[PVFS-developers] Kernel Oops prevention

[Rob Ross]

Hey,

I notice that this only addresses the kpvfsd mode, but that should be fine 
since errors dealt with by the pvfsd shouldn't cause kernel oopses :).

Applied.  Thanks!

Rob

On Thu, 10 Jun 2004, Porter Don wrote:

> 
> In dorking with the size checking stuff I accidentally caused a garbage
> errno to be sent back to the client in a stat call.  This, in turn, was
> passed by the kernel module to ERR_PTR, which in some cases interpreted the
> errno as a real pointer and tried to dereference it, thus oopsing the
> kernel.
> 
> Granted, this is probably the last time in all of pvfs1 history that someone
> will do something so foolish, but I thought it would be good if the kernel
> module failed with something a little less painful than an oops.  Put
> another way, it would be good if a broken manager would not oops the client
> kernel.
> 
> Attached is a patch that catches errno's greater than 1000 and changes them
> to EIO and prints a message to the syslog.  Thus, commands will fail if the
> command is broken, but the kernel will not oops in this case.
> 
> Thoughts?
> 
> Thanks,
> Don
> 
>