Re: [PATCH 03/17] xfs: don't lock the dquot before return in xqcheck_commit_dquot

["Darrick J. Wong" <djwong@kernel.org>]

On Wed, Oct 15, 2025 at 07:00:43AM +0200, Christoph Hellwig wrote:
> On Tue, Oct 14, 2025 at 04:22:44PM -0700, Darrick J. Wong wrote:
> > On Mon, Oct 13, 2025 at 11:48:04AM +0900, Christoph Hellwig wrote:
> > > While xfs_qm_dqput requires the dquot to be locked, the callers can use
> > > the more common xfs_qm_dqrele helper that takes care of locking the dquot
> > > instead.
> > 
> > But the start of xqcheck_commit_dquot does:
> > 
> > 	/* Unlock the dquot just long enough to allocate a transaction. */
> > 	xfs_dqunlock(dq);
> > 	error = xchk_trans_alloc(xqc->sc, 0);
> > >>	xfs_dqlock(dq);
> > 	if (error)
> > 		return error;
> > 
> > So that'll cause a livelock in xfs_qm_dqrele in the loop body below if
> > the transaction allocation fails.
> 
> I think livelock is the wrong term, it's just a leaked lock.  This gets
> fixed towards the end of the series, and I'll fix it by reshuffling
> the series so that this only happens later and at the same time as those
> changes.  Same for the next patch.

Ok let's see how far I get with ignoring dqrele/dqput for the rest of
the series then. ;)

--D