On 7/8/2010 9:31 AM, Matt Renzelmann wrote: > Hello, > > I've observed the following issue with OpenAFS. Platform is Windows 7 > x64 "Ultimate" with all the latest Windows Update patches. The behavior > occurs with the last three stable releases of OpenAFS recommended for > Windows: 1.5.75, 1.5.74, and 1.5.73. Using Network Identity Manager > 2.0.0.304 per Help -> About – the latest. > > Details of the behavior: > > - If I disable and then reenable the main network adapter--the one that > AFS is ultimately using to access my AFS data--I observe that windows > Explorer gets "stuck." It appears to be stuck in some kind of busy > live-lock state. > > - I suspect that if I lose my Internet connection on the same adapter > for any reason, I get a similar symptom, but I've not confirmed this. > > - Attempting to terminate the explorer process once it's in this state > fails. It will not terminate. Task Manager and Process Explorer + > administrative escalation is not sufficient. > > - All applications that use Explorer functionality, e.g. file open/save > windows, will hang as soon as they invoke said functionality. > > - Rebooting resolves the problem, though I often have some difficulty > rebooting cleanly in this scenario.
This is a known issue which can only be fixed by Microsoft. It is documented in the OpenAFS 1.5.75 release notes. In brief, if any network adapter link state changes or the DHCP IP address registration is altered after the OpenAFS SMB Server registers the "AFS <20>" netbios name, the Microsoft SMB redirector will drop the connection to the "AFS <20>" SMB server and will never again be able to resolve the name even though the name is locally registered. The registration can be checked with nbtstat -n and the active netbios connections can be reported with nbtstat -S If you examine the stack (Sysinternals Process Explorer) of any thread that is accessing \\AFS after the failure occurs you will find that it is blocked in the Microsoft SMB redirector (mrxsmb.sys) in a function that is attempting to re-establish a connection to the SMB server. Since the name "AFS <20>" is never found, the connection is never established and the process will block indefinitely. This problem cannot be addressed within OpenAFS except by replacing the AFS SMB server interface with a native kernel file system driver. The work to do so is quite far along. However, the AFS redirector has uncovered other problems in Microsoft Office that prevents its use with non-SMB file system interfaces. Microsoft is working on that problem but has no commitment as to when Office 2007 and 2010 might be patched to correct the failure. > More background: > > - I'm using the DEBUG version of AFS currently in an effort to resolve > this. I've had the problem with 1.5.74/73 using the standard "release" > version. The debug version of OpenAFS will not help and in fact in most cases running the debug version will hurt. Please do not use the debug version unless you are actively debugging a problem with a debugger. The only thing that can be done is to file reports with Microsoft in an effort to get them to escalate the problem and decide it is worth fixing. Several reports have been filed but as yet no one with sufficient clout has made enough noise. Jeffrey Altman
smime.p7s
Description: S/MIME Cryptographic Signature
