[CRIU] [PATCH] mnt: Carry run-time device ID in mount_info

[Pavel Emelyanov]

On 12/03/2015 04:31 PM, Cyrill Gorcunov wrote:
> On Thu, Dec 03, 2015 at 03:48:08PM +0300, Pavel Emelyanov wrote:
>>>
>>> How is that? There is two points where @s_dev and @s_dev_rt are
>>> assigned
>>>
>>> 1) When we parse procfs => @s_dev = @s_dev_rt always
>>> 2) When we do mount the device and fetch new run-time @s_dev_rt
>>>    from the stat call, so I really don't see how coincide may happen/
>>
>> Consider you restore two mountpoints that have devices 1.11 and 2.22 in 
>> the images. When you mount them they get 2.22 and 3.33 devices respectively.
>>
>> Now you try to open the 2nd mountpoint, but due to the BUG() in criu code
>> you effectively open the 1st one. So with your patch you will compare
>>
>> 	if (opened_dev != 2.22 and opened_dev != 3.33)
>> 		goto error.
>>
>> and will NOT detect this error, but should.
> 
> So we have to test @s_dev_rt only, right?

Yup.

>>
>>>>
>>>> This stat() should in in do_mount_one() after the 
>>>
>>> Why? Every mount point created go through restore_shared_options, don't they?
>>
>> Yes, but it's quite strange that a routine that has to restore shared options
>> (that's its name) suddenly picks up the device maj:min pair.
> 
> I see. OK, will do. You know, while it looks indeed a bit strange in lexical
> context it allows me to made code simplier ;)

If you move this into the do_mount_one() it will be the same size.

-- Pavel