Re: [PATCH 1/2] Add discard/nodiscard mount option for ext3

[Eric Sandeen <sandeen@redhat.com>]

On 07/12/2010 11:15 AM, Lukas Czerner wrote:
> On Mon, 12 Jul 2010, Lukas Czerner wrote:
> 
>> On Mon, 12 Jul 2010, Ric Wheeler wrote:
>>
>>> On 07/12/2010 11:19 AM, Jan Kara wrote:
>>>>> Those mount option has the same meaning as in ext4 file system. It
>>>>> provide a way to enable/disable file system's trim support. The trim
>>>>> support is off by default, thus nodiscard option is not actually
>>>>> necessary.
>>>>    I kind of miss why ext3 should have a 'discard' mount option. When
>>>> user calls DISCARD ioctl on the filesystem, then he probably wants
>>>> discard to be performed.
>>>>
>>>> 								Honza
>>>>
>>>
>>> Sorry I misunderstood your original question.
>>>
>>> One reason that you might want to have a "discard" option is to allow a system
>>> admin to mount without barriers to protect flaky hardware (we have had some
>>> mixed results for example). As you say, the user probably wants to have the
>>> ioctl do the discard and should be reasonable for doing it only on solid
>>> devices,
>>
>> The question is what in does on device other than SSD. I know it does
>> not harm the deivce, but is there some kernel logic preventing the trim
>> command to be send to device that does not support it ? I hope so.
>>
> 
> Yes, there is a check whether device support trim in blkdev_issue_discard
> code.

I also sent an ext4 patch so that a failed discard clears the flag so we
don't keep attempting more.

However, the LVM guys aren't too happy with that because they point to
cases like:

a) pvmove off of, and back onto, a trim-capable device
b) composite devices of trim-capable and -incapable devices

I don't care much either way, I don't think we do much harm in
continuing to send discards after a failure; the original motivation for
the patch was so that if a user asked for discard and the (simple)
device didn't support it, they'd get a message, and we'd stop trying.

-Eric

> -Lukas