+++ This bug was initially created as a clone of Bug #577824 +++

+++ This bug was initially created as a clone of Bug #576145 +++

The following was filed automatically by anaconda:
anaconda 13.21.20.1 exception report
Traceback (most recent call first):
  File "/usr/lib64/python2.6/site-packages/parted/disk.py", line 213, in commit
    return self.__disk.commit()
  File "/usr/lib64/python2.6/site-packages/parted/decorators.py", line 30, in localeC
    ret = fn(*args, **kwds)
  File "<string>", line 2, in commit
  File "/usr/lib/anaconda/storage/formats/disklabel.py", line 245, in commit
    self.partedDisk.commit()
  File "/usr/lib/anaconda/storage/formats/disklabel.py", line 235, in destroy
    self.commit()
  File "/usr/lib/anaconda/storage/deviceaction.py", line 313, in execute
    self.origFormat.destroy()
  File "/usr/lib/anaconda/storage/devicetree.py", line 670, in processActions
    action.execute(intf=self.intf)
  File "/usr/lib/anaconda/storage/__init__.py", line 287, in doIt
    self.devicetree.processActions()
  File "/usr/lib/anaconda/packages.py", line 110, in turnOnFilesystems
    anaconda.id.storage.doIt()
  File "/usr/lib/anaconda/dispatch.py", line 205, in moveStep
    rc = stepFunc(self.anaconda)
  File "/usr/lib/anaconda/dispatch.py", line 126, in gotoNext
    self.moveStep()
  File "/usr/lib/anaconda/gui.py", line 1396, in setScreen
    self.anaconda.dispatch.gotoNext()
  File "/usr/lib/anaconda/gui.py", line 1309, in nextClicked
    self.setScreen ()
IOException: Error opening /dev/dm-1: No such file or directory

<snip>

--- Additional comment from hdegoede on 2010-03-29 07:41:15 EDT ---

There are 2 issues here:

1) We should not call partedDisk.commit on non partitionable devices which
   hold a disklabel, as the partition table reload ioctl will fail.

2) We should not store /dev/dm-# device node names in actions, but
   rather /dev/mapper/xxxxx-yyyyy names, as the /dev/dm-# name can change
   when we deactivate and re-activate things.

I'm going to clone this bug, and keep this one for issue 1 (for which I'll send a patch to the list for review).

I'm going to assign issue 2 to Dave Lehman who knows that part of the code
best.

--- Additional comment from hdegoede on 2010-03-29 07:44:45 EDT ---

Dave,

Quoting the last comment from the cloned bug:

"2) We should not store /dev/dm-# device node names in actions, but
    rather /dev/mapper/xxxxx-yyyyy names, as the /dev/dm-# name can change
    when we deactivate and re-activate things."

Regards,

Hans

--- Additional comment from dlehman on 2010-03-29 12:15:05 EDT ---

We aren't storing dm-# anywhere. Parted is reporting dm-# because it has resolved the /dev/mapper/%name symlink to the real device name which is dm-#.

--- Additional comment from hdegoede on 2010-03-29 13:50:16 EDT ---

(In reply to comment #4)
> We aren't storing dm-# anywhere. Parted is reporting dm-# because it has
> resolved the /dev/mapper/%name symlink to the real device name which is dm-#.    

You are probably right, but even then we still have an issue, parted is returning the following error:
IOException: Error opening /dev/dm-1: No such file or directory

This means that the symlink is pointing to a non existing device node ?

Maybe some issue with the new udev driven lvm stuff ?

Note that the:

        if not os.access(self.device, os.W_OK):
            raise DeviceFormatError("device path does not exist")

check in disklabel.py does not trigger, maybe that is only checking the
existence of the symlink ?

Re-opening this so that it does not fall through the cracks (sorry).

--- Additional comment from akozumpl on 2010-04-01 09:54:30 EDT ---

One thing is for sure: according to the traceback we give parted the correct name of the device:

DiskLabel.destroy: device: /dev/mapper/vg00-lvRHEL5 ; status: True ; type: disklabel ;
12

--- Additional comment from akozumpl on 2010-04-02 08:50:43 EDT ---

I have a reproducing vmware image and there's actually no symlinks pointing to /dev/dm-2. So I am still wondering.

--- Additional comment from akozumpl on 2010-04-06 05:13:40 EDT ---

Here's whats happening: (it is only reproducible without Hans's patch for 576145, but it really can be an issue now that /dev/mapper/VolGrou00-* are udev manged symlinks)

1) during storageinit we initialize DiskLabel's parted devices through a call to pyparted:
self._partedDevice = parted.Device(path=self.device)

where self.device is '/dev/mapper/VolGroup00-lve1' links to '/dev/dm-2'

2) pyparted talks to libparted, it calls ped_device_get(path)
the first think ped_device_get does is to canonicalize the path. from then on it works with and remembers only the absolute path.

3) during the enablefilesystems step, at the point where we try to destroy disklabel on /dev/mapper/VolGroup00-lve1, /dev/mapper/VolGroup00-lve1 now points to /dev/dm-0 as the other lvm devices were deactivated/disappeared in the meantime (is this the correct behavior?) but /dev/dm-2 that libparted remembers no longer exists.

The fix should be to reinitialize disklabel._partedDevice whenever the given lvm is reactivated to keep the internal libparted object consistent with the disklabel object.

--- Additional comment from hdegoede on 2010-04-06 11:40:05 EDT ---

This is fixed in parted-2.1-6.el6, moving to modified.