On 12 Dec 2017, at 9:29, Christoph Hellwig wrote:
On Fri, Dec 08, 2017 at 12:52:56PM -0500, Benjamin Coddington wrote:
This set improves the blocklayoutdriver's handling of layouts when
devices
are not present on the a client, or become unavailable. Ideally,
clients
should not continually spam a server with LAYOUTGET for known
layouts, and
GETDEVINFO for known devices, since it may be a common scenario to
have some
clients that have access to the block devices, and some clients that
do not.
In addition to better handling situations where devices are
unavailable,
patches 2 and 3 fix crashes if SCSI devices do not exist or are
already have
existing reservations.
What testcases do you have for this?
The simplest reproduction of a problem would be to try to mount without
having the block devices present at all. That pops -EIO to the
application
instead of falling back to the MDS.
I've run this through the NFS cthon and xfstests while using iscsiadm to
login/logout of a discovered target every second. This conveniently
adds/removes the block devices repeatedly during the test.. something
like:
while true; do
iscsiadm -m node -T
iqn.2003-01.org.linux-iscsi.godel.x8664:sn.0770b0253b97 --logout
sleep 1;
iscsiadm -m node -T
iqn.2003-01.org.linux-iscsi.godel.x8664:sn.0770b0253b97 --login
done
I don't have recent memory of a way to reproduce the crashes I saw.. I'd
have to go back several months to when I first noted a problem. If
necessary I can do some work to dredge them up.
Ben
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
