Re: [PATCH 1/4] block: Allow devices to indicate whether discarded blocks are zeroed

[Ric Wheeler <rwheeler@redhat.com>]

On 11/23/2009 11:54 AM, Greg Freemyer wrote:
> On Mon, Nov 23, 2009 at 11:37 AM, Ric Wheeler<rwheeler@redhat.com>  wrote:
>> On 11/21/2009 09:43 PM, Mark Lord wrote:
>>>
>>> Matthew Wilcox wrote:
>>>>
>>>> On Sat, Nov 21, 2009 at 05:13:56AM -0500, Christoph Hellwig wrote:
>>>>>
>>>>> On Fri, Nov 20, 2009 at 09:45:21PM -0500, Martin K. Petersen wrote:
>>>>>>
>>>>>> The discard ioctl is used by mkfs utilities to clear a block device
>>>>>> prior to putting metadata down.  However, not all devices return zeroed
>>>>>> blocks after a discard.  Some drives return stale data, potentially
>>>>>> containing old superblocks.  It is therefore important to know whether
>>>>>> discarded blocks are properly zeroed.
>>>>>
>>>>> At least for mkfs.xfs we make sure to still zero the important areas
>>>>> after the TRIM ioctl anyway.
>>>>
>>>> Could you change that to zero _before_ the TRIM?
>>>
>>> ..
>>>
>>>
>>> Hopefully not for the drives that *don't* guarantee zeros after TRIM.
>>> Eg. most existing ones, including all of the Indilinx chipset drives.
>>>
>>> Cheers
>>>
>>
>> Note that writing to a block after a discard (write or trim) changes the
>> state of that block back to allocated (mapped) in the target. Basically,
>> that would seem to make the trim redundant and irrelevant.
>>
>> You can zero before the discard but that can still fail is the device
>> returns garbage for a trim'ed block I think,
>>
>> Ric
>
> I understood the mkfs process to be:
>
> - discard / trim 100% of the block device
> - write zeros to the relatively small group of blocks that the
> filesystem assumes will be zero'd.
>
> Seems like the only reasonable way to handle it.
>
> Greg

I agree - the discard of the whole device is a good idea.

I just want to make clear that "discard block X; write block X with zeroed data" 
undoes the discard in general :-)

ric