First page Back Continue Last page Graphics
Barriers and Device-mapper (2)
Secondly, we added a target barrier interface whereby each target instructs the device-mapper core to call it N times in succession. Each of the N requests is numbered, so each target can easily map the requests onto the devices that need barriers sent to them. This neatly dodges the memory problem – if we're short, later barrier request will have to wait for the earlier ones to complete but otherwise we can issue them in parallel.
This code is in linux-next, with the exception of the raid1 target which is still awaiting review.