On 2015/7/29 4:51, Andrew Morton wrote:> On Mon, 27 Jul 2015 11:39:16 +0800 Joseph Qi <joseph.qi at huawei.com> wrote: > >> cluster pos is defined as u32, when calculate corresponding sector it >> should be converted to u64 first, otherwise it may overflow. > > What are the runtime effects of this change? >The issue is when using large volume, for example, 9T volume with 2T already used, frequently create small files with O_DIRECT and the IO is not cluster aligned, it may clear sectors in the wrong place.> Please always include this info when fixing bugs so that others can > work out which kernel version(s) should be patched. >Sorry, I'll bear it in mind.> >
On Wed, 29 Jul 2015 08:30:59 +0800 Joseph Qi <joseph.qi at huawei.com> wrote:> On 2015/7/29 4:51, Andrew Morton wrote: > > On Mon, 27 Jul 2015 11:39:16 +0800 Joseph Qi <joseph.qi at huawei.com> wrote: > > > >> cluster pos is defined as u32, when calculate corresponding sector it > >> should be converted to u64 first, otherwise it may overflow. > > > > What are the runtime effects of this change? > > > The issue is when using large volume, for example, 9T volume with 2T already > used, frequently create small files with O_DIRECT and the IO is not cluster > aligned, it may clear sectors in the wrong place.You mean it corrupts the filesystem? So this wants to be merged asap and backported as far as possible?