On Tue, Dec 22, 2020 at 06:24:09PM +0100, Hannes Reinecke wrote:

However, lookup_bdev() now always recurses into the filesystem, causing 
multipath to stall in an all-paths-down scenario.

 
I have not read the background about whatever the new problem is - I'm
jumping in cold seeing this message - but from the very beginning of
device-mapper we have strongly recommended that userspace supplies the
block device in the form MAJOR:MINOR and all our own tools do that.  We
guarantee not to deadlock in these places when this is done.

We also accept the device in the form of a path name as we know there
are times when this is safe and convenient, but then we offer no
guarantees - we place the responsibility upon userspace only to do this
when it knows it is safe to do so i.e. no race and no deadlock.

Alasdair