Re: unionfs and kernel core dumps

[Thor Lancelot Simon <tls%panix.com@localhost>]

On Sat, Jul 03, 2010 at 05:00:59PM +0200, Michael Felke wrote:
> Martin S. Weber wrote:
>> On Fri, Jul 02, 2010 at 10:31:48PM -0400, Thor Lancelot Simon wrote:
>>> On Fri, Jul 02, 2010 at 10:42:18PM +0000, Taylor R Campbell wrote:
>>>> (I am not subscribed to this list, so please cc me in replies.)
>>>>
>>>> I would like to use unionfs for various purposes.
>>> Good luck.  Have you noticed some of the oddities about file and
>>> directory search and creation yet?
>>>
>>> The original Ficus paper described unionfs as not-ready-for-prime-time,
>>> the DMFS paper describing the fixes that made our stacking filesystems
>>> work at all said the same thing, and it's basically still true today.
>>
>> Having tried basing a live-cd on unionfs I can only agree. Don't use it
>> for anything serious, it'll just burn your time. It sounds great, but
>> there's a big list of problems. Andrew Doran, Iirc, was once sending
>> around a compilation of things that need to be solved so it could at
>> least be considered less experimental. That was a couple years ago. It's
>> not been resolved since.
>>
>> Regards,
>>
>> -Martin
> It would have been nice, when some had added the list to the man page on
> unionfs. Because we are am it on our server.
> I recognizes that the kernel hangs, if you try to unmount the stacked  
> file systems in the wrong order. But besides this, it seems to work.

It doesn't.  Nodes, for instance, are often looked up or created in the
wrong layer.  This is a particular nuisance with directory creation.

Thor