[Bug 480444] Re: packet storm with linux NFSv4 client when calling ftruncate()

[Harry Flink]

I'm not an NFS/Bazaar developer so I'm not familiar with the details but
this bug is pretty serious in every day use when using NFS shares.

User "gpk" above mentioned this has something to do with sent
write/locking requests. Could it be possible to have an easy workaround
to this bug like modifying some write sequences etc? I haven't faced
this same bug with any other software than Bazaar so other applications
access the NFS shares differently and don't trigger this bug.

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to bzr in Ubuntu.
https://bugs.launchpad.net/bugs/480444

Title:
  packet storm with linux NFSv4 client when calling ftruncate()

Status in “bzr” package in Ubuntu:
  Invalid
Status in “linux” package in Ubuntu:
  Triaged
Status in “bzr” source package in Lucid:
  Confirmed
Status in “linux” source package in Lucid:
  Confirmed
Status in “bzr” source package in Oneiric:
  Invalid
Status in “linux” source package in Oneiric:
  Triaged

Bug description:
  linux-image-2.6.31-15-generic 2.6.31-15.49 on amd64 - bug filed
  manually through LP instead of using apport, because the bug in
  question seems to have caused apport to hang while trying to collect
  information.

  I'm recently experiencing a regression when using karmic as an NFSv4
  client with gssapi authentication.  I don't believe this problem was
  present at beta time, but I don't know if this is the result of a
  kernel change or a change in other userspace software on the client
  side.  The server hasn't changed recently, it has an uptime of 54
  days; the server is running a Linux 2.6.30 kernel.

  Share is mounted with options 'sec=krb5i,proto=tcp'.  I've tried
  modifying the mount options by adding 'soft,rsize=1024,wsize=1024';
  this doesn't appear to have any effect except to make network traces
  more readable.

  When using bzr over NFS, I'm getting random unbreakable hangs on the
  client side.  Debugging appears to show that this happens when calling
  ftruncate() (sample size == 1 - very time consuming to reproduce this
  since I have to reboot the clent each time and the client is my
  laptop).  Network traces show a crazy packet storm - 13Mbps in
  constant traffic over the wireless, pushing data at the NFS server.

  Network trace to follow.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/bzr/+bug/480444/+subscriptions