[OpenAFS-devel] Cache inconsistency in client 1.4.8 and above

[Felix Frank]

On Tue, 5 May 2009, Simon Wilkinson wrote:

> So here's my question ...
>
> What is the acceptable solution for this problem, either for 1.4.11, or for a 
> RPM only patch release? Is it to just revert 
> linux-mmap-antirecursion-20081020, or is one of the patches in RT appropriate 
> for kernels with the BDI issues resolved?

The patches in RT are just variations on the theme of 
linux-mmap-antirecursion-20081020. They prevent deadlock at the risk of 
data loss. The fixes in RT solve a cache inconsistency, but data 
corruption is still possible.

People here unanimously voted for machine lockups rather than 
unpredictable (and largely transparent) file corruption. That's why we 
won't send any antirecursion mechanism into production.

So no, none of the solutions present so far seem to be ready for real use. 
Accepting the possibility of deadlocking the virtual memory seems to be 
the healthiest choice at the moment (which I accept to be a matter or 
opinion).

  - Felix