Re: [PATCH 2/3] HWPOISON: undo memory error handling for dirty pagecache

["Jun'ichi Nomura" <j-nomura@ce.jp.nec.com>]

On 08/11/12 08:09, Andi Kleen wrote:
> Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> writes:
> 
>> Current memory error handling on dirty pagecache has a bug that user
>> processes who use corrupted pages via read() or write() can't be aware
>> of the memory error and result in discarding dirty data silently.
>>
>> The following patch is to improve handling/reporting memory errors on
>> this case, but as a short term solution I suggest that we should undo
>> the present error handling code and just leave errors for such cases
>> (which expect the 2nd MCE to panic the system) to ensure data consistency.
> 
> Not sure that's the right approach. It's not worse than any other IO 
> errors isn't it? 

IMO, it's worse in certain cases.  For example, producer-consumer type
program which uses file as a temporary storage.
Current memory-failure.c drops produced data from dirty pagecache
and allows reader to consume old or empty data from disk (silently!),
that's what I think HWPOISON should prevent.

Similar thing could happen theoretically with disk I/O errors,
though, practically those errors are often persistent and reader will
likely get errors again instead of bad data.
Also, ext3/ext4 has an option to panic when an error is detected,
for people who want to avoid corruption on intermittent errors.

-- 
Jun'ichi Nomura, NEC Corporation

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>