On Tue, Apr 16, 2019 at 08:00:59AM -0400, Brian Foster wrote:
> On Mon, Apr 15, 2019 at 06:53:28PM -0700, Darrick J. Wong wrote:
> > Hi Brian,
> >
> > When running fstests against what will become the xfs-5.2-merge tree, I
> > hit an assert in xfs/104; the dmesg for that is pasted below.
> >
> > I'm pretty sure that what's going on here is that we hit a deferred AGFL
> > free while growing the filesystem and need to do a defer_roll, but
> > tr_growdata is not one of the permanent reservation transactions and
> > that's why we trip some asserts.
> >
>
> Yeah, it looks like we can hit this in the case where a growfs extends
> the last pre-existing AG in the fs. We "free" the additional space into
> the AG, happen to come across an overpopulated AGFL and end up doing
> deferred ops from a non-permanent transaction.
>
> I think we should probably just update the tr_growdata transaction to
> reflect how it's used and mark it permanent with the default permanent
> log count (2). That obviously increases the reservation size, but it's
> not like grow is a high frequency or performance critical operation.
>
> > FSTYP -- xfs (debug)
> > PLATFORM -- Linux/x86_64 mtr01 5.1.0-rc5-djw
> > MKFS_OPTIONS -- -f -m reflink=1,rmapbt=1, -i sparse=1, -b size=1024, /dev/sdd
> > MOUNT_OPTIONS -- -o usrquota,grpquota,prjquota, /dev/sdd /opt
> >
> > Granted this is the first time I've seen this in several months of
> > nightly fstests runs, so chances are this one is hard to reproduce.
> >
>
> I don't recall seeing this and don't reproduce in a few iterations of
> xfs/104 with the above geometry. I'll keep trying for a bit. In the
> meantime, a diff is appended if you'd like to give that a try. Thoughts?
That looks reasonable, though I think we end up playing whackamole
trying to find all the transactions that could result in a deferred agfl
free on a non-permanent transaction type.
Maybe a debugging patch along the lines of:
diff --git a/fs/xfs/libxfs/xfs_alloc.c b/fs/xfs/libxfs/xfs_alloc.c
index 857a53e58b94..e597164641a1 100644
--- a/fs/xfs/libxfs/xfs_alloc.c
+++ b/fs/xfs/libxfs/xfs_alloc.c
@@ -2324,6 +2324,8 @@ xfs_alloc_fix_freelist(
targs.oinfo = XFS_RMAP_OINFO_SKIP_UPDATE;
else
targs.oinfo = XFS_RMAP_OINFO_AG;
+ ASSERT((flags & XFS_ALLOC_FLAG_NOSHRINK) ||
+ (tp->t_flags & XFS_TRANS_PERM_LOG_RES));
while (!(flags & XFS_ALLOC_FLAG_NOSHRINK) && pag->pagf_flcount > need) {
error = xfs_alloc_get_freelist(tp, agbp, &bno, 0);
if (error)
Though this is a blunt instrument since as we've seen with x/104 not all
of the freelist fix calls ever result in shrinking the agfl.
--D
> Brian
>
> --- 8< ---
>
> diff --git a/fs/xfs/libxfs/xfs_trans_resv.c b/fs/xfs/libxfs/xfs_trans_resv.c
> index f99a7aefe418..83f4ee2afc49 100644
> --- a/fs/xfs/libxfs/xfs_trans_resv.c
> +++ b/fs/xfs/libxfs/xfs_trans_resv.c
> @@ -876,9 +876,13 @@ xfs_trans_resv_calc(
> resp->tr_sb.tr_logres = xfs_calc_sb_reservation(mp);
> resp->tr_sb.tr_logcount = XFS_DEFAULT_LOG_COUNT;
>
> + /* growdata requires permanent res; it can free space to the last AG */
> + resp->tr_growdata.tr_logres = xfs_calc_growdata_reservation(mp);
> + resp->tr_growdata.tr_logcount = XFS_DEFAULT_PERM_LOG_COUNT;
> + resp->tr_growdata.tr_logflags |= XFS_TRANS_PERM_LOG_RES;
> +
> /* The following transaction are logged in logical format */
> resp->tr_ichange.tr_logres = xfs_calc_ichange_reservation(mp);
> - resp->tr_growdata.tr_logres = xfs_calc_growdata_reservation(mp);
> resp->tr_fsyncts.tr_logres = xfs_calc_swrite_reservation(mp);
> resp->tr_writeid.tr_logres = xfs_calc_writeid_reservation(mp);
> resp->tr_attrsetrt.tr_logres = xfs_calc_attrsetrt_reservation(mp);
