[Bug 286179] Passing a vlan interface to a vnet jail result in a stuck dying state

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=286179

--- Comment #30 from Peter Ganzhorn <[email protected]> ---
Thank you very much for your explainations!

I don't use any drivers from ports, so my igb driver is the one from the
kernel.
That idea of a difference between virtual and 'real' interfaces probably wasn't
correct and just a wild guess.

I recall using ifconfig to assign an ip address to the igb interface in the
jail, but I am not sure if I actually physically connected it to a network.
If it never received any traffic, isc_rxd_pkt_get may have returned without an
error causing CURVNET_RESTORE() to be called in iflib_rxeof()? I probably
should have made sure to test with traffic...
The port of the X710 I created the virtual function on was definitely connected
and therefore received traffic.

After your explanation of the sequence I understand you beeing skeptical now.
If the refcount needs to reach zero before actually beginning the teardown of
the vnet, my patch shouldn't have made a difference if there is another
(earlier) issue at hand.

I will try to do some tests with different configurations for the jail
filesystem and share what causes the jail to get stuck and if/what (e.g.
forcefully unmounting) will let it finally stop or not.
Hopefully that will help to narrow down where a reference may be held.


And finally: I think I created a situation while testing where the dataset was
set jailed=on and delegated to the jail.
Still, it was mounted on the host and visible with zfs list at the same time.
I will try to recreate that configuration and share it with you. I didn't keep
copies of all the different configurations I tested with...

-- 
You are receiving this mail because:
You are on the CC list for the bug.