[PATCH] gfs2: No quotas for meta inodes

[Andreas Gruenbacher <agruenba@redhat.com>]

Inodes on the meta filesystem (mount -o meta) can never have quotas, so skip
them entirely in the quota code.  This avoids allocating ip->i_qadata
unnecessarily.

Without this fix, unlinking meta inodes like "statfs" on the meta filesystem
will result in a NULL pointer dereference at unmount time:

  gfs2_evict_inode() -> gfs2_dinode_dealloc() -> gfs2_quota_hold()
    -> qdsb_get() -> slot_get() -> find_first_zero_bit(sd_quota_bitmap)

Unlinking inodes on the meta filesystem isn't a useful operation, but it still
shouldn't cause the kernel to misbehave.

Reported-by: syzbot+cb79de2cc8b76fbf474f@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=cb79de2cc8b76fbf474f
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
---
 fs/gfs2/quota.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/fs/gfs2/quota.c b/fs/gfs2/quota.c
index b62431724eea..dbfc21693dd5 100644
--- a/fs/gfs2/quota.c
+++ b/fs/gfs2/quota.c
@@ -562,6 +562,8 @@ int gfs2_qa_get(struct gfs2_inode *ip)
 
 	if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
 		return 0;
+	if (ip->i_diskflags & GFS2_DIF_SYSTEM)
+		return 0;
 
 	spin_lock(&inode->i_lock);
 	if (ip->i_qadata == NULL) {
@@ -605,7 +607,7 @@ int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
 		return 0;
 
 	error = gfs2_qa_get(ip);
-	if (error)
+	if (error || !ip->i_qadata)
 		return error;
 
 	qd = ip->i_qadata->qa_qd;
-- 
2.55.0