[Ocfs2-devel] [bug report] ocfs2/cluster: Pin/unpin o2hb regions

[Dan Carpenter via Ocfs2-devel <ocfs2-devel@oss.oracle.com>]

Hello OCFS2 developers,

The patch 58a3158a5d17: "ocfs2/cluster: Pin/unpin o2hb regions" from
Dec 14, 2010, leads to the following Smatch static checker warning:

	./include/linux/fs.h:777 inode_lock()
	warn: sleeping in atomic context

The call tree is:

o2hb_heartbeat_group_drop_item() <- disables preempt
o2hb_region_inc_user() <- disables preempt
-> o2hb_region_pin()
   -> o2nm_depend_item()
      -> configfs_depend_item()
         -> inode_lock()

fs/ocfs2/cluster/heartbeat.c
  2265        static int o2hb_region_pin(const char *region_uuid)
  2266        {
  2267                int ret = 0, found = 0;
  2268                struct o2hb_region *reg;
  2269                char *uuid;
  2270
  2271                assert_spin_locked(&o2hb_live_lock);
                      ^^^^^^^^^^^^^^^^^^

This code is holding a spin lock.

  2272
  2273                list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
  2274                        if (reg->hr_item_dropped)
  2275                                continue;
  2276
  2277                        uuid = config_item_name(&reg->hr_item);
  2278
  2279                        /* local heartbeat */
  2280                        if (region_uuid) {
  2281                                if (strcmp(region_uuid, uuid))
  2282                                        continue;
  2283                                found = 1;
  2284                        }
  2285
  2286                        if (reg->hr_item_pinned || reg->hr_item_dropped)
  2287                                goto skip_pin;
  2288
  2289                        /* Ignore ENOENT only for local hb (userdlm domain) */
  2290                        ret = o2nm_depend_item(&reg->hr_item);
                                    ^^^^^^^^^^^^^^^^
The o2nm_depend_item() fucntion cannot succeed without taking a mutex.

  2291                        if (!ret) {
  2292                                mlog(ML_CLUSTER, "Pin region %s\n", uuid);
  2293                                reg->hr_item_pinned = 1;
  2294                        } else {
  2295                                if (ret == -ENOENT && found)
  2296                                        ret = 0;
  2297                                else {
  2298                                        mlog(ML_ERROR, "Pin region %s fails with %d\n",
  2299                                             uuid, ret);
  2300                                        break;
  2301                                }
  2302                        }
  2303        skip_pin:
  2304                        if (found)
  2305                                break;
  2306                }
  2307
  2308                return ret;
  2309        }

regards,
dan carpenter

_______________________________________________
Ocfs2-devel mailing list
Ocfs2-devel@oss.oracle.com
https://oss.oracle.com/mailman/listinfo/ocfs2-devel