[illumos-Developer] Review --> iSCSI UNMAP support

[Bill Sommerfeld]

On 02/11/11 07:44, Sumit Gupta wrote:
> There is an expectation that unmapped space is available for other
> zpool clients. And thats not a narrow view :)

you need to allow for a creative interpretation of the specs when your 
implementation includes features not contemplated by the standards; the 
user is always better off when the features interact sensibly than when 
two features are treated as incompatible and you get to choose to use 
only one.

in this case there is a very clear way for the features to interact 
sensibly: just allow unmap on all volumes, and the refreservation 
feature will cause the right thing to happen in all cases.

when you free allocated blocks in a volume, whether or not the volume 
has a refreservation, the actual blocks freed become available for reuse 
by the same volume or a different volume.  the refreservation controls 
the total number of blocks that can be used by other volumes but it 
doesn't prevent the actual blocks freed from being reused.

also, consider the case where there is a refreservation, but it's 
smaller than the volume size - it would be unreasonable to disallow 
unmap on a sparse volume with a refreservation.

> If folks want to enable a certain feature because of its merit, you
> can always add an option to sbd. The implementer of the array (e.g.
> Nexenta) may choose the default (enable that option or not).

I don't see any value in a tunable that disables unmap for volumes with 
a refreservation.

					- Bill