mdadm: Fix hang race condition in wait_for_zero_forks()

mdadm: Fix hang race condition in wait_for_zero_forks()

Running a create operation with --write-zeros can randomly hang
forever waiting for child processes. This happens roughly on in
ten runs with when running with small (20MB) loop devices.

The bug is caused by the fact that signals can be coallesced into
one if they are not read by signalfd quick enough. So if two children
finish at exactly the same time, only one SIGCHLD will be received
by the parent.

To fix this, wait on all processes with WNOHANG every time a SIGCHLD
is received and exit when all processes have been waited on.

Reported-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
Signed-off-by: Mariusz Tkaczyk <mariusz.tkaczyk@linux.intel.com>