It might be possible to yield an incorrect snapshot if trim isn't handled correctly.

[nixomose]

Need to check if trim commands are cowed, but if they are not, then the drive can conceivably write over the block before we cow it and when we do go to cow it, we will not be producing the snapshot of that block at the point in time, but some write-after-trimmed version instead.

[tcaputi]

I think handling this correctly should just be a matter of removing the !bio_is_discard() check from tracer_should_trace_bio(). We will need to ensure this works before committing it though.