/tmp/autoXXXXXX artifacts in /etc/mtab

/tmp/autoXXXXXX artifacts in /etc/mtab

[Jim Dennis]

 All,

 Running autofs version 5 (versions from RHEL4u7, RHEL5 and the upstream 5.0.4 with almost
 all of the patches applied) we're seeing quite a few issues with stray /tmp/auto* entries being
 left in the /etc/mtab file.

 The symptom is lots of error messages from df commands.

 Looking through the sources I see that autofs version 5 seems to implement its own internal
 mount handling (it's not just spawning process to run the /sbin/mount external command
 for any of the types of mounts for which it has a .so module).  I noticed that there are  three
 different places in the code which use tempfile template strings matching /tmp/autoXXXXXX.

 I haven't get been able to track the issue down further ... but I suspect there's a bug in the
 code that's updating /etc/mtab and which is supposed to be cleaning out these entries.
 We have a horrendously busy and complex automount infrastucture and we've uncovered
 locking issues on /etc/mtab in the past (from the /sbin/mount code as well as during the
 period of time when autofs4 was trying to wrap its own locking around its calls to the external
 mount command.

 I'll try to get a more detailed formal bug report together in the next week or so.  However, I
 figured I should put out an early message in case others are seeing this or already working on it.


--
Jim Dennis,
Senior Staff Linux Engineer, IT
Cadence Design Systems, Inc.

Re: /tmp/autoXXXXXX artifacts in /etc/mtab

[Ian Kent]

Jim Dennis wrote:
> 
>  All,
> 
>  Running autofs version 5 (versions from RHEL4u7, RHEL5 and the upstream 5.0.4 with almost
>  all of the patches applied) we're seeing quite a few issues with stray /tmp/auto* entries being
>  left in the /etc/mtab file.

But what OS are you using for this and what autofs v5 revisions are you
using?

We know RHEL-4 has a broken move mount and a util-linux dependency has
been added in later revisions. RHEL-4 U7 will be a problem with
autofs-5.0.4 without an updated util-linux.

Anyway, you can check.

From the bug for this:

Steps to Reproduce:
1. mkdir /tmp/mnt /tmp/tmp
2. mount <server>:/<path> /tmp/tmp
3. mount --move /tmp/tmp /tmp/mnt

Actual results:
After the steps above /etc/matb has:
<server>:/<path> /tmp/tmp nfs ....
/tmp/tmp /tmp/mnt none ...

but /proc/mounts has:
<server>:/<path> /tmp/mnt nfs ....

Expected results:
/etc/matb should have a single entry
<server>:/<path> /tmp/mnt nfs ....

> 
>  The symptom is lots of error messages from df commands.
> 
>  Looking through the sources I see that autofs version 5 seems to implement its own internal
>  mount handling (it's not just spawning process to run the /sbin/mount external command
>  for any of the types of mounts for which it has a .so module).  I noticed that there are  three
>  different places in the code which use tempfile template strings matching /tmp/autoXXXXXX.
> 
>  I haven't get been able to track the issue down further ... but I suspect there's a bug in the
>  code that's updating /etc/mtab and which is supposed to be cleaning out these entries.
>  We have a horrendously busy and complex automount infrastucture and we've uncovered
>  locking issues on /etc/mtab in the past (from the /sbin/mount code as well as during the
>  period of time when autofs4 was trying to wrap its own locking around its calls to the external
>  mount command.
> 
>  I'll try to get a more detailed formal bug report together in the next week or so.  However, I
>  figured I should put out an early message in case others are seeing this or already working on it.
> 
> 
> --
> Jim Dennis,
> Senior Staff Linux Engineer, IT
> Cadence Design Systems, Inc.
> 
> _______________________________________________
> autofs mailing list
> autofs@linux.kernel.org
> http://linux.kernel.org/mailman/listinfo/autofs