[Devel] [PATCH 05/11] fuse: Trust kernel i_mtime only -v2

Maxim Patlasov mpatlasov at parallels.com
Mon Jan 20 03:33:37 PST 2014


On 01/06/2014 08:22 PM, Miklos Szeredi wrote:
> On Thu, Dec 26, 2013 at 07:41:41PM +0400, Maxim Patlasov wrote:
>>>> +
>>>> +	if (!err)
>>>> +		clear_bit(FUSE_I_MTIME_DIRTY, &fi->state);
>>> Doing the test and the clear separately opens a huge race window when i_mtime
>>> modifications are bound to get lost.
>> No. Because the whole operation is protected by i_mutex (see
>> fuse_fsync_common()).
> fuse_release_common() doesn't have i_mutex.  It's probably safe to acquire it,
> but is that really needed?

No, that's not needed, I think. Because by the time of calling 
fuse_release(), file->f_count is already zero and no userspace activity 
is possible on the file.

Are you OK about -v3 version of the patch (sent 12/26/2013)?

Thanks,
Maxim

>
> Thanks,
> Miklos
>




More information about the Devel mailing list