Re: [PATCH 0/2] Fix blkdev_issue_discard()

[Hannes Reinecke <hare@suse.de>]

On 12/16/2015 09:44 AM, Jan Kara wrote:
> On Wed 16-12-15 00:12:23, Mike Snitzer wrote:
>> On Tue, Dec 15 2015 at  9:38pm -0500,
>> Martin K. Petersen <martin.petersen@oracle.com> wrote:
>>
>>>>>>>> "Bart" == Bart Van Assche <bart.vanassche@sandisk.com> writes:
>>>
>>> Bart> Should the caller of blkdev_issue_discard() implement this or
>>> Bart> should this functionality be added in blkdev_issue_discard()
>>> Bart> itself ?
>>>
>>> I'm very much against dropping information in the ioctl/filesystem
>>> submission path. But I am not against having a helper function of some
>>> sort that DM and target could use to handle heads and tails.
>>>
>>> Right now DM does not handle this and I think it would be worthwhile to
>>> have. Mike, what are your requirements?
>>
>> If the discard doesn't conform to the device's stacked discard limits
>> then yes the head and/or tail will get dropped on the floor.
>>
>> Not sure why DM is being made the focal point on this.  Are you thinking
>> specifically about DM-thinp and its discard_granularity?  DM thinp
>> doesn't support partial thinp block discards -- so misaligned/partial
>> discards are dropped by DM thinp.  There isn't a compelling case for
>> fixing this (by adding a bitset for each thinp block to support finer
>> grained discards) -- at least not that I'm aware of.
>>
>> But I'm missing exactly what it is your helper function would do... and
>> yet you're asking me what my requirements are.. sorry ;)
>
> I agree that just dropping head & tail from misaligned discard request
> looks reasonable. However I think the original motivation for this thread
> was actually blkdev_issue_zeroout(). That ends up being implemented using
> blkdev_issue_discard() and in that case silently dropping head and tail is
> not an option. Supporting non-aligned blkdev_issue_zeroout() via submitting
> properly aligned blkdev_issue_discard() and then just submitting writes
> with zeros for head & tail would look useful to me.
>
And we need a similar thing for SMR drives, too.
SMR drives might end up supporting discard only for a certain range 
of blocks, _and_ set 'discard_zeroes_data', too, for those ranges.
So for those we'd need a similar thing.

Cheers,

Hannes
-- 
Dr. Hannes Reinecke		               zSeries & Storage
hare@suse.de			               +49 911 74053 688
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: F. Imendörffer, J. Smithard, J. Guild, D. Upmanyu, G. Norton
HRB 21284 (AG Nürnberg)