Re: sparse loop devices and ext3

["Stephen C. Tweedie" <sct redhat com>]

Hi,

On Tue, Feb 06, 2001 at 04:50:01PM -0800, Peter J. Braam wrote:
> 
> The problem I saw was that the allocation of blocks was not handled
> all that well and seriously rolled back during recovery.

Loop doesn't allocate blocks as far as I know.  At least, not on files
using bmap.  bmap() is a read-only API.
 
> So shouldn't bmap in ext3 start a transaction to eliminate that?

No, because bmap() never allocates anything.  Ever.

One thing on the todo list for bmap() is to make it force any
data-journaled updates which are still in the log to be migrated back
to their main home location on disk before returning.  That would be
needed to make "lilo; sync; reboot -f" safe on data-journaled
partitions (the sync only forces the data to the journal, not
necessarily to the primary disk blocks).  But that's not an issue for
loop devices, as far as I can see.

The loop device _does_ check for sparse files: if it finds a zero
return on bmap(), it then calls the filesystem's write function to
allocate that block of the file.  In the ext3 case, that *will* be
journaled.

Cheers,
 Stephen