Re: [PATCH, RFC 2/2] dax: use range_lock instead of i_mmap_lock

[Boaz Harrosh]

On 08/11/2015 11:26 PM, Kirill A. Shutemov wrote:
> On Tue, Aug 11, 2015 at 07:17:12PM +0300, Boaz Harrosh wrote:
>> On 08/11/2015 06:28 PM, Kirill A. Shutemov wrote:
>>> We also used lock_page() to make sure we shoot out all pages as we don't
>>> exclude page faults during truncate. Consider this race:
>>>
>>> 	<fault>			<truncate>
>>> 	get_block
>>> 	check i_size
>>>     				update i_size
>>> 				unmap
>>> 	setup pte
>>>
>>
>> Please consider this senario then:
>>
>>  	<fault>			<truncate>
>> 	read_lock(inode)
>>
>>  	get_block
>>  	check i_size
>> 	
>> 	read_unlock(inode)
>>
>> 				write_lock(inode)
>>
>>      				update i_size
>> 				* remove allocated blocks
>>  				unmap
>>
>> 				write_unlock(inode)
>>
>>  	setup pte
>>
>> IS what you suppose to do in xfs
> 
> Do you realize that you describe a race? :-P
> 
> Exactly in this scenario pfn your pte point to is not belong to the file
> anymore. Have fun.
> 

Sorry yes I have written it wrong, I have now returned to read the actual code
and the setup pte part is also part of the read lock inside the fault handler
before the release of the r_lock.
Da of course it is, it is the page_fault handler that does the
vm_insert_mixed(vma,,pfn) and in the case of concurrent faults the second
call to vm_insert_mixed will return -EBUSY which means all is well.

So the only thing left is the fault-to-fault zero-the-page race as Matthew described
and as Dave and me think we can make this part of the FS's get_block where it is
more natural.

Thanks
Boaz

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/