On Mon, Mar 16, 2020 at 02:32:23PM -0700, Darrick J. Wong wrote:
> On Mon, Mar 16, 2020 at 01:00:32PM -0400, Brian Foster wrote:
> > AIL removal of the quotaoff start intent and free of both quotaoff
> > intents is currently limited to the ->iop_committed() handler of the
> > end intent. This executes when the end intent is committed to the
> > on-disk log and marks the completion of the operation. The problem
> > with this is it assumes the success of the operation. If a shutdown
> > or other error occurs during the quotaoff, it's possible for the
> > quotaoff task to exit without removing the start intent from the
> > AIL. This results in an unmount hang as the AIL cannot be emptied.
> > Further, no other codepath frees the intents and so this is also a
> > memory leak vector.
>
> And I'm guessing that you'd rather we taught the quota items to be
> self-releasing under error rather than making the quotaoff code be smart
> enough to free the quotaoff-start item?
>
It's a combination of both because they are separate transactions... If
the item is "owned" by a transaction and the transaction aborts, we
"release" all attached items as we would for any other transaction/item.
Once the start transaction commits, the start item is AIL resident and
there's no guarantee we'll ever get to the end transaction (i.e. a
shutdown could cause the transaction allocation to fail). The quotaoff
code handles cleaning up the start item in that scenario. This is
similar to the whole EFD holding a reference to the EFI model, except
quotaoff is a bit more hacky..
>From a development perspective, my approach was to fix up the intent
handling such that adheres to common practice wrt to transactions and
then address the gap(s) in the quotaoff code (i.e. the separate
start/end transactions being an implementation detail of quotaoff).
> > First, update the high level quotaoff error path to directly remove
> > and free the quotaoff start intent if it still exists in the AIL at
> > the time of the error. Next, update both of the start and end
> > quotaoff intents with an ->iop_release() callback to properly handle
> > transaction abort.
>
> I wonder, does this mean that we can drop the if (->io_release) check in
> xfs_trans_free_items? ISTR we were wondering at one point if there ever
> was a real use case for items that don't have a release function.
>
Hmm.. this was reworked fairly recently IIRC to condense some of the
callbacks. This used to be an ->iop_unlock() call. It was renamed to
->iop_release() to primarily handle the abort scenario (yet can also be
called via log I/O completion based on a magic flag) and the non-abort
->iop_unlock() call was folded into ->iop_committing(). Clear as mud? :)
In any event, I'm not aware of any further effort to remove
->iop_release() from xfs_trans_free_items() as that is the typical abort
path that this patch relies on..
Brian
> > This means that If the quotaoff start transaction aborts, it frees
> > the start intent in the transaction commit path. If the filesystem
> > shuts down before the end transaction allocates, the quotaoff
> > sequence removes and frees the start intent. If the end transaction
> > aborts, it removes the start intent and frees both. This ensures
> > that a shutdown does not result in a hung unmount and that memory is
> > not leaked regardless of when a quotaoff error occurs.
> >
> > Signed-off-by: Brian Foster <bfoster@xxxxxxxxxx>
>
> FWIW, the code looks reasonable.
>
> --D
>
> > ---
> > fs/xfs/xfs_dquot_item.c | 15 +++++++++++++++
> > fs/xfs/xfs_qm_syscalls.c | 13 +++++++------
> > 2 files changed, 22 insertions(+), 6 deletions(-)
> >
> > diff --git a/fs/xfs/xfs_dquot_item.c b/fs/xfs/xfs_dquot_item.c
> > index 2b816e9b4465..cf65e2e43c6e 100644
> > --- a/fs/xfs/xfs_dquot_item.c
> > +++ b/fs/xfs/xfs_dquot_item.c
> > @@ -315,17 +315,32 @@ xfs_qm_qoffend_logitem_committed(
> > return (xfs_lsn_t)-1;
> > }
> >
> > +STATIC void
> > +xfs_qm_qoff_logitem_release(
> > + struct xfs_log_item *lip)
> > +{
> > + struct xfs_qoff_logitem *qoff = QOFF_ITEM(lip);
> > +
> > + if (test_bit(XFS_LI_ABORTED, &lip->li_flags)) {
> > + if (qoff->qql_start_lip)
> > + xfs_qm_qoff_logitem_relse(qoff->qql_start_lip);
> > + xfs_qm_qoff_logitem_relse(qoff);
> > + }
> > +}
> > +
> > static const struct xfs_item_ops xfs_qm_qoffend_logitem_ops = {
> > .iop_size = xfs_qm_qoff_logitem_size,
> > .iop_format = xfs_qm_qoff_logitem_format,
> > .iop_committed = xfs_qm_qoffend_logitem_committed,
> > .iop_push = xfs_qm_qoff_logitem_push,
> > + .iop_release = xfs_qm_qoff_logitem_release,
> > };
> >
> > static const struct xfs_item_ops xfs_qm_qoff_logitem_ops = {
> > .iop_size = xfs_qm_qoff_logitem_size,
> > .iop_format = xfs_qm_qoff_logitem_format,
> > .iop_push = xfs_qm_qoff_logitem_push,
> > + .iop_release = xfs_qm_qoff_logitem_release,
> > };
> >
> > /*
> > diff --git a/fs/xfs/xfs_qm_syscalls.c b/fs/xfs/xfs_qm_syscalls.c
> > index 1ea82764bf89..5d5ac65aa1cc 100644
> > --- a/fs/xfs/xfs_qm_syscalls.c
> > +++ b/fs/xfs/xfs_qm_syscalls.c
> > @@ -29,8 +29,6 @@ xfs_qm_log_quotaoff(
> > int error;
> > struct xfs_qoff_logitem *qoffi;
> >
> > - *qoffstartp = NULL;
> > -
> > error = xfs_trans_alloc(mp, &M_RES(mp)->tr_qm_quotaoff, 0, 0, 0, &tp);
> > if (error)
> > goto out;
> > @@ -62,7 +60,7 @@ xfs_qm_log_quotaoff(
> > STATIC int
> > xfs_qm_log_quotaoff_end(
> > struct xfs_mount *mp,
> > - struct xfs_qoff_logitem *startqoff,
> > + struct xfs_qoff_logitem **startqoff,
> > uint flags)
> > {
> > struct xfs_trans *tp;
> > @@ -73,9 +71,10 @@ xfs_qm_log_quotaoff_end(
> > if (error)
> > return error;
> >
> > - qoffi = xfs_trans_get_qoff_item(tp, startqoff,
> > + qoffi = xfs_trans_get_qoff_item(tp, *startqoff,
> > flags & XFS_ALL_QUOTA_ACCT);
> > xfs_trans_log_quotaoff_item(tp, qoffi);
> > + *startqoff = NULL;
> >
> > /*
> > * We have to make sure that the transaction is secure on disk before we
> > @@ -103,7 +102,7 @@ xfs_qm_scall_quotaoff(
> > uint dqtype;
> > int error;
> > uint inactivate_flags;
> > - struct xfs_qoff_logitem *qoffstart;
> > + struct xfs_qoff_logitem *qoffstart = NULL;
> >
> > /*
> > * No file system can have quotas enabled on disk but not in core.
> > @@ -228,7 +227,7 @@ xfs_qm_scall_quotaoff(
> > * So, we have QUOTAOFF start and end logitems; the start
> > * logitem won't get overwritten until the end logitem appears...
> > */
> > - error = xfs_qm_log_quotaoff_end(mp, qoffstart, flags);
> > + error = xfs_qm_log_quotaoff_end(mp, &qoffstart, flags);
> > if (error) {
> > /* We're screwed now. Shutdown is the only option. */
> > xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
> > @@ -261,6 +260,8 @@ xfs_qm_scall_quotaoff(
> > }
> >
> > out_unlock:
> > + if (error && qoffstart)
> > + xfs_qm_qoff_logitem_relse(qoffstart);
> > mutex_unlock(&q->qi_quotaofflock);
> > return error;
> > }
> > --
> > 2.21.1
> >
>
