xen-blkfront: don't call talk_to_blkback when already connected to blkback
authorBob Liu <bob.liu@oracle.com>
Tue, 7 Jun 2016 14:43:15 +0000 (10:43 -0400)
committerKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Wed, 8 Jun 2016 17:54:39 +0000 (13:54 -0400)
Sometimes blkfront may twice receive blkback_changed() notification
(XenbusStateConnected) after migration, which will cause
talk_to_blkback() to be called twice too and confuse xen-blkback.

The flow is as follow:
   blkfront                                        blkback
blkfront_resume()
 > talk_to_blkback()
  > Set blkfront to XenbusStateInitialised
                                                front changed()
                                                 > Connect()
                                                  > Set blkback to XenbusStateConnected

blkback_changed()
 > Skip talk_to_blkback()
   because frontstate == XenbusStateInitialised
 > blkfront_connect()
  > Set blkfront to XenbusStateConnected

-----
And here we get another XenbusStateConnected notification leading
to:
-----
blkback_changed()
 > because now frontstate != XenbusStateInitialised
   talk_to_blkback() is also called again
  > blkfront state changed from
  XenbusStateConnected to XenbusStateInitialised
    (Which is not correct!)

front_changed():
                                                 > Do nothing because blkback
                                                   already in XenbusStateConnected

Now blkback is in XenbusStateConnected but blkfront is still
in XenbusStateInitialised - leading to no disks.

Poking of the XenbusStateConnected state is allowed (to deal with
block disk change) and has to be dealt with. The most likely
cause of this bug are custom udev scripts hooking up the disks
and then validating the size.

Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
drivers/block/xen-blkfront.c

index ca13df8..6ba8891 100644 (file)
@@ -2485,10 +2485,23 @@ static void blkback_changed(struct xenbus_device *dev,
                break;
 
        case XenbusStateConnected:
-               if (dev->state != XenbusStateInitialised) {
+               /*
+                * talk_to_blkback sets state to XenbusStateInitialised
+                * and blkfront_connect sets it to XenbusStateConnected
+                * (if connection went OK).
+                *
+                * If the backend (or toolstack) decides to poke at backend
+                * state (and re-trigger the watch by setting the state repeatedly
+                * to XenbusStateConnected (4)) we need to deal with this.
+                * This is allowed as this is used to communicate to the guest
+                * that the size of disk has changed!
+                */
+               if ((dev->state != XenbusStateInitialised) &&
+                   (dev->state != XenbusStateConnected)) {
                        if (talk_to_blkback(dev, info))
                                break;
                }
+
                blkfront_connect(info);
                break;