zfs code - Aug 2007 - When does blkptr_t->blk_birth get filled in ? (IV generation for crypto)

I''ve been hoping to use elements of the blkptr_t as the initalisation 
vector (IV) for the AES crypto algorithms - specifically the offset and 
blk_birth.  When do these get filled in ?

I was expecting them to be filled in by the time we reached the 
ZIO_WRITE_ENCRYPT_STAGE (after compression but before checksum), but it 
appears that only blk_props is filled in at this point.

So maybe I should be looking at io_txg instead.

I need an IV that is guaranteed unique over the life time of the pool 
that is also unique per encryption key (which for phase 1 of zfs-crypto 
means unique per dataset).   Since there could be multiple blocks being 
encrypted for the same dataset in a given transaction group (highly 
likely in fact) the txg alone is not enough.

-- 
Darren J Moffat

Darren J Moffat wrote:
> I''ve been hoping to use elements of the blkptr_t as the
initalisation
> vector (IV) for the AES crypto algorithms - specifically the offset and 
> blk_birth.  When do these get filled in ?
This happens during the allocate phase of the pipeline (either 
zio_dva_allocate or zio_write_allocate_gang_members). The value will be 
set to whatever zio->io_txg holds.
> 
> I was expecting them to be filled in by the time we reached the 
> ZIO_WRITE_ENCRYPT_STAGE (after compression but before checksum), but it 
> appears that only blk_props is filled in at this point.
> 
> So maybe I should be looking at io_txg instead.
Should work.
> 
> I need an IV that is guaranteed unique over the life time of the pool 
> that is also unique per encryption key (which for phase 1 of zfs-crypto 
> means unique per dataset).   Since there could be multiple blocks being 
> encrypted for the same dataset in a given transaction group (highly 
> likely in fact) the txg alone is not enough.
>

George Wilson wrote:
> 
> Darren J Moffat wrote:
>> I''ve been hoping to use elements of the blkptr_t as the
initalisation
>> vector (IV) for the AES crypto algorithms - specifically the offset and
>> blk_birth.  When do these get filled in ?
> 
> This happens during the allocate phase of the pipeline (either 
> zio_dva_allocate or zio_write_allocate_gang_members). The value will be 
> set to whatever zio->io_txg holds.
Are there any possible cases where the blk_birth will not be io_txg once 
it is written out to disk ?

-- 
Darren J Moffat

Darren,

I believe the exception would be zil blocks.

Thanks,
George

Darren J Moffat wrote:
> George Wilson wrote:
>>
>> Darren J Moffat wrote:
>>> I''ve been hoping to use elements of the blkptr_t as the
initalisation
>>> vector (IV) for the AES crypto algorithms - specifically the offset
>>> and blk_birth.  When do these get filled in ?
>>
>> This happens during the allocate phase of the pipeline (either 
>> zio_dva_allocate or zio_write_allocate_gang_members). The value will 
>> be set to whatever zio->io_txg holds.
> 
> Are there any possible cases where the blk_birth will not be io_txg once 
> it is written out to disk ?
>