Re: [ndctl PATCH] ndctl, check: Add a sigbus handler to detect metadata corruption

["Verma, Vishal L" <vishal.l.verma@intel.com>]

On Fri, 2017-04-14 at 00:10 +0000, Rudoff, Andy wrote:
> > If we have poison/badblocks in the BTT metadata sections, the mmap-
> > reads
> > happening in the checker will trigger a SIGBUS, and the program
> > will
> > halt abruptly. Add a sigbus handler which notifies the user of
> > this, and
> > prints out a relevant error message:
> > 
> >  namespace5.0: namespace_check: checking namespace5.0
> >  namespace5.0: btt_discover_arenas: found 1 BTT arena
> >  namespace5.0: btt_check_arenas: checking arena 0
> >  namespace5.0: namespace_check: Received a SIGBUS
> >  namespace5.0: namespace_check: Metadata corruption found, recovery
> > is not possible
> >  error checking namespaces: Bad address
> 
> If in repair mode, should a SIGBUS cause you to set the error state
> in the associated arena?
> 
> -andy
> 
So I thought a bit about this -
While we can set the error state, that leaves the BTT with no way to
get any data off it (currently).

If we allow the kernel to enable the namespace, the user has some
chance to do at least partial recovery until they hit the badblock.
(assuming the badblock is in the map, which is the largest on-disk
structure). If it is on the flog or info block, then, yes, there is no
recovery.

	-Vishal
_______________________________________________
Linux-nvdimm mailing list
Linux-nvdimm@lists.01.org
https://lists.01.org/mailman/listinfo/linux-nvdimm