[Linux-cachefs] [PATCH v11 3/4] xfs: refactor the usage around xfs_trans_context_{set, clear}

[Darrick J. Wong]

On Wed, Dec 09, 2020 at 09:47:38AM +0800, Yafang Shao wrote:
> On Wed, Dec 9, 2020 at 2:59 AM Darrick J. Wong <darrick.wong at oracle.com> wrote:
> >
> > On Tue, Dec 08, 2020 at 08:28:23PM +0800, Yafang Shao wrote:
> > > The xfs_trans context should be active after it is allocated, and
> > > deactive when it is freed.
> > >
> > > So these two helpers are refactored as,
> > > - xfs_trans_context_set()
> > >   Used in xfs_trans_alloc()
> > > - xfs_trans_context_clear()
> > >   Used in xfs_trans_free()
> > >
> > > This patch is based on Darrick's work to fix the issue in xfs/141 in the
> > > earlier version. [1]
> > >
> > > 1. https://lore.kernel.org/linux-xfs/20201104001649.GN7123@magnolia
> > >
> > > Cc: Darrick J. Wong <darrick.wong at oracle.com>
> > > Cc: Matthew Wilcox (Oracle) <willy at infradead.org>
> > > Cc: Christoph Hellwig <hch at lst.de>
> > > Cc: Dave Chinner <david at fromorbit.com>
> > > Signed-off-by: Yafang Shao <laoar.shao at gmail.com>
> > > ---
> > >  fs/xfs/xfs_trans.c | 20 +++++++-------------
> > >  1 file changed, 7 insertions(+), 13 deletions(-)
> > >
> > > diff --git a/fs/xfs/xfs_trans.c b/fs/xfs/xfs_trans.c
> > > index 11d390f0d3f2..fe20398a214e 100644
> > > --- a/fs/xfs/xfs_trans.c
> > > +++ b/fs/xfs/xfs_trans.c
> > > @@ -67,6 +67,9 @@ xfs_trans_free(
> > >       xfs_extent_busy_sort(&tp->t_busy);
> > >       xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
> > >
> > > +     /* Detach the transaction from this thread. */
> > > +     xfs_trans_context_clear(tp);
> >
> > Don't you need to check if tp is still the current transaction before
> > you clear PF_MEMALLOC_NOFS, now that the NOFS is bound to the lifespan
> > of the transaction itself instead of the reservation?
> >
> 
> The current->journal_info is always the same with tp here in my verification.
> I don't know in which case they are different.

I don't know why you changed it from the previous version.

> It would be better if you could explain in detail.  Anyway I can add
> the check with your comment in the next version.

xfs_trans_alloc is called to allocate a transaction.  We set _NOFS and
save the old flags (which don't contain _NOFS) to this transaction.

thread logs some changes and calls xfs_trans_roll.

xfs_trans_roll calls xfs_trans_dup to duplicate the old transaction.

xfs_trans_dup allocates a new transaction, which sets PF_MEMALLOC_NOFS
and saves the current context flags (in which _NOFS is set) in the new
transaction.

xfs_trans_roll then commits the old transaction

xfs_trans_commit frees the old transaction

xfs_trans_free restores the old context (which didn't have _NOFS) and
now we've dropped NOFS incorrectly

now we move on with the new transaction, but in the wrong NOFS mode.

note that this becomes a lot more obvious once you start fiddling with
current->journal_info in the last patch.

--D

> 
> >
> > > +
> > >       trace_xfs_trans_free(tp, _RET_IP_);
> > >       if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
> > >               sb_end_intwrite(tp->t_mountp->m_super);
> > > @@ -153,9 +156,6 @@ xfs_trans_reserve(
> > >       int                     error = 0;
> > >       bool                    rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
> > >
> > > -     /* Mark this thread as being in a transaction */
> > > -     xfs_trans_context_set(tp);
> > > -
> > >       /*
> > >        * Attempt to reserve the needed disk blocks by decrementing
> > >        * the number needed from the number available.  This will
> > > @@ -163,10 +163,9 @@ xfs_trans_reserve(
> > >        */
> > >       if (blocks > 0) {
> > >               error = xfs_mod_fdblocks(mp, -((int64_t)blocks), rsvd);
> > > -             if (error != 0) {
> > > -                     xfs_trans_context_clear(tp);
> > > +             if (error != 0)
> > >                       return -ENOSPC;
> > > -             }
> > > +
> > >               tp->t_blk_res += blocks;
> > >       }
> > >
> > > @@ -241,8 +240,6 @@ xfs_trans_reserve(
> > >               tp->t_blk_res = 0;
> > >       }
> > >
> > > -     xfs_trans_context_clear(tp);
> > > -
> > >       return error;
> > >  }
> > >
> > > @@ -284,6 +281,8 @@ xfs_trans_alloc(
> > >       INIT_LIST_HEAD(&tp->t_dfops);
> > >       tp->t_firstblock = NULLFSBLOCK;
> > >
> > > +     /* Mark this thread as being in a transaction */
> > > +     xfs_trans_context_set(tp);
> > >       error = xfs_trans_reserve(tp, resp, blocks, rtextents);
> > >       if (error) {
> > >               xfs_trans_cancel(tp);
> > > @@ -878,7 +877,6 @@ __xfs_trans_commit(
> > >
> > >       xfs_log_commit_cil(mp, tp, &commit_lsn, regrant);
> > >
> > > -     xfs_trans_context_clear(tp);
> > >       xfs_trans_free(tp);
> > >
> > >       /*
> > > @@ -911,7 +909,6 @@ __xfs_trans_commit(
> > >               tp->t_ticket = NULL;
> > >       }
> > >
> > > -     xfs_trans_context_clear(tp);
> > >       xfs_trans_free_items(tp, !!error);
> > >       xfs_trans_free(tp);
> > >
> > > @@ -971,9 +968,6 @@ xfs_trans_cancel(
> > >               tp->t_ticket = NULL;
> > >       }
> > >
> > > -     /* mark this thread as no longer being in a transaction */
> > > -     xfs_trans_context_clear(tp);
> > > -
> > >       xfs_trans_free_items(tp, dirty);
> > >       xfs_trans_free(tp);
> > >  }
> > > --
> > > 2.18.4
> > >
> 
> 
> 
> -- 
> Thanks
> Yafang