md: remove check for MD_RECOVERY_NEEDED in action_store.
authorNeilBrown <neilb@suse.com>
Mon, 21 Dec 2015 00:01:21 +0000 (11:01 +1100)
committerNeilBrown <neilb@suse.com>
Mon, 21 Dec 2015 00:10:06 +0000 (11:10 +1100)
md currently doesn't allow a 'sync_action' such as 'reshape' to be set
while MD_RECOVERY_NEEDED is set.

This s a problem, particularly since commit 738a273806ee as that can
cause ->check_shape to call mddev_resume() which sets
MD_RECOVERY_NEEDED.  So by the time we come to start 'reshape' it is
very likely that MD_RECOVERY_NEEDED is still set.

Testing for this flag is not really needed and is in any case very
racy as it can be set at any moment - asynchronously.  Any race
between setting a sync_action and setting MD_RECOVERY_NEEDED must
already be handled properly in some locked code, probably
md_check_recovery(), so remove the test here.

The test on MD_RECOVERY_RUNNING is also racy in the 'reshape' case
so we should test it again after getting mddev_lock().

As this fixes a race and a regression which can cause 'reshape' to
fail, it is suitable for -stable kernels since 4.1

Reported-by: Xiao Ni <xni@redhat.com>
Fixes: 738a273806ee ("md/raid5: fix allocation of 'scribble' array.")
Cc: stable@vger.kernel.org (v4.1+)
Signed-off-by: NeilBrown <neilb@suse.com>
drivers/md/md.c

index dbedc58d8c00fdcdacb5a9e24d3a5c8571b988a8..61aacab424cf1860dbb5651f5a4fc6c9f7188027 100644 (file)
@@ -4326,8 +4326,7 @@ action_store(struct mddev *mddev, const char *page, size_t len)
                        }
                        mddev_unlock(mddev);
                }
-       } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
-                  test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
+       } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
                return -EBUSY;
        else if (cmd_match(page, "resync"))
                clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
@@ -4340,8 +4339,12 @@ action_store(struct mddev *mddev, const char *page, size_t len)
                        return -EINVAL;
                err = mddev_lock(mddev);
                if (!err) {
-                       clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
-                       err = mddev->pers->start_reshape(mddev);
+                       if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
+                               err =  -EBUSY;
+                       else {
+                               clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
+                               err = mddev->pers->start_reshape(mddev);
+                       }
                        mddev_unlock(mddev);
                }
                if (err)