[Bug 1811795] Re: Mdadm: external metadata arrays and containers are removed incorrectly

[Christian Ehrhardt ]

Thanks for the real test steps and checking them Mariusz.

This will now go to the review by the SRU Team.
Once accepted they will ask for a verification in -proposed (of both affected releases).
Once that is complete this can be released to all users.

** Description changed:

  [Impact]
  
-  * The response to device fails so far was to remove underlying 
-    containers. That failed if the device itself was not stopped
-    before doing so causing issues.
+  * The response to device fails so far was to remove underlying
+    containers. That failed if the device itself was not stopped
+    before doing so causing issues.
  
-  * Fix by backporting upstream fix
+  * Fix by backporting upstream fix
  
  [Test Case]
  
-  * This needs a Intel(R) Matrix Storage Hardware which limits the 
-    systems this can be tested.
-    1. Create IMSM container on 2 SATA disks.
-    2. Create R0 IMSM volume
-    3. Wait for resync to complete.
-    4. Fail one of disks
+  * This needs a Intel(R) Matrix Storage Hardware which limits the
+    systems this can be tested.
+    1. Create IMSM Container:
+        mdadm -CR imsm0 -e imsm -n 2 /dev/sd[bc]
+    2. Create IMSM level 0 Array:
+        mdadm -CR vol0 -l 0 -n 2 /dev/sd[bc]
+    3. Wait for resync to complete
+       mdadm --wait /dev/md/vol0
+    4. Read /proc/mdstat and check /dev/md/ directory
+       cat /proc/mdstat
+       ls /dev/md/
+    5. Fail one disk incrementaly (simulate device disappearing
+       from system)
+       mdadm -If sdc
+    6. Check mdstat file and /dev/md/
+       cat /proc/mdstat
+       ls /dev/md/
+    Result:
+       After disk fail [5] raid 0 array should disappear from mdstat,
+       link /dev/md/imsm0 should exist.
  
-  * these should allow someone who is not familiar with the affected
-    package to reproduce the bug and verify that the updated package fixes
-    the problem.
+ 
+  * these should allow someone who is not familiar with the affected
+    package to reproduce the bug and verify that the updated package fixes
+    the problem.
  
  [Regression Potential]
  
-  * The patch changes the behavior on removing devices. So in that use-case 
-    lies the potential to regress somebody.
-    It adds stopping the device before removing the containers associated 
-    to it (that is the fix). OTOH if there are cases where the kernel 
-    signals a device fail it would formerly not have removed it (the bug) 
-    but there are (very low) chances that somebody relies on that behavior.
-    Since this avoids I/O errors on partially removed device elements I'd 
-    expect taking the change makes it better overall.
+  * The patch changes the behavior on removing devices. So in that use-case
+    lies the potential to regress somebody.
+    It adds stopping the device before removing the containers associated
+    to it (that is the fix). OTOH if there are cases where the kernel
+    signals a device fail it would formerly not have removed it (the bug)
+    but there are (very low) chances that somebody relies on that behavior.
+    Since this avoids I/O errors on partially removed device elements I'd
+    expect taking the change makes it better overall.
  
  [Other Info]
-  
-  * n/a
+ 
+  * n/a
  
  ---
  
  Steps to reproduce:
  
  1. Create IMSM container on 2 SATA disks.
  
  2. Create R0 IMSM volume
  
  3. Wait for resync to complete.
  
  4. Fail one of disks
  
  Expected result:
  R0 Array should disappear form /proc/mdstat.
  
  Actual result:
  R0 array is visible in mdstat, link to container in /dev/md/ is removed.

** Changed in: mdadm (Ubuntu Bionic)
       Status: Incomplete => Fix Committed

** Changed in: mdadm (Ubuntu Cosmic)
       Status: Incomplete => Fix Committed

** Changed in: mdadm (Ubuntu Bionic)
       Status: Fix Committed => In Progress

** Changed in: mdadm (Ubuntu Cosmic)
       Status: Fix Committed => In Progress

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to mdadm in Ubuntu.
https://bugs.launchpad.net/bugs/1811795

Title:
  Mdadm: external metadata arrays and containers are removed incorrectly

Status in mdadm package in Ubuntu:
  Fix Released
Status in mdadm source package in Bionic:
  In Progress
Status in mdadm source package in Cosmic:
  In Progress

Bug description:
  [Impact]

   * The response to device fails so far was to remove underlying
     containers. That failed if the device itself was not stopped
     before doing so causing issues.

   * Fix by backporting upstream fix

  [Test Case]

   * This needs a Intel(R) Matrix Storage Hardware which limits the
     systems this can be tested.
     1. Create IMSM Container:
         mdadm -CR imsm0 -e imsm -n 2 /dev/sd[bc]
     2. Create IMSM level 0 Array:
         mdadm -CR vol0 -l 0 -n 2 /dev/sd[bc]
     3. Wait for resync to complete
        mdadm --wait /dev/md/vol0
     4. Read /proc/mdstat and check /dev/md/ directory
        cat /proc/mdstat
        ls /dev/md/
     5. Fail one disk incrementaly (simulate device disappearing
        from system)
        mdadm -If sdc
     6. Check mdstat file and /dev/md/
        cat /proc/mdstat
        ls /dev/md/
     Result:
        After disk fail [5] raid 0 array should disappear from mdstat,
        link /dev/md/imsm0 should exist.

  
   * these should allow someone who is not familiar with the affected
     package to reproduce the bug and verify that the updated package fixes
     the problem.

  [Regression Potential]

   * The patch changes the behavior on removing devices. So in that use-case
     lies the potential to regress somebody.
     It adds stopping the device before removing the containers associated
     to it (that is the fix). OTOH if there are cases where the kernel
     signals a device fail it would formerly not have removed it (the bug)
     but there are (very low) chances that somebody relies on that behavior.
     Since this avoids I/O errors on partially removed device elements I'd
     expect taking the change makes it better overall.

  [Other Info]

   * n/a

  ---

  Steps to reproduce:

  1. Create IMSM container on 2 SATA disks.

  2. Create R0 IMSM volume

  3. Wait for resync to complete.

  4. Fail one of disks

  Expected result:
  R0 Array should disappear form /proc/mdstat.

  Actual result:
  R0 array is visible in mdstat, link to container in /dev/md/ is removed.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/mdadm/+bug/1811795/+subscriptions