umount -f /nfsmount hangs

NFS server has gone away, was mounted soft, intr:

bvrgsa.ibm.com:/gsa/bvrgsa on /bvrgsa type nfs (rw,soft,intr,nfsvers=2,tcp,rsize=8192,wsize=8192,timeo=30,addr=9.47.56.70)

but umount -f just hangs ... surely that's not the intended behaviour? from Alt+SysRq+t:

umount S 7FFFFFFF 0 11303 11284 (NOTLB)
c30d7e64 00000082 c35782c0 7fffffff c30d6000 002f961e c292cb50 c35785bc

        0016115b 53a40609 000279b2 c35782c0 c292cb50 c292cd00 c30d7ec0 c031f1c4 
        c35782c0 c30d7ec0 46382f09 c35784a8 c3578408 00000247 c35782c0 c02d0c21 
 Call Trace:
  [schedule_timeout+20/176] schedule_timeout+0x14/0xb0
  [release_sock+81/84] release_sock+0x51/0x54
  [inet_wait_for_connect+136/224] inet_wait_for_connect+0x88/0xe0
  [autoremove_wake_function+0/68] autoremove_wake_function+0x0/0x44
  [autoremove_wake_function+0/68] autoremove_wake_function+0x0/0x44
  [inet_stream_connect+229/384] inet_stream_connect+0xe5/0x180
  [sys_connect+99/128] sys_connect+0x63/0x80
  [do_page_fault+0/1211] do_page_fault+0x0/0x4bb
  [copy_from_user+48/88] copy_from_user+0x30/0x58
  [sys_socketcall+148/520] sys_socketcall+0x94/0x208
  [error_code+45/64] error_code+0x2d/0x40
  [syscall_call+7/11] syscall_call+0x7/0xb

På to , 19/08/2004 klokka 16:55, skreiv Martin J. Bligh:

Works fine for me with 2.6.8.1 and the Fedora Core2 2.6.7-based kernel. If you physically turn off the server as opposed to just killing the nfsd processes, then it takes a bit longer than for the networking layer to time out the sock_release etc (isn't that under the control of the tcp_fin_timeout sysctl?), but AFAICS it does eventually get there.

Are there any other details you're omitting?

Cheers,
Trond

Yeah, it did time out eventually and yes, the network got disconnected rather than killing nfsd. I guess I was expecting -f to mean "Do it. Now" ... However, the damned thing is still mounted as listed by "mount".

Viro pointed me to -l as well ... -f -l seems to work OK.

M.