Re: [PATCH v1] xfs: initialize error in xfs_defer_finish_one

[Darrick J. Wong]

On Wed, Aug 03, 2022 at 11:42:21PM +0000, Sherry Yang wrote:
> 
> > On Aug 2, 2022, at 9:31 PM, Darrick J. Wong <djwong@xxxxxxxxxx> wrote:
> > 
> > On Tue, Aug 02, 2022 at 06:49:02AM +1000, Dave Chinner wrote:
> >> On Mon, Aug 01, 2022 at 12:03:11PM -0700, Sherry Yang wrote:
> >>> Path through non-void function 'xfs_defer_finish_one' may return error
> >>> uninitialized if no iteration of 'list_for_each_safe' occurs. Fix this
> >>> by initializing error.
> >> 
> >> I didn't think this situation was possible - how do we get deferred
> >> work queued with no work items on it?
> >> 
> >> If we can return an uninitialised error from xfs_defer_finish_one()
> >> because of an empty queued work, then something else has gone wrong
> >> earlier in the work deferral process. If this can actually happen,
> >> then we need to fix whatever is creating the empty work rather than
> >> paper over it by initialising the error being returned for empty
> >> works...
> > 
> > /me bets this is a response to a static checker that doesn't know that
> > list_empty(&dfp->dfp_work) == false in all circumstances.  It's not
> > possible for tp->t_dfops to contain an xfs_defer_pending with no work
> > items.
> 
> Hi Darrick, 
> 
> You’re correct. This is a false positive bug detected by our static code 
> analysis tool. Sorry for the noise.

Well, thank /you/ for running smatch/sparse/whatever on the XFS code
base.  Let us know if you find any other oddities, since it does tend to
find things every now and then. :)

--D

> Sherry
> > 
> > --D
> > 
> >> Cheers,
> >> 
> >> Dave.
> >> -- 
> >> Dave Chinner
> >> david@xxxxxxxxxxxxx
>