Re: [PATCH] Fix over-zealous flush_disk when changing device size.

From
Christoph Hellwig <hch@infradead.org>
Date
2011-03-03 14:31:25
Also in
dm-devel, lkml
Reply to
NeilBrown, 2011-02-17
On Thu, Feb 17, 2011 at 04:50:57PM +1100, NeilBrown wrote:
Hi Andrew (and others)
 I wonder if you would review the following for me and comment.
Please send think in this area through -fsdevel next time, thanks!
There are two cases when we call flush_disk.
In one, the device has disappeared (check_disk_change) so any
data will hold becomes irrelevant.
In the oter, the device has changed size (check_disk_size_change)
so data we hold may be irrelevant.

In both cases it makes sense to discard any 'clean' buffers,
so they will be read back from the device if needed.
Does it?  If the device has disappeared we can't read them back anyway.
If the device has resized to a smaller size the same is true about
those buffers that have gone away, and if it has resized to a larger
size invalidating anything doesn't make sense at all.  I think this
area needs more love than a quick kill_dirty hackjob.
In the former case it makes sense to discard 'dirty' buffers
as there will never be anywhere safe to write the data.  In the
second case it *does*not* make sense to discard dirty buffers
as that will lead to file system corruption when you simply enlarge
the containing devices.
Doing anything like this at the buffer cache layer or inode cache layer
doesn't make any sense.  If a device goes away or shrinks below the
filesystem size the filesystem simply needs to be shut down and in te
former size the admin needs to start a manual repair.  Trying to do
any botch jobs in lower layer never works in practice.

For now I think the best short term fix is to simply revert commit
608aeef17a91747d6303de4df5e2c2e6899a95e8

	"Call flush_disk() after detecting an online resize."
Thread (10 messages) 10 messages, 4 authors, last reply 2011-03-08
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help