zfs code - Feb 2011 - All-zero blkptr_t in level 1 block

It seems ( via mdb)  that a level 1 parent blocks will have an all-zeroes
blkptr_t in it if the associated data block is a hole.

I find this surprising, because I would expect at least the logical size to be
set.

Could someone from the zfs team confirm that the above is indeed the convention?

TIA

Steve
-- 
This message posted from opensolaris.org

It seems ( via mdb)  that a level 1 parent block will have an all-zeroes
blkptr_t in it if the associated data block is a hole.

I find this surprising, because I would expect at least the logical size to be
set.

Could someone from the zfs team confirm that the above is indeed the convention?

TIA

Steve
-- 
This message posted from opensolaris.org

On Wed, Feb 09, 2011 at 06:21:01PM -0800, Steve Gonczi
wrote:
> It seems ( via mdb) that a level 1 parent blocks will have an
> all-zeroes blkptr_t in it if the associated data block is a hole.
>
> I find this surprising, because I would expect at least the logical
> size to be set.
>
> Could someone from the zfs team confirm that the above is indeed the
> convention?
Hi Steve,

Holes are represented by all-zero block pointers.  When compression is on,
all-zero blocks compress to holes, which means that an indirect block which
contains only holes will be all zero, which also compresses to a hole.  So with
compression on, you can write terrabytes of zero data to a file, without
using any actual bytes.

BP_IS_HOLE(bp) detects the "hole" block pointers by their zero
"blk_birth"
field, which on any normal block pointer contains the transaction group the
block was "born" in.

Cheers,
- jonathan