RFC: Pass TRIM through GELI

[Steven Hartland]

On 10/07/2015 21:01, Pawel Jakub Dawidek wrote:
> On Sun, Jun 28, 2015 at 08:38:41PM -0500, Matthew D. Fuller wrote:
>>>> Stuffed into bugzilla as
>>>> <https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=198860>
>>> [...]
>>>> After last round, everybody seems happy enough with this, so I've
>>>> filed it as
>>>> <https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=198863>.
>>> Does anybody have outstanding concerns on these?  Or, if not, what
>>> else do we need to move them along?  They're working fine for me
>>> here...
>> Ping...   still working fine here, and I'm pretty sure I've addressed
>> every concern anybody's raised.
> Matthew,
>
> I'm sorry that it took me so long to get to your patch.
>
> The good news is that I like the patch - it looks clean and complete.
> The bad news is that I like it a bit too much:) I think I'd prefer that
> BIO_DELETE is passed through by default and there is an option to turn
> it off. This would mean changing -t option to -T for init and onetime
> and renaming the G_ELI_FLAG_DELETE flag to G_ELI_FLAG_IGNORE_DELETE.
> OR... just removing the ability to ignore BIO_DELETEs. The latter is
> appealing especially if some days we will implement BIO_DELETEs as
> overwrites. Then we should have an option to turn that on, which would
> turn off TRIM/UNMAP.
>
> Thinking about it some more, I believe that if someone doesn't want
> TRIM/UNMAP to hit his SSDs it should be configurable on per-SSD basis
> and not on every layer above SSD. So at the end I'd change my preference
> to just passing BIO_DELETEs always.

This can already be achieved for devices attached via cam/scsi_da e.g.

kern.cam.da.1.delete_method="NONE"


>
> What do you think?
>