Em Sexta, 12 de Outubro de 2007 20:56, Steinar H. Gunderson escreveu: > On Thu, Oct 11, 2007 at 12:10:04PM +0100, Pedro Celestino dos Reis Rodrigues wrote: > > Automounted homes over nfs4 with sec=krb5 > > Start a session on the nfs client and let the user ticked (used to grant > > the nfs access) expire > > > > After this the user can not access is home (as should be expected) > > However, it also appens that, for example, any trial to start a new > > session of some other user will sucessfully mount the respective home but > > the login process will stay forever trying to read a file (.bash_profile > > for example). > > The client syslog presents messages like these > > So the real problem is that the new session doesn't refresh your user > credentials? That doesn't sound like an NFS bug to me. > Hello Thank you for your prompt answer Steinar
The reason I suspect that the problem is with NFS is that the problem comes out with users (designated by B) other than the user that have is ticket expired(designated by A). That is to say, a user B, with a fresh and valid ticket, can not access is own home at the nfs mount until the user A (that have is home mounted at same nfs server and is ticket was expired) does not refresh is (the user A) ticket. For me, it seems that a lock condition is generated, somewere in the client machine NFS code, when the opened files of user A become unreachable due to ticket expiration. The lock condition does not interfere with mount and unmount (made by the automounter or other mean) for the home of any user B on the nfs server. Read of directory entries on the user B home is also possible showing that the user B have valid credentials. Only user B processes reading and writing on files (or writing directory entries) hang on a locked state. After user A refreshes is ticket he can access the opened files again and the NFS lock condition is removed, letting the locked processes of the user B proceed on the read (or write) operation. Until now, I failed to figure out a subsystem other than NFS that can be responsible for this behaviour. However, I am only guessing becose I have not traced the code execution, so, if you have reasons to assign the bug to other package please let me know. Thanks again Pedro -- _____________________________________________________________ Pedro Celestino dos Reis Rodrigues Departamento de Química e Bioquímica Faculdade de Ciências da Universidade de Lisboa Tel: 21750000-28619
