Red Hat Bugzilla – Bug 70967
ftp login hangs while nfs is waiting to reconnect remote offline filesystem
Last modified: 2007-04-18 12:45:18 EDT
From Bugzilla Helper:
User-Agent: Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.0)
Description of problem:
How did it fail? Typing in 'ftp machinename' brought back a login prompt.
Responding to the login prompt with 'username' brought back a password prompt.
Responding with a correct password brought back no prompt at all for several
minutes. (Meanwhile telnet sessions to the server over the LAN could be
started without apparent difficulty).
One day ftp into server works, the next it fails. What has changed? Answer
seems to be, 'power failure in neighbouring office switched off machine hosting
an nfs volume mounted on the server's filesystem'. The nfs volume (running on
RH7.2 itself), was mounted using default options and explicit 'rw'. There was
an outstanding cron job trying to access the mounted filesystem to copy files
to it from the server.
Acting as root, unable to kill -9 pid for any of hung ftp sessions on the
server. Reading 'man' for fstab, mount, nfs, found that a 'hard' nfs mount
which isn't responding is doggedly chased down by the system, to the extent
that related processes 'cannot' be killed. I get the impression that something
is tying itself up in knots trying to re-establish the nfs mount, and is
blocking the ftpd.
The following might or might not be diagnostic: Whenever I reboot this server
I have to go to both this one and the one hosting the nfs volume and
do '/etc/rc.d/init.d/portmap stop', '/etc/rc.d/init.d/nfs stop' and the
corresponding 'start's, to get the mount working again. Whenever I do the nfs
stop, 3 of the 4 items listed are marked as 'failed' rather than 'OK'
The bug is 'failed nfs mount shouldn't preclude ftp login'.
Version-Release number of selected component (if applicable): 2.6.1-18
Steps to Reproduce:
1.Create a working mount of an nfs filesystem on another RH7.2 server
2.Kill power to the remote server
3.Have cron start a file copy job to the remote server (acting as a non-root
4.Try to log in to the ftpd on the local host from another host (used Windows
NT, but don't think this is significant).
5.Log in from other host hangs and ftpd processes are unkillable (even -9).
Actual Results: After typing in password, get endless wait for an ftp prompt.
Expected Results: Get ftp prompt from the server
It would be nice if this feature could be removed. The evidence of the problem
doesn't point out the problem itself very directly. Only when I gave up and
tried to shutdown did I get an idea of what was wrong. Shutdown messages
reporting nfs busy and prevented shutdown completing. Only this alerted me to
explore the status of the remote nfs server.
I see the same problem on a fully patched 7.3 system.
This is no wu-ftpd bug.