We have been seeing some process crashes when performing hot remove of our virtual NVMeF
target devices from the subsystem.
Looking at the code for ns_remove - I saw a comment that states that there is no real
quiesce of IO during subsystem pause, which seems to cause IO already in the system to
attempt completions when the underlying device is no longer valid.
Is there a plan to support this? Is there a design of how to quiesce the system without
locking in the fast path?
Show replies by date