OpenAFS 1.6 client can flood file servers with duplicate NAT ping traffic

[Sergio Gelato]

This issue affects OpenAFS 1.6 series releases prior to 1.6.1pre3. It was reported upstream as, e.g.,
https://rt.central.org/rt/Ticket/Display.html?id=130361
and addressed in a series of patches, the most recent of which appears to be
http://git.openafs.org/?p=openafs.git;a=commit;h=fcdd20e389282607616b23f912a22ff5c352ce70
The relevant keyphrase in patch titles is "nat ping" (case insensitive).

Both 1.6.0~pre1 (oneiric) and 1.6.1~pre1 (currently in precise) are affected. The final 1.6.1 release, now in Debian unstable, contains the relevant fixes.

Clients affected by this issue can cause a significant, unnecessary additional load on fileservers under some circumstances. It would therefore be very desirable to have this bug fixed in stable releases of Ubuntu.

[Launchpad Janitor (janitor)]

Status changed to 'Confirmed' because the bug affects multiple users.

[Anders Kaseorg (andersk)]

I filed bug 984422 for syncing the final 1.6.1 release into precise.

[Scott Kitterman (kitterman)]

This bug was fixed in the package openafs - 1.6.1-1
Sponsored for Anders Kaseorg (anders-kaseorg)

---------------
openafs (1.6.1-1) unstable; urgency=low

  * New upstream release.
    - Do not ignore all InlineBulkStatus errors in file server.
    - Support for Linux 3.3 and 3.4.
    - Fix incorrect kernel error handling in afs_notify_change.
    - Fix locking around RXS_PreparePacket.

 -- Russ Allbery <email address hidden> Wed, 28 Mar 2012 17:25:05 -0700

openafs (1.6.1~pre4-1) unstable; urgency=low

  * New upstream prerelease.
    - Always retry busy rx calls and retriable actions.
    - Do not abort the salvager on large volume IDs.
    - Ensure one NAT ping connection per srvAddr.
    - Improvements to the file server address tracking.
    - Improve locking in the host table in the file server.
    - Avoid discrepancy between lastBusy and PEER_BUSY.
    - Allow a null host for callback breaks.
    - Fix probing for noop_fsync in the Linux kernel.
    - Fix handling of expired locks in the file server.
    - Reset blacklisted servers on a hard-mount retry.
  * Update debhelper compatibility level to V9.
  * Update standards version to 3.9.3 (no changes required).

 -- Russ Allbery <email address hidden> Thu, 08 Mar 2012 18:24:53 -0800

openafs (1.6.1~pre2-1) unstable; urgency=low

  * New upstream prerelease.
    - Disable server-side Rx keepalive during disk I/O.
    - Avoid timing out on non-kernel busy Rx channels.
    - Add separate client connection pool for replicated volumes.
    - Remove SYNC client disable behavior on some error patterns.
    - Correctly test for the end of the Rx call queue.
    - Fix cleanup of special inodes on the server.
    - Provide a public Rx function for toggling keepalive.
    - Fix connection management in the client checkserver loop.
    - Discard cached state in the client when unsure of validity.
    - Fix thread safety of free inode management with demand-attach.
    - Fix xstat handling of time values on 64-bit platforms.
    - Increase the idledead time.
    - Fix return status of fsync on a directory.
  * Fix afs-newcell's creation of the file server process to pass
    -localauth. This was broken when demand-attach support was
    added. Thanks, Florian Hatat. (Closes: #652973)
  * Fix afs-newcell's unwind of file server process creation. Thanks,
    Davor. (Closes: #655136)
  * Description typo, thanks Jonas Smedegaard. (Closes: #657933)
  * Translation updates:
    - Danish, thanks Joe Hansen. (Closes: #654933)

 -- Russ Allbery <email address hidden> Wed, 08 Feb 2012 14:30:07 -0800