URGENT: 1.3.72 to release this Sunday
Jeffrey Altman <[email protected]>
| Newsgroups | gmane.comp.file-systems.openafs.devel.win32 |
|---|---|
| Organization | Mass |
| Message-ID | <[email protected]> |
Folks; For those who have either not been suffering with unexplained file server crashes or have not been following the events on openafs-info let me summarize what has been going on. Since the beginning of the Fall semester a number of organizations which deploy AFS file servers have been watching them periodically crash. It has been determined that the crashes are an indirect result of a broken behavior in the Windows AFS clients (all versions of OpenAFS since 1.0.4.) The Windows clients had effectively implemented a "use once and discard" policy for AFS RPC connections. For every operation the client would destroy the existing connection and create a new one whenever cryptall was set. This produced an unnecessary load on the servers. More importantly though is the fact that all of the rapid destruction and recreation of RPC connections between the same set of peers has brought a serious thread safety issue to the surface. In all of the existing clients and servers (IBM/Transarc and OpenAFS) there is a failure to properly reference count the rx_connection objects. This allows an rx_connection to be destroyed by one thread while another thread is still using it. This problem has now been fixed in the Windows client cache manager. The Unix cache manager still needs to be fixed. Unfortunately, we have been unable to identify where in the file server the reference counts are failing to be applied. It is hoped that we will be able to track this down sooner rather then later. In the meantime, we believe it is crucial that an updated OpenAFS client for Windows be released and that all organizations distributing the Windows client begin the process of pushing out the update. The 1.3.72 client will fix the use once and discard problem which is tickling the server bug. It will fix the local thread safety issues so that it won't crash. And as a side effect of fixing the bug, the overlapped write issues from Microsoft Office and other high performance applications has been fixed as well. The daily builds are located at: /afs/athena.mit.edu/user/j/a/jaltman/Public/OpenAFS/ \\afs\athena.mit.edu\user\j\a\jaltman\Public\OpenAFS\ http://web.mit.edu/~jaltman/Public/OpenAFS/ Please test ASAP. Thanks. Jeffrey Altman
smime.p7s
(application/x-pkcs7-signature, 3.2 KB) - not displayed