Mircea Ciocan wrote:

> IMHO, no matter what kerberos key  b..s is happening this should not
> produce this miserable kernel loop that kills the most powerful machines
> available today, it either should have some slower cadence so that
> eventually some could stop the AFS processes or it should give up after
> some time, I this regard I consider this behavior a bug.
> 
> Cheers,
> Mircea

In the good old days when server-A reported "bad kvno" the cache manager
would discard the token even though it was possible that the token was
in fact valid and would work on every other server in the cell.  This
is frequently a problem when a cell contains a mixture of pre-OpenAFS
1.2.8 or Transarc/IBM servers and post-OpenAFS 1.2.8 servers.

A few releases ago a modification was made to prevent the token from
being discarded and in the case of a replicated volume, to permit fail
over to an alternate copy.

I suspect the issue is somehow related to this because now the afs
client will return a bad kvno error to the application instead of
access denied.  If the application repeatedly retries the request
the kernel module will just ask the file servers again and produce the
same result.  If the application asks again, ....

It would be helpful to know if the infinite loop is being driven by
the application or is strictly within the kernel module.

Jeffrey Altman

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

Reply via email to