[CRIU] [PATCH] check: Make sure dirty tracking not used on unpatched kernel

[Pavel Emelyanov]

On 01/30/2014 10:08 PM, Cyrill Gorcunov wrote:
> On Thu, Jan 30, 2014 at 09:57:06PM +0400, Pavel Emelyanov wrote:
>> On 01/30/2014 09:40 PM, Cyrill Gorcunov wrote:
>>> Signed-off-by: Cyrill Gorcunov <gorcunov at openvz.org>
>>> ---
>>>  kerndat.c | 12 ++++++++++++
>>>  1 file changed, 12 insertions(+)
>>>
>>> diff --git a/kerndat.c b/kerndat.c
>>> index 27acf0cc2ed1..04c937f1814f 100644
>>> --- a/kerndat.c
>>> +++ b/kerndat.c
>>> @@ -88,6 +88,18 @@ int kerndat_get_dirty_track(void)
>>>  
>>>  	map[0] = '\0';
>>>  
>>> +	/*
>>> +	 * It's known some kernel versions may have a dirty
>>> +	 * page tracking bug which clears soft dirty bit if
>>> +	 * mprotect code is unpatched. So in such case don't
>>> +	 * allow to use dirty page tracking.
>>> +	 */
>>> +	if (mprotect(map, PAGE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC)) {
>>
>> O_o Why mprotect failure results in disabled tracker? And how does it
>> correlate with soft-dirty BUG()?
>> Anyway -- I'm not applying the patch, since the fix is going to be in
>> stables soon.
> 
> If mprotect failed it means we can't test for bug, this also means
> we can't be sure if traking is consistent, better to refuse traking
> than endup with incomplete data on restore. Are we sure all users
> to run criu with stable kernel updated?

No. But as we don't have stable kernel yet, this means we have no way
to use tracker.