When blkdev_open() races with device removal and creation it can happen
that unhashed bdev inode gets associated with newly created gendisk
like:
CPU0 CPU1
blkdev_open()
bdev = bd_acquire()
del_gendisk()
bdev_unhash_inode(bdev);
remove device
create new device with the same number
__blkdev_get()
disk = get_gendisk()
- gets reference to gendisk of the new device
Now another blkdev_open() will not find original 'bdev' as it got
unhashed, create a new one and associate it with the same 'disk' at
which point problems start as we have two independent page caches for
one device.
Fix the problem by verifying that the bdev inode didn't get unhashed
before we acquired gendisk reference. That way we make sure gendisk can
get associated only with visible bdev inodes.
Tested-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When two blkdev_open() calls for a partition race with device removal
and recreation, we can hit BUG_ON(!bd_may_claim(bdev, whole, holder)) in
blkdev_open(). The race can happen as follows:
CPU0 CPU1 CPU2
del_gendisk()
bdev_unhash_inode(part1);
blkdev_open(part1, O_EXCL) blkdev_open(part1, O_EXCL)
bdev = bd_acquire() bdev = bd_acquire()
blkdev_get(bdev)
bd_start_claiming(bdev)
- finds old inode 'whole'
bd_prepare_to_claim() -> 0
bdev_unhash_inode(whole);
<device removed>
<new device under same
number created>
blkdev_get(bdev);
bd_start_claiming(bdev)
- finds new inode 'whole'
bd_prepare_to_claim()
- this also succeeds as we have
different 'whole' here...
- bad things happen now as we
have two exclusive openers of
the same bdev
The problem here is that block device opens can see various intermediate
states while gendisk is shutting down and then being recreated.
We fix the problem by introducing new lookup_sem in gendisk that
synchronizes gendisk deletion with get_gendisk() and furthermore by
making sure that get_gendisk() does not return gendisk that is being (or
has been) deleted. This makes sure that once we ever manage to look up
newly created bdev inode, we are also guaranteed that following
get_gendisk() will either return failure (and we fail open) or it
returns gendisk for the new device and following bdget_disk() will
return new bdev inode (i.e., blkdev_open() follows the path as if it is
completely run after new device is created).
Reported-and-analyzed-by: Hou Tao <houtao1@huawei.com>
Tested-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When two blkdev_open() calls race with device removal and recreation,
__blkdev_get() can use looked up gendisk after it is freed:
CPU0 CPU1 CPU2
del_gendisk(disk);
bdev_unhash_inode(inode);
blkdev_open() blkdev_open()
bdev = bd_acquire(inode);
- creates and returns new inode
bdev = bd_acquire(inode);
- returns the same inode
__blkdev_get(devt) __blkdev_get(devt)
disk = get_gendisk(devt);
- got structure of device going away
<finish device removal>
<new device gets
created under the same
device number>
disk = get_gendisk(devt);
- got new device structure
if (!bdev->bd_openers) {
does the first open
}
if (!bdev->bd_openers)
- false
} else {
put_disk_and_module(disk)
- remember this was old device - this was last ref and disk is
now freed
}
disk_unblock_events(disk); -> oops
Fix the problem by making sure we drop reference to disk in
__blkdev_get() only after we are really done with it.
Reported-by: Hou Tao <houtao1@huawei.com>
Tested-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add a proper counterpart to get_disk_and_module() -
put_disk_and_module(). Currently it is opencoded in several places.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Rename get_disk() to get_disk_and_module() to make sure what the
function does. It's not a great name but at least it is now clear that
put_disk() is not it's counterpart.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit 8ddcd65325 "block: introduce GENHD_FL_HIDDEN" added handling of
hidden devices to get_gendisk() but forgot to drop module reference
which is also acquired by get_disk(). Drop the reference as necessary.
Arguably the function naming here is misleading as put_disk() is *not*
the counterpart of get_disk() but let's fix that in the follow up
commit since that will be more intrusive.
Fixes: 8ddcd65325
CC: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit e864f39569 "fs: add RWF_DSYNC aand RWF_SYNC" added additional
way for direct IO to become synchronous and thus trigger fsync from the
IO completion handler. Then commit 9830f4be15 "fs: Use RWF_* flags for
AIO operations" allowed these flags to be set for AIO as well. However
that commit forgot to update the condition checking whether the IO
completion handling should be defered to a workqueue and thus AIO DIO
with RWF_[D]SYNC set will call fsync() from IRQ context resulting in
sleep in atomic.
Fix the problem by checking directly iocb flags (the same way as it is
done in dio_complete()) instead of checking all conditions that could
lead to IO being synchronous.
CC: Christoph Hellwig <hch@lst.de>
CC: Goldwyn Rodrigues <rgoldwyn@suse.com>
CC: stable@vger.kernel.org
Reported-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Mark Rutland <mark.rutland@arm.com>
Fixes: 9830f4be15
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When allocation of underlying block for a page fault fails, we fail the
fault with SIGBUS. However we may well hit ENOSPC just due to lots of
free blocks being held by the running / committing transaction. So
propagate the error from ext4_iomap_begin() and implement do standard
allocation retry loop in ext4_dax_huge_fault().
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Ext4 needs to pass through error from its iomap handler to the page
fault handler so that it can properly detect ENOSPC and force
transaction commit and retry the fault (and block allocation). Add
argument to dax_iomap_fault() for passing such error.
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reportedly some UDF filesystems are recorded with bogus subsecond values
resulting in nanoseconds being over 10^9. Sanitize nanoseconds in time
stamps when loading them from disk.
Reported-by: Ian Turner <vectro@vectro.org>
Signed-off-by: Jan Kara <jack@suse.cz>
Currently we remove pages from the radix tree one by one. To speed up
page cache truncation, lock several pages at once and free them in one
go. This allows us to batch radix tree operations in a more efficient
way and also save round-trips on mapping->tree_lock. As a result we
gain about 20% speed improvement in page cache truncation.
Data from a simple benchmark timing 10000 truncates of 1024 pages (on
ext4 on ramdisk but the filesystem is barely visible in the profiles).
The range shows 1% and 95% percentiles of the measured times:
4.14-rc2 4.14-rc2 + batched truncation
248-256 209-219
249-258 209-217
248-255 211-239
248-255 209-217
247-256 210-218
[jack@suse.cz: convert delete_from_page_cache_batch() to pagevec]
Link: http://lkml.kernel.org/r/20171018111648.13714-1-jack@suse.cz
[akpm@linux-foundation.org: move struct pagevec forward declaration to top-of-file]
Link: http://lkml.kernel.org/r/20171010151937.26984-8-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Acked-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "Speed up page cache truncation", v1.
When rebasing our enterprise distro to a newer kernel (from 4.4 to 4.12)
we have noticed a regression in bonnie++ benchmark when deleting files.
Eventually we have tracked this down to a fact that page cache
truncation got slower by about 10%. There were both gains and losses in
the above interval of kernels but we have been able to identify that
commit 83929372f6 ("filemap: prepare find and delete operations for
huge pages") caused about 10% regression on its own.
After some investigation it didn't seem easily possible to fix the
regression while maintaining the THP in page cache functionality so
we've decided to optimize the page cache truncation path instead to make
up for the change. This series is a result of that effort.
Patch 1 is an easy speedup of cancel_dirty_page(). Patches 2-6 refactor
page cache truncation code so that it is easier to batch radix tree
operations. Patch 7 implements batching of deletes from the radix tree
which more than makes up for the original regression.
This patch (of 7):
cancel_dirty_page() does quite some work even for clean pages (fetching
of mapping, locking of memcg, atomic bit op on page flags) so it
accounts for ~2.5% of cost of truncation of a clean page. That is not
much but still dumb for something we don't need at all. Check whether a
page is actually dirty and avoid any work if not.
Link: http://lkml.kernel.org/r/20171010151937.26984-2-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Acked-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Andi Kleen <ak@linux.intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
xfs_filemap_pfn_mkwrite() duplicates a lot of __xfs_filemap_fault().
It will also need to handle flushing for synchronous page faults. So
just make that function use __xfs_filemap_fault().
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
We return IOMAP_F_DIRTY flag from ext4_iomap_begin() when asked to
prepare blocks for writing and the inode has some uncommitted metadata
changes. In the fault handler ext4_dax_fault() we then detect this case
(through VM_FAULT_NEEDDSYNC return value) and call helper
dax_finish_sync_fault() to flush metadata changes and insert page table
entry. Note that this will also dirty corresponding radix tree entry
which is what we want - fsync(2) will still provide data integrity
guarantees for applications not using userspace flushing. And
applications using userspace flushing can avoid calling fsync(2) and
thus avoid the performance overhead.
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
If transaction starting fails, just bail out of the function immediately
instead of checking for that condition throughout the function.
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Implement a function that filesystems can call to finish handling of
synchronous page faults. It takes care of syncing appropriare file range
and insertion of page table entry.
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Add a flag to iomap interface informing the caller that inode needs
fdstasync(2) for returned extent to become persistent and use it in DAX
fault code so that we don't map such extents into page tables
immediately. Instead we propagate the information that fdatasync(2) is
necessary from dax_iomap_fault() with a new VM_FAULT_NEEDDSYNC flag.
Filesystem fault handler is then responsible for calling fdatasync(2)
and inserting pfn into page tables.
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Define new MAP_SYNC flag and corresponding VMA VM_SYNC flag. As the
MAP_SYNC flag is not part of LEGACY_MAP_MASK, currently it will be
refused by all MAP_SHARED_VALIDATE map attempts and silently ignored for
everything else.
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Currently we dirty radix tree entry whenever dax_insert_mapping_entry()
gets called for a write fault. With synchronous page faults we would
like to insert clean radix tree entry and dirty it only once we call
fdatasync() and update page tables to save some unnecessary cache
flushing. Add 'dirty' argument to dax_insert_mapping_entry() for that.
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
For synchronous page fault dax_iomap_fault() will need to return PFN
which will then need to be inserted into page tables after fsync()
completes. Add necessary parameter to dax_iomap_fault().
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
dax_pmd_insert_mapping() has only one callsite and we will need to
further fine tune what it does for synchronous faults. Just inline it
into the callsite so that we don't have to pass awkward bools around.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
dax_insert_mapping() has only one callsite and we will need to further
fine tune what it does for synchronous faults. Just inline it into the
callsite so that we don't have to pass awkward bools around.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
dax_insert_mapping() has lots of arguments and a lot of them is actuall
duplicated by passing vm_fault structure as well. Change the function to
take the same arguments as dax_pmd_insert_mapping().
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
_calc_vm_trans() does not handle the situation when some of the passed
flags are 0 (which can happen if these VM flags do not make sense for
the architecture). Improve the _calc_vm_trans() macro to return 0 in
such situation. Since all passed flags are constant, this does not add
any runtime overhead.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
When fsnotify_add_mark_locked() fails it cleans up the mark it was
adding. Since the mark is already visible in group's list, we should
protect update of mark->flags with mark->lock. I'm not aware of any real
issues this could cause (since we also hold group->mark_mutex) but
better be safe and obey locking rules properly.
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Jan Kara <jack@suse.cz>
fsnotify_add_mark_locked() can fail but we do not check its return
value. This didn't matter before commit 9dd813c15b "fsnotify: Move
mark list head from object into dedicated structure" as none of possible
failures could happen for dnotify but after that commit -ENOMEM can be
returned. Handle this error properly in fcntl_dirnotify() as
otherwise we just hit BUG_ON(dn_mark->dn) in dnotify_free_mark().
Reviewed-by: Amir Goldstein <amir73il@gmail.com>
Reported-by: syzkaller
Fixes: 9dd813c15b
Signed-off-by: Jan Kara <jack@suse.cz>
Remove outdated references to maintainer and mailing list from the
documentation - reference MAINTAINERS instead. Also update reference to
current repository of udf-tools.
Signed-off-by: Jan Kara <jack@suse.cz>
When session starts beyond offset 2^31 the arithmetics in
udf_check_vsd() would overflow. Make sure the computation is done in
large enough type.
Reported-by: Cezary Sliwa <sliwa@ifpan.edu.pl>
Signed-off-by: Jan Kara <jack@suse.cz>
match_int() used in mount option parsing can allocate memory using
GFP_KERNEL and thus sleep. Avoid parsing mount options with sbi->s_lock
held.
Signed-off-by: Jan Kara <jack@suse.cz>
Instead of parsing mount options directly into the superblock (and
restoring options in case of error), parse the options into a dedicated
structure and only copy everything when we know we can safely switch
options. This will allow us to simplify locking and do option parsing
without holding sb->s_lock.
Signed-off-by: Jan Kara <jack@suse.cz>
Eryu has reported that since commit 7b9ca4c61b "quota: Reduce
contention on dq_data_lock" test generic/233 occasionally fails. This is
caused by the fact that since that commit we don't generate warning and
set grace time for quota allocations that have DQUOT_SPACE_NOFAIL set
(these are for example some metadata allocations in ext4). We need these
allocations to behave regularly wrt warning generation and grace time
setting so fix the code to return to the original behavior.
Reported-and-tested-by: Eryu Guan <eguan@redhat.com>
CC: stable@vger.kernel.org
Fixes: 7b9ca4c61b
Signed-off-by: Jan Kara <jack@suse.cz>
Eric has reported that since commit d2faa41516 "quota: Do not acquire
dqio_sem for dquot overwrites in v2 format" test generic/232
occasionally fails due to quota information being incorrect. Indeed that
commit was too eager to remove dqio_sem completely from the path that
just overwrites quota structure with updated information. Although that
is innocent on its own, another process that inserts new quota structure
to the same block can perform read-modify-write cycle of that block thus
effectively discarding quota information update if they race in a wrong
way.
Fix the problem by acquiring dqio_sem for reading for overwrites of
quota structure. Note that it *is* possible to completely avoid taking
dqio_sem in the overwrite path however that will require modifying path
inserting / deleting quota structures to avoid RMW cycles of the full
block and for now it is not clear whether it is worth the hassle.
Fixes: d2faa41516
Reported-and-tested-by: Eric Whitney <enwlinux@gmail.com>
Signed-off-by: Jan Kara <jack@suse.cz>
When new directory 'DIR1' is created in a directory 'DIR0' with SGID bit
set, DIR1 is expected to have SGID bit set (and owning group equal to
the owning group of 'DIR0'). However when 'DIR0' also has some default
ACLs that 'DIR1' inherits, setting these ACLs will result in SGID bit on
'DIR1' to get cleared if user is not member of the owning group.
Fix the problem by creating __orangefs_set_acl() function that does not
call posix_acl_update_mode() and use it when inheriting ACLs. That
prevents SGID bit clearing and the mode has been properly set by
posix_acl_create() anyway.
Fixes: 073931017b
CC: stable@vger.kernel.org
CC: Mike Marshall <hubcap@omnibond.com>
CC: pvfs2-developers@beowulf-underground.org
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mike Marshall <hubcap@omnibond.com>
We want only pages from given range in page_cache_seek_hole_data(). Use
pagevec_lookup_range() instead of pagevec_lookup() and remove
unnecessary code.
Note that the check for getting less pages than desired can be removed
because index gets updated by pagevec_lookup_range().
Link: http://lkml.kernel.org/r/20170726114704.7626-9-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Commit e64855c6cf ("fs: Add helper to clean bdev aliases under a bh
and use it") added a wrapper for clean_bdev_aliases() that invalidates
bdev aliases underlying a single buffer head.
However this has caused a performance regression for bonnie++ benchmark
on ext4 filesystem when delayed allocation is turned off (ext3 mode) -
average of 3 runs:
Hmean SeqOut Char 164787.55 ( 0.00%) 107189.06 (-34.95%)
Hmean SeqOut Block 219883.89 ( 0.00%) 168870.32 (-23.20%)
The reason for this regression is that clean_bdev_aliases() is slower
when called for a single block because pagevec_lookup() it uses will end
up iterating through the radix tree until it finds a page (which may
take a while) but we are only interested whether there's a page at a
particular index.
Fix the problem by using pagevec_lookup_range() instead which avoids the
needless iteration.
Fixes: e64855c6cf ("fs: Add helper to clean bdev aliases under a bh and use it")
Link: http://lkml.kernel.org/r/20170726114704.7626-5-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Cc: Jens Axboe <axboe@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Implement a variant of find_get_pages() that stops iterating at given
index. This may be substantial performance gain if the mapping is
sparse. See following commit for details. Furthermore lots of users of
this function (through pagevec_lookup()) actually want a range lookup
and all of them are currently open-coding this.
Also create corresponding pagevec_lookup_range() function.
Link: http://lkml.kernel.org/r/20170726114704.7626-4-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patch series "Ranged pagevec lookup", v2.
In this series I make pagevec_lookup() update the index (to be
consistent with pagevec_lookup_tag() and also as a preparation for
ranged lookups), provide ranged variant of pagevec_lookup() and use it
in places where it makes sense. This not only removes some common code
but is also a measurable performance win for some use cases (see patch
4/10) where radix tree is sparse and searching & grabing of a page after
the end of the range has measurable overhead.
This patch (of 10):
The callback doesn't ever get called. Remove it.
Link: http://lkml.kernel.org/r/20170726114704.7626-2-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If blkdev_get_by_path() in register_bcache() fails, we try to lookup the
block device using lookup_bdev() to detect which situation we are in to
properly report error. However we never drop the reference returned to
us from lookup_bdev(). Fix that.
Signed-off-by: Jan Kara <jack@suse.cz>
Acked-by: Coly Li <colyli@suse.de>
Cc: stable@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In recently_deleted() function we want to check whether inode is still
cached in buffer cache. Use sb_find_get_block() for that instead of
sb_getblk() to avoid unnecessary allocation of bdev page and buffer
heads.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
dq_data_lock is currently used to protect all modifications of quota
accounting information, consistency of quota accounting on the inode,
and dquot pointers from inode. As a result contention on the lock can be
pretty heavy.
Reduce the contention on the lock by protecting quota accounting
information by a new dquot->dq_dqb_lock and consistency of quota
accounting with inode usage by inode->i_lock.
This change reduces time to create 500000 files on ext4 on ramdisk by 50
different processes in separate directories by 6% when user quota is
turned on. When those 50 processes belong to 50 different users, the
improvement is about 9%.
Signed-off-by: Jan Kara <jack@suse.cz>
Provide helper __inode_get_bytes() which assumes i_lock is already
acquired. Quota code will need this to be able to use i_lock to protect
consistency of quota accounting information and inode usage.
Signed-off-by: Jan Kara <jack@suse.cz>
inode_incr_space() and inode_decr_space() have only two callsites.
Inline them there as that will make locking changes simpler.
Signed-off-by: Jan Kara <jack@suse.cz>
inode_add_rsv_space() and inode_sub_rsv_space() had only one callsite.
Inline them there directly. inode_claim_rsv_space() and
inode_reclaim_rsv_space() had two callsites so inline them there as
well. This will simplify further locking changes.
Signed-off-by: Jan Kara <jack@suse.cz>
When journalling quotas, we writeback all dquots immediately after
changing them as part of current transation. Thus there's no need to
write anything in dquot_writeback_dquots() and so we can avoid updating
list of dirty dquots to reduce dq_list_lock contention.
This change reduces time to create 500000 files on ext4 on ramdisk by 50
different processes in separate directories by 15% when user quota is
turned on.
Signed-off-by: Jan Kara <jack@suse.cz>
Filesystems that are journalling quotas generally don't need tracking of
dirty dquots in a list since forcing a transaction commit flushes all
quotas anyway. Allow filesystem to say it doesn't want dquots to be
tracked as it reduces contention on the dq_list_lock.
Signed-off-by: Jan Kara <jack@suse.cz>
Currently every dquot carries a wait_queue_head_t used only when we are
turning quotas off to wait for last users to drop dquot references.
Since such rare case is not performance sensitive in any means, just use
a global waitqueue for this and save space in struct dquot. Also convert
the logic to use wait_event() instead of open-coding it.
Signed-off-by: Jan Kara <jack@suse.cz>
Move locking of dq_list_lock into clear_dquot_dirty(). It makes the
function more self-contained and will simplify our life later.
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
Currently we mark dirty even dquots that are not active (i.e.,
initialization or reading failed for them). Thus later we have to check
whether dirty dquot is really active and just clear the dirty bit if
not. Avoid this complication by just never marking non-active dquot as
dirty.
Signed-off-by: Jan Kara <jack@suse.cz>
dqi_flags modifications are protected by dq_data_lock. However the
modifications in vfs_load_quota_inode() and in mark_info_dirty() were
not which could lead to corruption of dqi_flags. Since modifications to
dqi_flags are rare, this is hard to observe in practice but in theory it
could happen. Fix the problem by always using dq_data_lock for
protection.
Signed-off-by: Jan Kara <jack@suse.cz>
Currently we return -EIO on any error (or short read) from
->quota_read() while reading quota info. Propagate the error code
instead.
Suggested-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
v2_read_file_info() returned -1 instead of proper error codes on error.
Luckily this is not easily visible from userspace as we have called
->check_quota_file shortly before and thus already verified the quota
file is sane. Still set the error codes to proper values.
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
Push down acquisition of dqio_sem into ->read_file_info() callback. This
is for consistency with other operations and it also allows us to get
rid of an ugliness in OCFS2.
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
Push down acquisition of dqio_sem into ->write_file_info() callback.
Mostly for consistency with other operations.
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
Push down acquisition of dqio_sem into ->get_next_id() callback. Mostly
for consistency with other operations.
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
Push down acquisition of dqio_sem into ->release_dqblk() callback. It
will allow quota formats to decide whether they need it or not.
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
The old quota quota format has fixed offset in quota file based on ID so
there's no locking needed against concurrent modifications of the file
(locking against concurrent IO on the same dquot is still provided by
dq_lock).
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
When dquot has space already allocated in a quota file, we just
overwrite that place when writing dquot. So we don't need any protection
against other modifications of quota file as these keep dquot in place.
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
Push down acquisition of dqio_sem into ->write_dqblk() callback. It will
allow quota formats to decide whether they need it or not.
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
The old quota format has fixed offset in quota file based on ID so
there's no locking needed against concurrent modifications of the file
(locking against concurrent IO on the same dquot is still provided by
dq_lock).
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
Push down acquisition of dqio_sem into ->read_dqblk() callback. It will
allow quota formats to decide whether they need it or not.
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
Currently dquot writeout is only protected by dqio_sem held for writing.
As we transition to a finer grained locking we will use dquot->dq_lock
instead. So acquire it in dquot_commit() and move dqio_sem just around
->commit_dqblk() call as it is still needed to serialize quota file
changes.
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
vfs_load_quota_inode() needs dqio_sem only for reading. In fact dqio_sem
is not needed there at all since the function can be called only during
quota on when quota file cannot be modified but let's leave the
protection there since it is logical and the path is in no way
performance critical.
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
dquot_get_next_id() needs dqio_sem only for reading to protect against
racing with modification of quota file structure.
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Jan Kara <jack@suse.cz>
Jan Kara