AutoFS mount point is managed by user space daemon, which receives
requests from kernel via pipe, passed on mount operation.
Kernel hold write end, while user space process - read end.
Thus, for successful AutoFS migration, this connection has to be
restored.
During this stage, all the autofs parameters and parsed and collected in
per-mount autofs structure. It also does check that autofs mounts is not in
interim state.

Below are major tricks to be performed to dump AutoFS mount:

1) Read end of the pipe in the process have to be discovered: it will be used
to restore the pipe in case of write pipe end in closed in the process.
Note: migration of AutoFS mounts, which process group leader doesn't have read
pipe end opened is not supported.

2) Read pipe end has to be empty. Otherwise autofs is an interim state.
Interim state is determined by control pipe contents: if it's not empty, then
it means, that kernel requested some operation (mount/umount), which is in
process right now.
In case of pipe is not empty, dump is aborted.

3) Mount points are collected (parsed) from initial pid namespace, which leads
to real pgrp value in AutoFS options instead of virtual one. This have to be
fixed.
Note: we don't care about virtual pgrp in case of catatonic mounts
Signed-off-by: Stanislav Kinsburskiy <skinsbursky@virtuozzo.com>
Signed-off-by: Pavel Emelyanov <xemul@virtuozzo.com>