Hanjun Guo has reported that a CMA stress test causes broken accounting of
CMA and free pages:
> Before the test, I got:
> -bash-4.3# cat /proc/meminfo | grep Cma
> CmaTotal: 204800 kB
> CmaFree: 195044 kB
>
>
> After running the test:
> -bash-4.3# cat /proc/meminfo | grep Cma
> CmaTotal: 204800 kB
> CmaFree: 6602584 kB
>
> So the freed CMA memory is more than total..
>
> Also the the MemFree is more than mem total:
>
> -bash-4.3# cat /proc/meminfo
> MemTotal: 16342016 kB
> MemFree: 22367268 kB
> MemAvailable: 22370528 kB
Laura Abbott has confirmed the issue and suspected the freepage accounting
rewrite around 3.18/4.0 by Joonsoo Kim. Joonsoo had a theory that this is
caused by unexpected merging between MIGRATE_ISOLATE and MIGRATE_CMA
pageblocks:
> CMA isolates MAX_ORDER aligned blocks, but, during the process,
> partialy isolated block exists. If MAX_ORDER is 11 and
> pageblock_order is 9, two pageblocks make up MAX_ORDER
> aligned block and I can think following scenario because pageblock
> (un)isolation would be done one by one.
>
> (each character means one pageblock. 'C', 'I' means MIGRATE_CMA,
> MIGRATE_ISOLATE, respectively.
>
> CC -> IC -> II (Isolation)
> II -> CI -> CC (Un-isolation)
>
> If some pages are freed at this intermediate state such as IC or CI,
> that page could be merged to the other page that is resident on
> different type of pageblock and it will cause wrong freepage count.
This was supposed to be prevented by CMA operating on MAX_ORDER blocks,
but since it doesn't hold the zone->lock between pageblocks, a race
window does exist.
It's also likely that unexpected merging can occur between
MIGRATE_ISOLATE and non-CMA pageblocks. This should be prevented in
__free_one_page() since commit 3c605096d3 ("mm/page_alloc: restrict
max order of merging on isolated pageblock"). However, we only check
the migratetype of the pageblock where buddy merging has been initiated,
not the migratetype of the buddy pageblock (or group of pageblocks)
which can be MIGRATE_ISOLATE.
Joonsoo has suggested checking for buddy migratetype as part of
page_is_buddy(), but that would add extra checks in allocator hotpath
and bloat-o-meter has shown significant code bloat (the function is
inline).
This patch reduces the bloat at some expense of more complicated code.
The buddy-merging while-loop in __free_one_page() is initially bounded
to pageblock_border and without any migratetype checks. The checks are
placed outside, bumping the max_order if merging is allowed, and
returning to the while-loop with a statement which can't be possibly
considered harmful.
This fixes the accounting bug and also removes the arguably weird state
in the original commit 3c605096d3 where buddies could be left
unmerged.
Fixes: 3c605096d3 ("mm/page_alloc: restrict max order of merging on isolated pageblock")
Link: https://lkml.org/lkml/2016/3/2/280
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Reported-by: Hanjun Guo <guohanjun@huawei.com>
Tested-by: Hanjun Guo <guohanjun@huawei.com>
Acked-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Debugged-by: Laura Abbott <labbott@redhat.com>
Debugged-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: <stable@vger.kernel.org> [3.18+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The r592 driver relies on behavior of the DMA mapping API that is
normally observed but not guaranteed by the API. Instead it uses a
runtime check to fail transfers if the API ever behaves
When CONFIG_NEED_SG_DMA_LENGTH is not set, one of the checks turns into a
comparison of a variable with itself, which gcc-6.0 now warns about:
drivers/memstick/host/r592.c: In function 'r592_transfer_fifo_dma':
drivers/memstick/host/r592.c:302:31: error: self-comparison always evaluates to false [-Werror=tautological-compare]
(sg_dma_len(&dev->req->sg) < dev->req->sg.length)) {
^
The check itself is not a problem, so this patch just rephrases the
condition in a way that gcc does not consider an indication of a mistake.
We already know that dev->req->sg.length was initially R592_LFIFO_SIZE, so
we can compare it to that constant again.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Maxim Levitsky <maximlevitsky@gmail.com>
Cc: Quentin Lambert <lambert.quentin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Now function ocfs2_replay_truncate_records() first modifies tl_used,
then calls ocfs2_extend_trans() to extend transactions for gd and alloc
inode used for freeing clusters. jbd2_journal_restart() may be called
and it may happen that tl_used in truncate log is decreased but the
clusters are not freed, which means these clusters are lost. So we
should avoid extending transactions in these two operations.
Signed-off-by: joyce.xue <xuejiufei@huawei.com>
Reviewed-by: Mark Fasheh <mfasheh@suse.de>
Acked-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
I found that jbd2_journal_restart() is called in some places without
keeping things consistently before. However, jbd2_journal_restart() may
commit the handle's transaction and restart another one. If the first
transaction is committed successfully while another not, it may cause
filesystem inconsistency or read only. This is an effort to fix this
kind of problems.
This patch (of 3):
The following functions will be called while truncating an extent:
ocfs2_remove_btree_range
-> ocfs2_start_trans
-> ocfs2_remove_extent
-> ocfs2_truncate_rec
-> ocfs2_extend_rotate_transaction
-> jbd2_journal_restart if jbd2_journal_extend fail
-> ocfs2_rotate_tree_left
-> ocfs2_remove_rightmost_path
-> ocfs2_extend_rotate_transaction
-> ocfs2_unlink_subtree
-> ocfs2_update_edge_lengths
-> ocfs2_extend_trans
-> jbd2_journal_restart if jbd2_journal_extend fail
-> ocfs2_et_update_clusters
-> ocfs2_commit_trans
jbd2_journal_restart() may be called and it may happened that the buffers
dirtied in ocfs2_truncate_rec() are committed while buffers dirtied in
ocfs2_et_update_clusters() are not, the total clusters on extent tree and
i_clusters in ocfs2_dinode is inconsistency. So the clusters got from
ocfs2_dinode is incorrect, and it also cause read-only problem when call
ocfs2_commit_truncate() with the error message: "Inode %llu has empty
extent block at %llu".
We should extend enough credits for function ocfs2_remove_rightmost_path
and ocfs2_update_edge_lengths to avoid this inconsistency.
Signed-off-by: joyce.xue <xuejiufei@huawei.com>
Acked-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We have found a bug when two nodes doing umount one after another.
1) Node 1 migrate a lockres that has 3 locks in grant queue such as
N2(PR)<->N3(NL)<->N4(PR) to N2. After migration, lvb of the lock
N3(NL) and N4(PR) are empty on node 2 because migration target do not
copy lvb to these two lock.
2) Node 3 want to convert to PR, it can be granted in
__dlmconvert_master(), and the order of these locks is unchanged. The
lvb of the lock N3(PR) on node 2 is copyed from lockres in function
dlm_update_lvb() while the lvb of lock N4(PR) is still empty.
3) Node 2 want to leave domain, it will migrate this lockres to node 3.
Then node 2 will trigger the BUG in dlm_prepare_lvb_for_migration()
when adding the lock N4(PR) to mres with the following message because
the lvb of mres is already copied from lock N3(PR), but the lvb of lock
N4(PR) is empty.
"Mismatched lvb in lock cookie=%u:%llu, name=%.*s, node=%u"
[akpm@linux-foundation.org: tweak comment]
Signed-off-by: xuejiufei <xuejiufei@huawei.com>
Acked-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In update_backups() there exists a problem of crossing the boundary as
follows:
we assume that lun will be resized to 1TB(cluster_size is 32kb), it will
include 0~33554431 cluster, in update_backups func, it will backup super
block in location of 1TB which is the 33554432th cluster, so the
phenomenon of crossing the boundary happens.
Signed-off-by: Yiwen Jiang <jiangyiwen@huawei.com>
Reviewed-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Xue jiufei <xuejiufei@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch fixes a deadlock, as follows:
Node 1 Node 2 Node 3
1)volume a and b are only mount vol a only mount vol b
mounted
2) start to mount b start to mount a
3) check hb of Node 3 check hb of Node 2
in vol a, qs_holds++ in vol b, qs_holds++
4) -------------------- all nodes' network down --------------------
5) progress of mount b the same situation as
failed, and then call Node 2
ocfs2_dismount_volume.
but the process is hung,
since there is a work
in ocfs2_wq cannot beo
completed. This work is
about vol a, because
ocfs2_wq is global wq.
BTW, this work which is
scheduled in ocfs2_wq is
ocfs2_orphan_scan_work,
and the context in this work
needs to take inode lock
of orphan_dir, because
lockres owner are Node 1 and
all nodes' nework has been down
at the same time, so it can't
get the inode lock.
6) Why can't this node be fenced
when network disconnected?
Because the process of
mount is hung what caused qs_holds
is not equal 0.
Because all works in the ocfs2_wq are relative to the super block.
The solution is to change the ocfs2_wq from global to local. In other
words, move it into struct ocfs2_super.
Signed-off-by: Yiwen Jiang <jiangyiwen@huawei.com>
Reviewed-by: Joseph Qi <joseph.qi@huawei.com>
Cc: Xue jiufei <xuejiufei@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Cc: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When master handles convert request, it queues ast first and then
returns status. This may happen that the ast is sent before the request
status because the above two messages are sent by two threads. And
right after the ast is sent, if master down, it may trigger BUG in
dlm_move_lockres_to_recovery_list in the requested node because ast
handler moves it to grant list without clear lock->convert_pending. So
remove BUG_ON statement and check if the ast is processed in
dlmconvert_remote.
Signed-off-by: Joseph Qi <joseph.qi@huawei.com>
Reported-by: Yiwen Jiang <jiangyiwen@huawei.com>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Tariq Saeed <tariq.x.saeed@oracle.com>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There is a race window between dlmconvert_remote and
dlm_move_lockres_to_recovery_list, which will cause a lock with
OCFS2_LOCK_BUSY in grant list, thus system hangs.
dlmconvert_remote
{
spin_lock(&res->spinlock);
list_move_tail(&lock->list, &res->converting);
lock->convert_pending = 1;
spin_unlock(&res->spinlock);
status = dlm_send_remote_convert_request();
>>>>>> race window, master has queued ast and return DLM_NORMAL,
and then down before sending ast.
this node detects master down and calls
dlm_move_lockres_to_recovery_list, which will revert the
lock to grant list.
Then OCFS2_LOCK_BUSY won't be cleared as new master won't
send ast any more because it thinks already be authorized.
spin_lock(&res->spinlock);
lock->convert_pending = 0;
if (status != DLM_NORMAL)
dlm_revert_pending_convert(res, lock);
spin_unlock(&res->spinlock);
}
In this case, check if res->state has DLM_LOCK_RES_RECOVERING bit set
(res is still in recovering) or res master changed (new master has
finished recovery), reset the status to DLM_RECOVERING, then it will
retry convert.
Signed-off-by: Joseph Qi <joseph.qi@huawei.com>
Reported-by: Yiwen Jiang <jiangyiwen@huawei.com>
Reviewed-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Tariq Saeed <tariq.x.saeed@oracle.com>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The code should call ocfs2_free_alloc_context() to free meta_ac &
data_ac before calling ocfs2_run_deallocs(). Because
ocfs2_run_deallocs() will acquire the system inode's i_mutex hold by
meta_ac. So try to release the lock before ocfs2_run_deallocs().
Fixes: af1310367f41 ("ocfs2: fix sparse file & data ordering issue in direct io.")
Signed-off-by: Ryan Ding <ryan.ding@oracle.com>
Acked-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When doing append direct write in an already allocated cluster, and fast
path in ocfs2_dio_get_block() is triggered, function
ocfs2_dio_end_io_write() will be skipped as there is no context
allocated.
As a result, the disk file size will not be changed as it should be.
The solution is to skip fast path when we are about to change file size.
Fixes: af1310367f41 ("ocfs2: fix sparse file & data ordering issue in direct io.")
Signed-off-by: Ryan Ding <ryan.ding@oracle.com>
Acked-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Take ip_alloc_sem to prevent concurrent access to extent tree, which may
cause the extent tree in an unstable state.
Signed-off-by: Ryan Ding <ryan.ding@oracle.com>
Reviewed-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In the current implementation of unaligned aio+dio, lock order behave as
follow:
in user process context:
-> call io_submit()
-> get i_mutex
<== window1
-> get ip_unaligned_aio
-> submit direct io to block device
-> release i_mutex
-> io_submit() return
in dio work queue context(the work queue is created in __blockdev_direct_IO):
-> release ip_unaligned_aio
<== window2
-> get i_mutex
-> clear unwritten flag & change i_size
-> release i_mutex
There is a limitation to the thread number of dio work queue. 256 at
default. If all 256 thread are in the above 'window2' stage, and there
is a user process in the 'window1' stage, the system will became
deadlock. Since the user process hold i_mutex to wait ip_unaligned_aio
lock, while there is a direct bio hold ip_unaligned_aio mutex who is
waiting for a dio work queue thread to be schedule. But all the dio
work queue thread is waiting for i_mutex lock in 'window2'.
This case only happened in a test which send a large number(more than
256) of aio at one io_submit() call.
My design is to remove ip_unaligned_aio lock. Change it to a sync io
instead. Just like ip_unaligned_aio lock, serialize the unaligned aio
dio.
[akpm@linux-foundation.org: remove OCFS2_IOCB_UNALIGNED_IO, per Junxiao Bi]
Signed-off-by: Ryan Ding <ryan.ding@oracle.com>
Reviewed-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Clean up ocfs2_file_write_iter & ocfs2_prepare_inode_for_write:
* remove append dio check: it will be checked in ocfs2_direct_IO()
* remove file hole check: file hole is supported for now
* remove inline data check: it will be checked in ocfs2_direct_IO()
* remove the full_coherence check when append dio: we will get the
inode_lock in ocfs2_dio_get_block, there is no need to fall back to
buffer io to ensure the coherence semantics.
Now the drop dio procedure is gone. :)
[akpm@linux-foundation.org: remove unused label]
Signed-off-by: Ryan Ding <ryan.ding@oracle.com>
Reviewed-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There are mainly three issues in the direct io code path after commit
24c40b329e ("ocfs2: implement ocfs2_direct_IO_write"):
* Does not support sparse file.
* Does not support data ordering. eg: when write to a file hole, it
will alloc extent first. If system crashed before io finished, data
will corrupt.
* Potential risk when doing aio+dio. The -EIOCBQUEUED return value is
likely to be ignored by ocfs2_direct_IO_write().
To resolve above problems, re-design direct io code with following ideas:
* Use buffer io to fill in holes. And this will make better
performance also.
* Clear unwritten after direct write finished. So we can make sure
meta data changes after data write to disk. (Unwritten extent is
invisible to user, from user's view, meta data is not changed when
allocate an unwritten extent.)
* Clear ocfs2_direct_IO_write(). Do all ending work in end_io.
This patch has passed fs,dio,ltp-aiodio.part1,ltp-aiodio.part2,ltp-aiodio.part4
test cases of ltp.
For performance improvement, see following test result:
ocfs2 cluster size 1MB, ocfs2 volume is mounted on /mnt/.
The original way:
+ rm /mnt/test.img -f
+ dd if=/dev/zero of=/mnt/test.img bs=4K count=1048576 oflag=direct
1048576+0 records in
1048576+0 records out
4294967296 bytes (4.3 GB) copied, 1707.83 s, 2.5 MB/s
+ rm /mnt/test.img -f
+ dd if=/dev/zero of=/mnt/test.img bs=256K count=16384 oflag=direct
16384+0 records in
16384+0 records out
4294967296 bytes (4.3 GB) copied, 582.705 s, 7.4 MB/s
After this patch:
+ rm /mnt/test.img -f
+ dd if=/dev/zero of=/mnt/test.img bs=4K count=1048576 oflag=direct
1048576+0 records in
1048576+0 records out
4294967296 bytes (4.3 GB) copied, 64.6412 s, 66.4 MB/s
+ rm /mnt/test.img -f
+ dd if=/dev/zero of=/mnt/test.img bs=256K count=16384 oflag=direct
16384+0 records in
16384+0 records out
4294967296 bytes (4.3 GB) copied, 34.7611 s, 124 MB/s
Signed-off-by: Ryan Ding <ryan.ding@oracle.com>
Reviewed-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
To support direct io in ocfs2_write_begin_nolock & ocfs2_write_end_nolock.
There is still one issue in the direct write procedure.
phase 1: alloc extent with UNWRITTEN flag
phase 2: submit direct data to disk, add zero page to page cache
phase 3: clear UNWRITTEN flag when data has been written to disk
When there are 2 direct write A(0~3KB),B(4~7KB) writing to the same
cluster 0~7KB (cluster size 8KB). Write request A arrive phase 2 first,
it will zero the region (4~7KB). Before request A enter to phase 3,
request B arrive phase 2, it will zero region (0~3KB). This is just like
request B steps request A.
To resolve this issue, we should let request B knows this cluster is already
under zero, to prevent it from steps the previous write request.
This patch will add function ocfs2_unwritten_check() to do this job. It
will record all clusters that are under direct write(it will be recorded
in the 'ip_unwritten_list' member of inode info), and prevent the later
direct write writing to the same cluster to do the zero work again.
Signed-off-by: Ryan Ding <ryan.ding@oracle.com>
Reviewed-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
To support direct io in ocfs2_write_begin_nolock & ocfs2_write_end_nolock.
Direct io needs to get the physical address from write_begin, to map the
user page. This patch is to change the arg 'phys' of
ocfs2_write_cluster to a pointer, so it can be retrieved to write_begin.
And we can retrieve it to the direct io procedure.
Signed-off-by: Ryan Ding <ryan.ding@oracle.com>
Reviewed-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
To support direct io in ocfs2_write_begin_nolock & ocfs2_write_end_nolock.
Append direct io do not change i_size in get block phase. It only move
to orphan when starting write. After data is written to disk, it will
delete itself from orphan and update i_size. So skip i_size change
section in write_begin for direct io.
And when there is no extents alloc, no meta data changes needed for
direct io (since write_begin start trans for 2 reason: alloc extents &
change i_size. Now none of them needed). So we can skip start trans
procedure.
Signed-off-by: Ryan Ding <ryan.ding@oracle.com>
Reviewed-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
To support direct io in ocfs2_write_begin_nolock & ocfs2_write_end_nolock.
Direct io data will not appear in buffer. The w_target_page member will
not be filled by direct io. So avoid to use it when it's NULL. Unlinke
buffer io and mmap, direct io will call write_begin with more than 1
page a time. So the target_index is not sufficient to describe the
actual data. change it to a range start at target_index, end in
end_index.
Signed-off-by: Ryan Ding <ryan.ding@oracle.com>
Reviewed-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
To support direct io in ocfs2_write_begin_nolock & ocfs2_write_end_nolock.
There is a problem in ocfs2's direct io implement: if system crashed
after extents allocated, and before data return, we will get a extent
with dirty data on disk. This problem violate the journal=order
semantics, which means meta changes take effect after data written to
disk. To resolve this issue, direct write can use the UNWRITTEN flag to
describe a extent during direct data writeback. The direct write
procedure should act in the following order:
phase 1: alloc extent with UNWRITTEN flag
phase 2: submit direct data to disk, add zero page to page cache
phase 3: clear UNWRITTEN flag when data has been written to disk
This patch is to change the 'c_unwritten' member of
ocfs2_write_cluster_desc to 'c_clear_unwritten'. Means whether to clear
the unwritten flag. It do not care if a extent is allocated or not.
And use 'c_new' to specify a newly allocated extent. So the direct io
procedure can use c_clear_unwritten to control the UNWRITTEN bit on
extent.
Signed-off-by: Ryan Ding <ryan.ding@oracle.com>
Reviewed-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Patchset: fix ocfs2 direct io code patch to support sparse file and data
ordering semantics
The idea is to use buffer io(more precisely use the interface
ocfs2_write_begin_nolock & ocfs2_write_end_nolock) to do the zero work
beyond block size. And clear UNWRITTEN flag until direct io data has
been written to disk, which can prevent data corruption when system
crashed during direct write.
And we will also archive a better performance: eg. dd direct write new
file with block size 4KB: before this patchset:
2.5 MB/s
after this patchset:
66.4 MB/s
This patch (of 8):
To support direct io in ocfs2_write_begin_nolock &
ocfs2_write_end_nolock.
Remove unused args filp & flags. Add new arg type. The type is one of
buffer/direct/mmap. Indicate 3 way to perform write. buffer/mmap type
has implemented. direct type will be implemented later.
Signed-off-by: Ryan Ding <ryan.ding@oracle.com>
Reviewed-by: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <joseph.qi@huawei.com>
Cc: Mark Fasheh <mfasheh@suse.de>
Cc: Joel Becker <jlbec@evilplan.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
INPUT_COMPAT_TEST became much simpler after commit f4056b5284
("input: redefine INPUT_COMPAT_TEST as in_compat_syscall()") so we can
cleanly eliminate it altogether.
Acked-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
"oom, oom_reaper: disable oom_reaper for oom_kill_allocating_task" tried
to protect oom_reaper_list using MMF_OOM_KILLED flag. But we can do it
by simply checking tsk->oom_reaper_list != NULL.
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
After "oom: clear TIF_MEMDIE after oom_reaper managed to unmap the
address space" oom_reaper will call exit_oom_victim on the target task
after it is done. This might however race with the PM freezer:
CPU0 CPU1 CPU2
freeze_processes
try_to_freeze_tasks
# Allocation request
out_of_memory
oom_killer_disable
wake_oom_reaper(P1)
__oom_reap_task
exit_oom_victim(P1)
wait_event(oom_victims==0)
[...]
do_exit(P1)
perform IO/interfere with the freezer
which breaks the oom_killer_disable semantic. We no longer have a
guarantee that the oom victim won't interfere with the freezer because
it might be anywhere on the way to do_exit while the freezer thinks the
task has already terminated. It might trigger IO or touch devices which
are frozen already.
In order to close this race, make the oom_reaper thread freezable. This
will work because
a) already running oom_reaper will block freezer to enter the
quiescent state
b) wake_oom_reaper will not wake up the reaper after it has been
frozen
c) the only way to call exit_oom_victim after try_to_freeze_tasks
is from the oom victim's context when we know the further
interference shouldn't be possible
Signed-off-by: Michal Hocko <mhocko@suse.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Entries are only added/removed from oom_reaper_list at head so we can
use a single linked list and hence save a word in task_struct.
Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Signed-off-by: Michal Hocko <mhocko@suse.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Tetsuo has reported that oom_kill_allocating_task=1 will cause
oom_reaper_list corruption because oom_kill_process doesn't follow
standard OOM exclusion (aka ignores TIF_MEMDIE) and allows to enqueue
the same task multiple times - e.g. by sacrificing the same child
multiple times.
This patch fixes the issue by introducing a new MMF_OOM_KILLED mm flag
which is set in oom_kill_process atomically and oom reaper is disabled
if the flag was already set.
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
wake_oom_reaper has allowed only 1 oom victim to be queued. The main
reason for that was the simplicity as other solutions would require some
way of queuing. The current approach is racy and that was deemed
sufficient as the oom_reaper is considered a best effort approach to
help with oom handling when the OOM victim cannot terminate in a
reasonable time. The race could lead to missing an oom victim which can
get stuck
out_of_memory
wake_oom_reaper
cmpxchg // OK
oom_reaper
oom_reap_task
__oom_reap_task
oom_victim terminates
atomic_inc_not_zero // fail
out_of_memory
wake_oom_reaper
cmpxchg // fails
task_to_reap = NULL
This race requires 2 OOM invocations in a short time period which is not
very likely but certainly not impossible. E.g. the original victim
might have not released a lot of memory for some reason.
The situation would improve considerably if wake_oom_reaper used a more
robust queuing. This is what this patch implements. This means adding
oom_reaper_list list_head into task_struct (eat a hole before embeded
thread_struct for that purpose) and a oom_reaper_lock spinlock for
queuing synchronization. wake_oom_reaper will then add the task on the
queue and oom_reaper will dequeue it.
Signed-off-by: Michal Hocko <mhocko@suse.com>
Cc: Vladimir Davydov <vdavydov@virtuozzo.com>
Cc: Andrea Argangeli <andrea@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Inform about the successful/failed oom_reaper attempts and dump all the
held locks to tell us more who is blocking the progress.
[akpm@linux-foundation.org: fix CONFIG_MMU=n build]
Signed-off-by: Michal Hocko <mhocko@suse.com>
Cc: Andrea Argangeli <andrea@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When oom_reaper manages to unmap all the eligible vmas there shouldn't
be much of the freable memory held by the oom victim left anymore so it
makes sense to clear the TIF_MEMDIE flag for the victim and allow the
OOM killer to select another task.
The lack of TIF_MEMDIE also means that the victim cannot access memory
reserves anymore but that shouldn't be a problem because it would get
the access again if it needs to allocate and hits the OOM killer again
due to the fatal_signal_pending resp. PF_EXITING check. We can safely
hide the task from the OOM killer because it is clearly not a good
candidate anymore as everyhing reclaimable has been torn down already.
This patch will allow to cap the time an OOM victim can keep TIF_MEMDIE
and thus hold off further global OOM killer actions granted the oom
reaper is able to take mmap_sem for the associated mm struct. This is
not guaranteed now but further steps should make sure that mmap_sem for
write should be blocked killable which will help to reduce such a lock
contention. This is not done by this patch.
Note that exit_oom_victim might be called on a remote task from
__oom_reap_task now so we have to check and clear the flag atomically
otherwise we might race and underflow oom_victims or wake up waiters too
early.
Signed-off-by: Michal Hocko <mhocko@suse.com>
Suggested-by: Johannes Weiner <hannes@cmpxchg.org>
Suggested-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Andrea Argangeli <andrea@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch (of 5):
This is based on the idea from Mel Gorman discussed during LSFMM 2015
and independently brought up by Oleg Nesterov.
The OOM killer currently allows to kill only a single task in a good
hope that the task will terminate in a reasonable time and frees up its
memory. Such a task (oom victim) will get an access to memory reserves
via mark_oom_victim to allow a forward progress should there be a need
for additional memory during exit path.
It has been shown (e.g. by Tetsuo Handa) that it is not that hard to
construct workloads which break the core assumption mentioned above and
the OOM victim might take unbounded amount of time to exit because it
might be blocked in the uninterruptible state waiting for an event (e.g.
lock) which is blocked by another task looping in the page allocator.
This patch reduces the probability of such a lockup by introducing a
specialized kernel thread (oom_reaper) which tries to reclaim additional
memory by preemptively reaping the anonymous or swapped out memory owned
by the oom victim under an assumption that such a memory won't be needed
when its owner is killed and kicked from the userspace anyway. There is
one notable exception to this, though, if the OOM victim was in the
process of coredumping the result would be incomplete. This is
considered a reasonable constrain because the overall system health is
more important than debugability of a particular application.
A kernel thread has been chosen because we need a reliable way of
invocation so workqueue context is not appropriate because all the
workers might be busy (e.g. allocating memory). Kswapd which sounds
like another good fit is not appropriate as well because it might get
blocked on locks during reclaim as well.
oom_reaper has to take mmap_sem on the target task for reading so the
solution is not 100% because the semaphore might be held or blocked for
write but the probability is reduced considerably wrt. basically any
lock blocking forward progress as described above. In order to prevent
from blocking on the lock without any forward progress we are using only
a trylock and retry 10 times with a short sleep in between. Users of
mmap_sem which need it for write should be carefully reviewed to use
_killable waiting as much as possible and reduce allocations requests
done with the lock held to absolute minimum to reduce the risk even
further.
The API between oom killer and oom reaper is quite trivial.
wake_oom_reaper updates mm_to_reap with cmpxchg to guarantee only
NULL->mm transition and oom_reaper clear this atomically once it is done
with the work. This means that only a single mm_struct can be reaped at
the time. As the operation is potentially disruptive we are trying to
limit it to the ncessary minimum and the reaper blocks any updates while
it operates on an mm. mm_struct is pinned by mm_count to allow parallel
exit_mmap and a race is detected by atomic_inc_not_zero(mm_users).
Signed-off-by: Michal Hocko <mhocko@suse.com>
Suggested-by: Oleg Nesterov <oleg@redhat.com>
Suggested-by: Mel Gorman <mgorman@suse.de>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Andrea Argangeli <andrea@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This will be needed in the patch "mm, oom: introduce oom reaper".
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The permissions of this file were modified by commit (f447671b9e PM /
AVS: rockchip-io: add io selectors and supplies for rk3399) by mistake,
so fix them.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Use KMEM_CACHE() instead of kmem_cache_create() to simplify the code.
Signed-off-by: Geliang Tang <geliangtang@163.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Use kmem_cache_zalloc() instead of kmem_cache_alloc() with flag GFP_ZERO.
Signed-off-by: Geliang Tang <geliangtang@163.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Use KMEM_CACHE() instead of kmem_cache_create() to simplify the code.
Signed-off-by: Geliang Tang <geliangtang@163.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
If dentry has no lease, ceph_d_revalidate() previously return 0.
This causes VFS to invalidate the dentry and create a new dentry
for later lookup. Invalidating a dentry also detach any underneath
mount points. So mount point inside cephfs can disapear mystically
(even the mount point is not modified by other hosts).
The fix is using lookup request to revalidate dentry without lease.
This can partly solve the mount points disapear issue (as long as
the mount point is not modified by other hosts)
Signed-off-by: Yan, Zheng <zyan@redhat.com>
When security is enabled, security module can call filesystem's
getxattr/setxattr callbacks during d_instantiate(). For cephfs,
d_instantiate() is usually called by MDS' dispatch thread, while
handling MDS reply. If the MDS reply does not include xattrs and
corresponding caps, getxattr/setxattr need to send a new request
to MDS and waits for the reply. This makes MDS' dispatch sleep,
nobody handles later MDS replies.
The fix is make sure lookup/atomic_open reply include xattrs and
corresponding caps. So getxattr can be handled by cached xattrs.
This requires some modification to both MDS and request message.
(Client tells MDS what caps it wants; MDS encodes proper caps in
the reply)
Smack security module may call setxattr during d_instantiate().
Unlike getxattr, we can't force MDS to issue CEPH_CAP_XATTR_EXCL
to us. So just make setxattr return error when called by MDS'
dispatch thread.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
If page->mapping is NULL, releasepage() callback does not get called.
Remove the unnecessary NULL check to make static code analysis tool
happy
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Readdir cache uses page cache to save dentry pointers. When adding
dentry pointers to middle of a page, we need to make sure the page
already exists. Otherwise the beginning part of the page will be
invalid pointers.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Don't open-code sizeof_footer() in read_partial_message() and
ceph_msg_revoke(). Also, after switching to sizeof_footer(), it's now
possible to use con_out_kvec_add() in prepare_write_message_footer().
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Alex Elder <elder@linaro.org>
ceph_empty_snapc->num_snaps == 0 at all times. Passing such a snapc to
ceph_osdc_alloc_request() (possibly through ceph_osdc_new_request()) is
equivalent to passing NULL, as ceph_osdc_alloc_request() uses it only
for sizing the request message.
Further, in all four cases the subsequent ceph_osdc_build_request() is
passed NULL for snapc, meaning that 0 is encoded for seq and num_snaps
and making ceph_empty_snapc entirely useless. The two cases where it
actually mattered were removed in commits 8605609049 ("ceph: avoid
sending unnessesary FLUSHSNAP message") and 23078637e0 ("ceph: fix
queuing inode to mdsdir's snaprealm").
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Yan, Zheng <zyan@redhat.com>
A negative value rc compared to the positive value ENOENT in the
finish_read() function.
Signed-off-by: Anton Protopopov <a.s.protopopov@gmail.com>
Signed-off-by: Yan, Zheng <zyan@redhat.com>
CURRENT_TIME macro is not appropriate for filesystems as it
doesn't use the right granularity for filesystem timestamps.
Use current_fs_time() instead.
Signed-off-by: Deepa Dinamani <deepa.kernel@gmail.com>
Signed-off-by: Yan, Zheng <zyan@redhat.com>
This patch makes ceph_writepages_start() try using single OSD request
to write all dirty pages within a strip unit. When a nonconsecutive
dirty page is found, ceph_writepages_start() tries starting a new write
operation to existing OSD request. If it succeeds, it uses the new
operation to writeback the dirty page.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
Vlastimil Babka