Re: [PATCH] xfs: fix in the setting of logbsize

[Ales Novak <alnovak@xxxxxxx>]

On 2015-6-8 15:04, Mark Tinguely wrote:

> On 06/07/15 04:16, Ales Novak wrote:
> > On 2015-6-6 00:22, Dave Chinner wrote:
> >
> > > Can you give more information about the crashes? From this
> > > description, I do not know whether this is a work around or a fix
> > > because I don't know what code is actually causing the problem or
> > > the circumstances in which the crash occurs. Hence I cannot
> > > determine if your change is the right change to make or whether the
> > > documetnation is simply wrong and we have a real bug we shoul dbe
> > > fixing.
> > The crashes occurred with logsunit=192k and logbsize=256k. xlog_sync()
> > is calculating in logsunit units:
> >
> > count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
> >
> > With count_init=256k, the calculated roundoff will try to cover that in
> > two 192k units. This roundoff is passed to the xlog_pack_data() which
> > walks the buffer and (hopefully) fails when it hits its 256k border.
> Talked about this once before:
>
> http://oss.sgi.com/archives/xfs/2013-03/msg00039.html
>
> Mark.
Thank you! I missed that.

Still I'd advocate for at least simple solution. Sure it's a root's fault 
if he tries to set logbsize not being integer multiple of logsunit, but 
crashing a kernel, that's bad...
--
Ales Novak