10.2-RC1/STABLE VIMAGE memory leak progress?

Paul Mather freebsd-lists at gromit.dlib.vt.edu
Mon Jul 27 15:50:05 UTC 2015


I'm running 10-STABLE (which currently identifies itself as 10.2-PRERELEASE) and lately I've started experimenting with using iocage for Jail management.  (I really like it so far.)  In particular, I've been trying out iocage's support for VNET networking that relies on VIMAGE and uses bridge and epair for the network.

I find that when I stop a VNET Jail using iocage that I get something akin to the following in my console logs:

=====
Jul 27 10:59:47 chumby kernel: vnet0:1: link state changed to DOWN
Jul 27 10:59:47 chumby kernel: vnet0: link state changed to DOWN
Jul 27 10:59:47 chumby kernel: re0: link state changed to DOWN
Jul 27 10:59:47 chumby kernel: bridge0: link state changed to DOWN
Jul 27 10:59:47 chumby kernel: vnet1:1: link state changed to DOWN
Jul 27 10:59:47 chumby kernel: vnet1: link state changed to DOWN
Jul 27 10:59:47 chumby kernel: bridge1: link state changed to DOWN
Jul 27 10:59:47 chumby kernel: ifa_del_loopback_route: deletion failed: 48
Jul 27 10:59:47 chumby kernel: Freed UMA keg (udp_inpcb) was not empty (120 items).  Lost 12 pages of memory.
Jul 27 10:59:47 chumby kernel: Freed UMA keg (udpcb) was not empty (1169 items).  Lost 7 pages of memory.
Jul 27 10:59:47 chumby kernel: hhook_vnet_uninit: hhook_head type=1, id=1 cleanup required
Jul 27 10:59:47 chumby kernel: hhook_vnet_uninit: hhook_head type=1, id=0 cleanup required
=====

I found a PR (kern/164763) dating back to 2012-02-04 describing the problem.  The last update on that PR was on 2014-10-17, which made reference to some unmerged fixes that would be good to merge into HEAD.

Does anyone know whether this was done?

One of the entries in that PR states, "Our current rule of thumb is 'if you need to shutdown one vimage jail then you need to reboot the host.'"  It then goes on to observe, "Of course this is far from optimal."

Is it safe to say, then, that dynamic VNET Jails are not recommended still under FreeBSD?

Judging by that PR, this is not likely to be fixed for 10.2-RELEASE?

Cheers,

Paul.




More information about the freebsd-stable mailing list