struct o2hb_bio_wait_ctxt {
        atomic_t          wc_num_reqs;
        struct completion wc_io_complete;
+       int               wc_error;
 };
 
 static void o2hb_write_timeout(void *arg)
 {
        atomic_set(&wc->wc_num_reqs, num_ios);
        init_completion(&wc->wc_io_complete);
+       wc->wc_error = 0;
 }
 
 /* Used in error paths too */
 {
        struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
 
-       if (error)
+       if (error) {
                mlog(ML_ERROR, "IO Error %d\n", error);
+               wc->wc_error = error;
+       }
 
        if (bio->bi_size)
                return 1;
 
 bail_and_wait:
        o2hb_wait_on_io(reg, &wc);
+       if (wc.wc_error && !status)
+               status = wc.wc_error;
 
        if (bios) {
                for(i = 0; i < num_bios; i++)
        return highest;
 }
 
-static void o2hb_do_disk_heartbeat(struct o2hb_region *reg)
+static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
 {
        int i, ret, highest_node, change = 0;
        unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
        struct bio *write_bio;
        struct o2hb_bio_wait_ctxt write_wc;
 
-       if (o2nm_configured_node_map(configured_nodes, sizeof(configured_nodes)))
-               return;
+       ret = o2nm_configured_node_map(configured_nodes,
+                                      sizeof(configured_nodes));
+       if (ret) {
+               mlog_errno(ret);
+               return ret;
+       }
 
        highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
        if (highest_node >= O2NM_MAX_NODES) {
                mlog(ML_NOTICE, "ocfs2_heartbeat: no configured nodes found!\n");
-               return;
+               return -EINVAL;
        }
 
        /* No sense in reading the slots of nodes that don't exist
        ret = o2hb_read_slots(reg, highest_node + 1);
        if (ret < 0) {
                mlog_errno(ret);
-               return;
+               return ret;
        }
 
        /* With an up to date view of the slots, we can check that no
        ret = o2hb_issue_node_write(reg, &write_bio, &write_wc);
        if (ret < 0) {
                mlog_errno(ret);
-               return;
+               return ret;
        }
 
        i = -1;
         */
        o2hb_wait_on_io(reg, &write_wc);
        bio_put(write_bio);
+       if (write_wc.wc_error) {
+               /* Do not re-arm the write timeout on I/O error - we
+                * can't be sure that the new block ever made it to
+                * disk */
+               mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
+                    write_wc.wc_error, reg->hr_dev_name);
+               return write_wc.wc_error;
+       }
+
        o2hb_arm_write_timeout(reg);
 
        /* let the person who launched us know when things are steady */
                if (atomic_dec_and_test(®->hr_steady_iterations))
                        wake_up(&o2hb_steady_queue);
        }
+
+       return 0;
 }
 
 /* Subtract b from a, storing the result in a. a *must* have a larger
                 * likely to time itself out. */
                do_gettimeofday(&before_hb);
 
-               o2hb_do_disk_heartbeat(reg);
+               i = 0;
+               do {
+                       ret = o2hb_do_disk_heartbeat(reg);
+               } while (ret && ++i < 2);
 
                do_gettimeofday(&after_hb);
                elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);