Re: [Qemu-devel] [PATCH] report read/write errors to IDE guest driver as ECC errors

[Anthony Liguori <anthony@codemonkey.ws>]

Ian Jackson wrote:
> Daniel P. Berrange writes ("Re: [Qemu-devel] [PATCH] report read/write errors to IDE guest driver as ECC errors"):
>   
>> Unless someone wants to implement the ENOSPC handling right now, I'd 
>> like to see this patch just committed as is, so we at least get
>> incremental benefit over current behaviour, which definitely *does*
>> corrupt guest filesystems by silently pretending the write succeeed.
>>     
>
> Absolutely.
>
>   
>> Special ENOSPC handling can be added on top.
>> I agree that pausing the guest is probably best option in that scenario,
>>     
>
> I disagree.  Most reasonable guests will have special handling for
> write failures on their disks.  For example, Linux will (in the
> default setup) remount the fs readonly precisely to prevent
> corruption.
>   

But corruption isn't the issue.  The issue is that you're out of disk 
space.  If using qcow, an attempt to remount the fs may result in even 
more errors if the qcow file has to expand (due to CoW).

Consider the case of guest RAID.  Once the ECC errors are detected, the 
guest will start trying to rebuild the failed drive and since we're 
already out of space, this is going to be catastrophic.  This is why 
ENOSPC should be handled differently than EIO.

> Pausing the guest denies the guest the ability to take whatever action
> it really wants to.
>   

I agree for EIO, but a guest cannot take any actions when ENOSPC occurs 
because it has no concept that such a thing can occur.

Regards,

Anthony Liguori

> Ian.
>
>
>