Storage Foundation for Windows DMP Device Specific Module returns a bogus path during a path failover operation

Article:TECH75004  |  Created: 2009-01-29  |  Updated: 2014-01-24  |  Article URL http://www.symantec.com/docs/TECH75004
Article Type
Technical Solution

Product(s)

Environment

Issue



Storage Foundation for Windows (SFW) Dynamic Multi-Pathing (DMP) Device Specific Module (DSM) returns a bogus path during a path failover operation

If a path fails or is removed during I/O, DMP DSM may return a bogus path which causes IO to the Volume to fail.


Error



Log Name: System
Source: mpio
Date: DD/MM/YYYY HH:MM:SS
Event ID: 33
Task Category: None
Level: Error
Keywords: Classic
User: N/A
Computer: HOSTNAME
Description:
DSM NAME returned a bogus Path to \Device\MPIODisk5.


Environment



SFW with DMP

  • There are disks within the same array which have a different number of paths.
  • There are disks within the same array which have some paths in common, but not all paths in common.

For example:
* Harddisk1 has 2 paths to the array, denoted 1-0-0, 2-0-0.
* Harddisk2 has 4 paths to the array, denoted 1-0-0, 2-0-0, 1-0-1, 2-0-1.
 

In this case, Harddisk1 and Harddisk2 share two paths, but Harddisk2 also has two paths that are not shared with Harddisk1.


Cause



DMP DSM uses the DsmInvalidatePath API from the Microsoft MPIO framework.  In the event of a path failure, a surviving path is returned without knowledge of which specific disks have a path that is being removed.  If the disks therefore share some paths and not other paths, there is a possibility that a disk may be given a path that does not belong to it.


Solution



In order to avoid this issue, the following conditions should be met:
  • If a disk shares any paths with another disk in the same array, they must share exactly the same paths.
  • If a disk does not share a path with a disk in the same array, then they must not share any paths.

That is, they must either share all paths, or share no paths.  This is known as a share-all or share-nothing configuration.

Exceptions:
 
Arrays that use control devices such as the EMC Symmetrix family of arrays and the Hitachi VSP type arrays can have configurations whereby the control Devices are zoned for many (or all) configured paths to an array; and the data devices are configured to use different subsets of these paths. In these configurations, there is the possibility that during a path failure, MPIO requests an available path from the control device which could result in the "Bogus Path" error being reported for a data device.
 
Example:
 
Control Devices: Paths 1-10
Harddisks1-5: Paths 1-5
Harddisks6-10: Paths 6-10
 
In this scenario, if path 1 fails and MPIO requests an available failover path from a Control device, it can choose any path from paths 2-10. If it chooses path 6, then all disks with failed I/O on path 1 may attempt to move to path 6. In this case, since Harddisks 1-5 do NOT have path 6 available, this could result in failed I/O and the "Bogus Path" message being logged.
 
There have been changes made in the vhdsaa (Hitachi AA) and vemcsymm (Symmetrix) DSMs to allow these configurations to work by ignoring the control device for path selection.

For vemcsymm:  This fix has been included in the latest DDI packages for 5.1 Sp1 and all later versions.  Please see the related articles for details on how to find the latest DDI packages.

For vhdsaa: A private fix has been created which addresses this problem for version 6.0 and higher.  Please contact support and a support representative will further assist in obtaining the private fix. For a complete list of Symantec Enterprise Technical Support contact numbers, go to http://www.symantec.com/enterprise/support/index.jsp

Disclaimer: This fix specifically addresses the problem identified above. It has not been fully tested and should be applied in a test environment before placing into production. If the systems are not critically impaired, it is recommended to delay the installation of this private fix until the next scheduled maintenance release. Before applying this private fix, systems may be required to be upgraded to the latest code base. The support representative will help in determining the best course of action.

Note:

The above fixes do not remove the requirement for shared-all or shared-none pathing for non-control devices.


Supplemental Materials

SourceEvent ID
Value33
Description

<DSM_name> returned a bogus Path to \Device\MPIODiskX


SourceETrack
Value2774415
Description

During path failure in an Active/Active DMP configuration, 2 luns did not properly move I/O to surviving path and reported 'Bogus Path' errors to the Event Log


SourceETrack
Value3370740
Description

Bogus path returned to DSM due to VSP -CM device path configuration



Legacy ID



333474


Article URL http://www.symantec.com/docs/TECH75004


Terms of use for this information are found in Legal Notices