Re: [PATCH 17/34] libext2fs: support allocating uninit blocks in bmap2()

["Darrick J. Wong" <darrick.wong@oracle.com>]

On Mon, Oct 13, 2014 at 10:35:26AM -0400, Theodore Ts'o wrote:
> On Sat, Sep 13, 2014 at 03:13:06PM -0700, Darrick J. Wong wrote:
> > In order to support fallocate, we need to be able to have
> > ext2fs_bmap2() allocate blocks and put them into uninitialized
> > extents.  There's a flag to do this in the extent code, but it's not
> > exposed to the bmap2 interface, so plumb that in.  Eventually
> > fallocate or fuse2fs or somebody will use it.
> > 
> > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > ---
> >  lib/ext2fs/bmap.c   |   24 ++++++++++++++++++++++--
> >  lib/ext2fs/ext2fs.h |    1 +
> >  2 files changed, 23 insertions(+), 2 deletions(-)
> > 
> > 
> > diff --git a/lib/ext2fs/bmap.c b/lib/ext2fs/bmap.c
> > index c1d0e6f..a4dc8ef 100644
> > --- a/lib/ext2fs/bmap.c
> > +++ b/lib/ext2fs/bmap.c
> > @@ -72,6 +72,11 @@ static _BMAP_INLINE_ errcode_t block_ind_bmap(ext2_filsys fs, int flags,
> >  					    block_buf + fs->blocksize, &b);
> >  		if (retval)
> >  			return retval;
> > +		if (flags & BMAP_UNINIT) {
> > +			retval = ext2fs_zero_blocks2(fs, b, 1, NULL, NULL);
> > +			if (retval)
> > +				return retval;
> > +		}
> 
> What I think we should do is to have two separate new BMAP_ flags;
> BMAP_UNINIT, which sets the uninit bit, and BMAP_ZERO, which requests
> that the block be zeroed.  I don't think it should follow that whe you
> set the uninit bit via the libext2fs, the block wil automatically be
> zeroed.  After all, userspace can't assume that if the uninit bit is
> set, that the block will be pre-zeroed, since files fallocated by the
> kernel won't meet that guarantee.

On an extent based file, we can record the BLOCK_UNINIT status in the extent
flags so that subsequent reads return zeroes.  On a block mapped file it's not
possible to record the uninitialized status (short of unmapping the block), so
here I was trying to emulate the read behavior you'd get with an extent file.

Kernel fallocate() refuses to service non-extent files, so there's not much
precedent there unless you want to block BMAP_UNINIT on such files.

So... I agree that BMAP_ZERO would be a useful feature anyway.  My question is,
if we pass in a non-extent file with BMAP_UNINIT but not BMAP_ZERO, should we
simply return -EINVAL?

--D

> 
> 					- Ted
> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html