying this now before we
have statx() in any released kernel, before anybody starts using the
NULL special case. - Linus ]
Fixes: a528d35e8b ("statx: Add a system call to make enhanced file info available")
Reported-by: Michael Kerrisk <mtk.manpages@gmail.com>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Eric Sandeen <sandeen@sandeen.net>
cc: fstests@vger.kernel.org
cc: linux-api@vger.kernel.org
cc: linux-man@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If we write zero bytes to this debugfs file, then it will cause an
underflow when we do copy_from_user(buf, ubuf, count - 1). Debugfs can
normally only be written to by root so the impact of this is low.
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
When the computer is turned off, all the processes are killed and then
all the filesystems are umounted. OrangeFS should not wait for the
userspace daemon to come back in that case.
This only works for plain umount(2). To actually take advantage of this
interactively, `umount -f' is needed; otherwise umount will issue a
statfs first, which will wait for the userspace daemon to come back.
Signed-off-by: Martin Brandenburg <martin@omnibond.com>
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
It is not necessary to take the lock and search through the request list
if the list is empty.
Signed-off-by: Martin Brandenburg <martin@omnibond.com>
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
If the mount is aborted after userspace has been asked to mount,
userspace must be told to unmount.
Ordinarily orangefs_kill_sb does the unmount. However it cannot be
called if the superblock has not been set up. This is a very narrow
window.
The NULL fs_id is not unmounted.
Signed-off-by: Martin Brandenburg <martin@omnibond.com>
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
Let the server figure this out because our size might be out of date or
not present.
The bug was that
xfs_io -f -t -c "pread -v 0 100" /mnt/foo
echo "Test" > /mnt/foo
xfs_io -f -t -c "pread -v 0 100" /mnt/foo
fails because the second truncate did not happen if nothing had
requested the size after the write in echo. Thus i_size was zero (not
present) and the orangefs_setattr though i_size was zero and there was
nothing to do.
Signed-off-by: Martin Brandenburg <martin@omnibond.com>
Cc: stable@vger.kernel.org
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
Fortunately OrangeFS has had a getattr request mask for a long time.
The server basically has two difficulty levels for attributes. Fetching
any attribute except size requires communicating with the metadata
server for that handle. Since all the attributes are right there, it
makes sense to return them all. Fetching the size requires
communicating with every I/O server (that the file is distributed
across). Therefore if asked for anything except size, get everything
except size, and if asked for size, get everything.
Signed-off-by: Martin Brandenburg <martin@omnibond.com>
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
They are clones of the ORANGEFS_ITERATE macros in use elsewhere. Delete
ORANGEFS_ITERATE_NEXT which is a hack previously used by readdir.
Signed-off-by: Martin Brandenburg <martin@omnibond.com>
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
This works by maintaining a linked list of pages which the directory
has been read into rather than one giant fixed-size buffer.
This replaces code which limits the total directory size to the total
amount that could be returned in one server request. Since filenames
are usually considerably shorter than the maximum, the old code could
usually handle several server requests before running out of space.
Signed-off-by: Martin Brandenburg <martin@omnibond.com>
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
This and the previous commit fix xfstests generic/257.
Signed-off-by: Martin Brandenburg <martin@omnibond.com>
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
In the past, readdir assumed that the user buffer will be large enough
that all entries from the server will fit. If this was not true,
entries would be skipped.
Since it works now, request 512 entries rather than 96 per server
operation.
Signed-off-by: Martin Brandenburg <martin@omnibond.com>
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
Also don't check flags as this has been validated by the VFS already.
Fix an off-by-one error in the max size checking.
Stop logging just because userspace wants to write attributes which do
not fit.
This and the previous commit fix xfstests generic/020.
Signed-off-by: Martin Brandenburg <martin@omnibond.com>
Cc: stable@vger.kernel.org
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
If the client receives a fatal server error from nfs_pageio_add_request(),
then we should always truncate the page on which the error occurred.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
EACCES, EDQUOT, EFBIG and ESTALE are all fatal errors as far as NFS
I/O is concerned. They need to be reported back to the application.
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Currently when there are buffered writes that were not yet flushed and
they fall within allocated ranges of the file (that is, not in holes or
beyond eof assuming there are no prealloc extents beyond eof), btrfs
simply reports an incorrect number of used blocks through the stat(2)
system call (or any of its variants), regardless of mount options or
inode flags (compress, compress-force, nodatacow). This is because the
number of blocks used that is reported is based on the current number
of bytes in the vfs inode plus the number of dealloc bytes in the btrfs
inode. The later covers bytes that both fall within allocated regions
of the file and holes.
Example scenarios where the number of reported blocks is wrong while the
buffered writes are not flushed:
$ mkfs.btrfs -f /dev/sdc
$ mount /dev/sdc /mnt/sdc
$ xfs_io -f -c "pwrite -S 0xaa 0 64K" /mnt/sdc/foo1
wrote 65536/65536 bytes at offset 0
64 KiB, 16 ops; 0.0000 sec (259.336 MiB/sec and 66390.0415 ops/sec)
$ sync
$ xfs_io -c "pwrite -S 0xbb 0 64K" /mnt/sdc/foo1
wrote 65536/65536 bytes at offset 0
64 KiB, 16 ops; 0.0000 sec (192.308 MiB/sec and 49230.7692 ops/sec)
# The following should have reported 64K...
$ du -h /mnt/sdc/foo1
128K /mnt/sdc/foo1
$ sync
# After flushing the buffered write, it now reports the correct value.
$ du -h /mnt/sdc/foo1
64K /mnt/sdc/foo1
$ xfs_io -f -c "falloc -k 0 128K" -c "pwrite -S 0xaa 0 64K" /mnt/sdc/foo2
wrote 65536/65536 bytes at offset 0
64 KiB, 16 ops; 0.0000 sec (520.833 MiB/sec and 133333.3333 ops/sec)
$ sync
$ xfs_io -c "pwrite -S 0xbb 64K 64K" /mnt/sdc/foo2
wrote 65536/65536 bytes at offset 65536
64 KiB, 16 ops; 0.0000 sec (260.417 MiB/sec and 66666.6667 ops/sec)
# The following should have reported 128K...
$ du -h /mnt/sdc/foo2
192K /mnt/sdc/foo2
$ sync
# After flushing the buffered write, it now reports the correct value.
$ du -h /mnt/sdc/foo2
128K /mnt/sdc/foo2
So the number of used file blocks is simply incorrect, unlike in other
filesystems such as ext4 and xfs for example, but only while the buffered
writes are not flushed.
Fix this by tracking the number of delalloc bytes that fall within holes
and beyond eof of a file, and use instead this new counter when reporting
the number of used blocks for an inode.
Another different problem that exists is that the delalloc bytes counter
is reset when writeback starts (by clearing the EXTENT_DEALLOC flag from
the respective range in the inode's iotree) and the vfs inode's bytes
counter is only incremented when writeback finishes (through
insert_reserved_file_extent()). Therefore while writeback is ongoing we
simply report a wrong number of blocks used by an inode if the write
operation covers a range previously unallocated. While this change does
not fix this problem, it does minimizes it a lot by shortening that time
window, as the new dealloc bytes counter (new_delalloc_bytes) is only
decremented when writeback finishes right before updating the vfs inode's
bytes counter. Fully fixing this second problem is not trivial and will
be addressed later by a different patch.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Normally we don't have inline extents followed by regular extents, but
there's currently at least one harmless case where this happens. For
example, when the page size is 4Kb and compression is enabled:
$ mkfs.btrfs -f /dev/sdb
$ mount -o compress /dev/sdb /mnt
$ xfs_io -f -c "pwrite -S 0xaa 0 4K" -c "fsync" /mnt/foobar
$ xfs_io -c "pwrite -S 0xbb 8K 4K" -c "fsync" /mnt/foobar
In this case we get a compressed inline extent, representing 4Kb of
data, followed by a hole extent and then a regular data extent. The
inline extent was not expanded/converted to a regular extent exactly
because it represents 4Kb of data. This does not cause any apparent
problem (such as the issue solved by commit e1699d2d7b
("btrfs: add missing memset while reading compressed inline extents"))
except trigger an unexpected case in the incremental send code path
that makes us issue an operation to write a hole when it's not needed,
resulting in more writes at the receiver and wasting space at the
receiver.
So teach the incremental send code to deal with this particular case.
The issue can be currently triggered by running fstests btrfs/137 with
compression enabled (MOUNT_OPTIONS="-o compress" ./check btrfs/137).
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
If the call to btrfs_qgroup_reserve_data() failed, we were leaking an
extent map structure. The failure can happen either due to an -ENOMEM
condition or, when quotas are enabled, due to -EDQUOT for example.
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
