tree 2f7254e7117eac0fc0d974029e5bb5422866a73c
parent 8b1afc3d6751063d3f0cdefe55719b1cd2f7edcc
author NeilBrown <neilb@suse.de> 1412028642 +1000
committer NeilBrown <neilb@suse.de> 1413252508 +1100

md: don't start resync thread directly from md thread.

The main 'md' thread is needed for processing writes, so if it blocks
write requests could be delayed.

Starting a new thread requires some GFP_KERNEL allocations and so can
wait for writes to complete.  This can deadlock.

So instead, ask a workqueue to start the sync thread.
There is no particular rush for this to happen, so any work queue
will do.

MD_RECOVERY_RUNNING is used to ensure only one thread is started.

Reported-by: BillStuff <billstuff2001@sbcglobal.net>
Signed-off-by: NeilBrown <neilb@suse.de>