Re: Fun with SSD and GPT wedges

[Greg Troxel <gdt%lexort.com@localhost>]

John Nemeth <jnemeth%cue.bc.ca@localhost> writes:

> } With disklabels, when the label is scanned then the various abcdefgh
> } partitions can be used.  Ideally, when writing the block with the
> } disklabel it would be rescanned.
>
>      What do you mean by "rescanned"?  Normally with disklabels
> there are no wedges, unless you create them yourself or use the
> DKWEDGE_METHOD_BSDLABEL kernel option.

There aren't wedges, but things like /dev/rwd0f seem to behave like
autodetected wedges in that they automatically refer to regions of the
disk based on reading a label when the disk is attached.

So I mean that when a label is written, that process or reading the
label and setting up how /dev/rwdN[a-h] map to regions should be
repeated.

Basically, my view is that if we're going to have any automatic
processing of the relationship between gpt contents and wedges, it
should be done in a consistent way that maintains invariants, so that
more or less in a given state, regardless of how we got there, the same
thing is true.

It's certainly fine to make things be an error if some of those state
transitions aren't allowed, similar to prohibiting writes to mounted
partitions at some securelevels.