[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

[dm-devel] Re: [PATCH 2.6.19 5/5] fs: freeze_bdev with semaphore not mutex

On Wednesday, 8 November 2006 00:49, Alasdair G Kergon wrote:
> On Wed, Nov 08, 2006 at 12:05:49AM +0100, Rafael J. Wysocki wrote:
> > But freeze_bdev() is supposed to return the result of get_super(bdev)
> > _unconditionally_.  Moreover, in its current form freeze_bdev() _cannot_
> > _fail_, so I don't see how this change doesn't break any existing code.
> > For example freeze_filesystems() (recently added to -mm) will be broken
> > if the down_trylock() is unsuccessful.
> I hadn't noticed that -mm patch.  I'll take a look. Up to now, device-mapper 
> (via dmsetup) and xfs (via xfs_freeze, which dates from before device-mapper
> handled this automatically) were the only users.  Only one freeze should be
> issued at once.  A freeze is a temporary thing, normally used while creating a
> snapshot.  (One problem we still have is lots of old documentation on the web
> advising people to run xfs_freeze before creating device-mapper snapshots.)
> You're right that the down_trylock idea is more trouble than it's worth and
> should be scrapped.

Well, having looked at it once again I think I was wrong that this change would
break freeze_filesystems(), because it only calls freeze_bdev() after checking
if sb->s_frozen is not set to SB_FREEZE_TRANS (freeze_filesystems() is only
called after all of the userspace processes have been frozen).

However, XFS_IOC_FREEZE happily returns 0 after calling freeze_bdev(),
apparetnly assuming that it won't fail.


You never change things by fighting the existing reality.
		R. Buckminster Fuller

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]