Re: [PATCH v3 2/3] xfs: enable xfs_trans_cancel() to report and error code

[Carlos Maiolino <cem@kernel.org>]

On Mon, Sep 21, 2026 at 10:33:56AM +0200, Christoph Hellwig wrote:
> On Thu, Sep 17, 2026 at 12:35:47PM +0200, cem@kernel.org wrote:
> > From: Carlos Maiolino <cem@kernel.org>
> > 
> > Once in a while it's useful to know exactly what kind of error caused a
> > transaction to be cancelled.
> > Enable xfs_trans_cancel() to receive an error code to be reported.
> > 
> > Signed-off-by: Carlos Maiolino <cmaiolino@redhat.com>
> > Reviewed-by: "Darrick J. Wong" <djwong@kernel.org>
> > ---
> > V3:
> > 	- Kill typedef from __xfs_trans_cancel
> > 	- Remove uneeded wrap of parameter 0
> > V2:
> > 	- Parenthesize macro parameters
> > 
> >  fs/xfs/xfs_trans.c | 8 +++++---
> >  fs/xfs/xfs_trans.h | 6 +++++-
> >  2 files changed, 10 insertions(+), 4 deletions(-)
> > 
> > diff --git a/fs/xfs/xfs_trans.c b/fs/xfs/xfs_trans.c
> > index 5c522790e5be..395c896673fe 100644
> > --- a/fs/xfs/xfs_trans.c
> > +++ b/fs/xfs/xfs_trans.c
> > @@ -943,8 +943,9 @@ xfs_trans_commit(
> >   * xfs_trans_commit().
> >   */
> >  void
> > -xfs_trans_cancel(
> > -	struct xfs_trans	*tp)
> > +__xfs_trans_cancel(
> > +	struct xfs_trans	*tp,
> > +	int			error)
> >  {
> >  	struct xfs_mount	*mp = tp->t_mountp;
> >  	struct xlog		*log = mp->m_log;
> > @@ -971,7 +972,8 @@ xfs_trans_cancel(
> >  	 * here.
> >  	 */
> >  	if (dirty && !xfs_is_shutdown(mp)) {
> > -		XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, 0, mp);
> > +		XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW,
> > +				 error, mp);
> >  		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
> >  	}
> >  #ifdef DEBUG
> > diff --git a/fs/xfs/xfs_trans.h b/fs/xfs/xfs_trans.h
> > index eb83c5dac032..36a49622b870 100644
> > --- a/fs/xfs/xfs_trans.h
> > +++ b/fs/xfs/xfs_trans.h
> > @@ -213,6 +213,11 @@ xfs_trans_read_buf(
> >  				      flags, bpp, ops);
> >  }
> >  
> > +void __xfs_trans_cancel(struct xfs_trans *, int);
> > +
> > +#define xfs_trans_cancel(tp) __xfs_trans_cancel((tp), 0)
> > +#define xfs_trans_cancel_error(tp, error) __xfs_trans_cancel((tp), (error))
> 
> Can we turn these into inlines instead of macros?

This sounds good.

> 
> And do away with the pointless __xfs_trans_cancel, just implement
> xfs_trans_cancel_error in xfs_trans.c, and wrap xfs_trans_cancel around
> it.

The main reason I created __xfs_trans_cancel was to avoid a fs/xfs/*
change of calls to xfs_trans_cancel().

If you think that's a must, I can do that in this patch, otherwise I'd
prefer to add it on a separate patch just removing the call, without any
logic change.
What do you think?

Cheers.