TCP route caching panic

[Gleb Smirnoff]

On Thu, Oct 20, 2016 at 12:32:17AM -0500, Mike Karels wrote:
M> > On Wed, Oct 19, 2016 at 07:26:44PM -0500, Mike Karels wrote:
M> > M> > On Wed, Oct 19, 2016 at 02:29:14PM -0700, Gleb Smirnoff wrote:
M> > M> > T>   Hi!
M> > M> > T> 
M> > M> > T>   I got this panic in a bhyve VM, which was just compiling stuff
M> > M> > T> and had two ssh sessions open. It has static network configuration,
M> > M> > T> one interfaces, one IP address, IPv4 only and it sees only hardware
M> > M> > T> address of the host. So it is minimal possible configuration. It
M> > M> > T> runs with INVARIANTS.
M> > M> 
M> > M> > Important note that kernel has FLOWTABLE option. And recently I refined
M> > M> > the original fix of FLOWTABLE+inpcb_route_cache. This might be related,
M> > M> > but I don't see how. The inp code explicitly nulls the ro_lle when
M> > M> > it frees it.
M> > M> 
M> > M> Hmm, I run a similar configuration, but may not run as many compiles.
M> > M> I'm running a recent head GENERIC system, so has INVARIANTS.  I've
M> > M> tested with FLOWTABLE in the past; I'll try it again now.  
M> > M> 
M> > M> But if I understand correctly, you have modifications?  Could we see
M> > M> them (assuming they are not in the tree, I haven't been watching closely)?
M> 
M> > No, I don't have modifications.
M> 
M> Sorry then, what did you mean by this:
M> 
M>     And recently I refined the original fix of FLOWTABLE+inpcb_route_cache.
M>     This might be related,

That is already checked in as r307234.

M> I've run one buildworld in s aimilar environment so far today, using NFS
M> to the host for the sources.  I'll try another.

I also failed to reproduce for the last several hours. Will keep you updated,
since I use the VM daily.

-- 
Totus tuus, Glebius.