Bug 448038

+++ This bug was initially created as a clone of Bug #442618 +++

Description of problem:
After some amount of time using /net for builds, the following shows up in the
log (after enabling debugging, because the mount points wouldn't expire):

Apr 15 15:29:12 xxxx automount[1932]: st_expire: state 1 path /net
Apr 15 15:29:12 xxxx automount[1932]: expire_proc: exp_proc =
3082759056 path /net
Apr 15 15:29:12 xxxx automount[1932]: expire_proc_indirect: expire
/net/yyyy/usr/src
Apr 15 15:29:12 xxxx automount[1932]: expire_indirect: fstat failed:
Bad file descriptor
Apr 15 15:29:12 xxxx automount[1932]: expire_proc_indirect: expire
/net/yyyy/usr/src
Apr 15 15:29:12 xxxx automount[1932]: expire_indirect: fstat failed:
Bad file descriptor
Apr 15 15:29:12 xxxx automount[1932]: expire_proc_indirect: expire
/net/yyyy
Apr 15 15:29:12 xxxx automount[1932]: 3 remaining in /net
Apr 15 15:29:12 xxxx automount[1932]: expire_cleanup: got thid
3082759056 path /net stat 4
Apr 15 15:29:12 xxxx automount[1932]: expire_cleanup: sigchld: exp
3082759056 finished, switching from 2 to 1

Version-Release number of selected component (if applicable):
Linux automount version 5.0.3-11

Directories:
       config dir:     /etc/sysconfig
       maps dir:       /etc
       modules dir:    /usr/lib/autofs

Compile options:
 DISABLE_MOUNT_LOCKING ENABLE_IGNORE_BUSY_MOUNTS WITH_HESIOD WITH_LDAP
 WITH_SASL

kernel 2.6.25-0.220

How reproducible:
Not sure.

-- Additional comment from ikent on 2008-04-16 02:21 EST --
I may be able to reproduce it given the export lists, as
long as there aren't too many servers.

-- Additional comment from staubach on 2008-04-16 09:32 EST --
The exports on the server look like:

/               *(ro,sync,fsid=0)
/usr/src        *(rw,sync,nohide,no_root_squash)

I am working in a directory inside of that /usr/src file system.
(/ is exported so that I can get NFSv4 working in a similar
fashion to NFSv2 and NFSv3.  I am not using NFSv4 here though.)

-- Additional comment from ikent on 2008-04-19 00:05 EST --
Just to let you know, I haven't forgotten this.

The cause of the descriptor messages is due to autofs
interaction with the "nohide" option. I've been expecting
something like this for some time and have thought about
how to deal with it on and off for a while.

What happens is that, since this mount shows up in the exports
list autofs thinks it is something it needs to automount but
because it is a "nohide" export the NFS client also mounts it
transparently. This confuses autofs as it then thinks it needs
to umount it more than once.

I can't see any way to discover the export options in user
space on the client, any ideas?

Anyway, I'm giving it more serious thought now we have a
reported case.

Ian


-- Additional comment from ikent on 2008-05-12 03:03 EST --
Created an attachment (id=305085)
Patcg to check for nohide mounts


-- Additional comment from ikent on 2008-05-12 03:06 EST --
Could you please check this to see if the patch functions correctly.
The built packages are available at:
http://koji.fedoraproject.org/packages/autofs/5.0.3/14.

-- Additional comment from staubach on 2008-05-12 18:31 EST --
I have installed the new rpm, but I am at Connectathon and it may take
me a few days to test.  I will update this bugzilla as soon as I have
some feedback.

-- Additional comment from staubach on 2008-05-13 18:31 EST --
It seems to work for me.  I accessed the server mentioned in Comment #2.
It appears that autofs caused the / file system to be mounted, but then
the NFS client mounted the /usr/src file system.  It all umounted
cleanly as far as I could tell.

-- Additional comment from staubach on 2008-05-13 18:35 EST --
One other thing, the "nohide" export option is a Linux thing.  Other
systems, such as Solaris, can export the hierarchy described in
Comment #2, without any special export options.

My suggestion is to avoid the mention of the "nohide" export option
because it is server specific.

-- Additional comment from ikent on 2008-05-13 21:57 EST --
(In reply to comment #8)
> One other thing, the "nohide" export option is a Linux thing.  Other
> systems, such as Solaris, can export the hierarchy described in
> Comment #2, without any special export options.
> 
> My suggestion is to avoid the mention of the "nohide" export option
> because it is server specific.

OK, I'll alter the description and internal comments and then
commit it upstream and push updates out.

Heads up, we've had another issue reported in Rawhide which
results in these messages and has similar symptoms. It appears
to be gvfsd that is the causing it but there must be a case
where the mount control file descriptor is incorrectly being
closed. Anyway I'm working on it.

Thanks.

-- Additional comment from fedora-triage-list on 2008-05-14 05:30 EST --
Changing version to '9' as part of upcoming Fedora 9 GA.
More information and reason for this action is here:
http://fedoraproject.org/wiki/BugZappers/HouseKeeping