Avoid panic in case of pool errors and missing L2ARC

In case an ARC buffer is allocated only on L2ARC, and there are
underlying errors in a pool with the cache device in faulty state, a
panic can occur in arc_read_done()->arc_hdr_destroy()->
arc_hdr_l2arc_destroy()->arc_hdr_clear_flags() when trying to free
the ARC buffer.

Fix this by discarding the buffer's identity in arc_hdr_destroy(), in
case the buffer is not empty, before calling arc_hdr_l2hdr_destroy().

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: George Amanakis <gamanakis@gmail.com>
Closes #12392

module/zfs/arc.c

@@ -3775,8 +3775,13 @@ arc_hdr_destroy(arc_buf_hdr_t *hdr)
 		 * to acquire the l2ad_mtx. If that happens, we don't
 		 * want to re-destroy the header's L2 portion.
 		 */
-		if (HDR_HAS_L2HDR(hdr))
+		if (HDR_HAS_L2HDR(hdr)) {
+
+			if (!HDR_EMPTY(hdr))
+				buf_discard_identity(hdr);
+
 			arc_hdr_l2hdr_destroy(hdr);
+		}
 
 		if (!buflist_held)
 			mutex_exit(&dev->l2ad_mtx);