Skip to contentSkip to conversation

automount 4.1 / mount issues and possible workarounds

DISCUSSION Anthony Menasse 2 participants 2 messages Feb 23, 2007 – Feb 28, 2007 Kernel automounter

lore
Share & tools
lore
#0 lore raw

Hi,

I was hoping for some feedback on wether the following actions are appropriate to fix the issues we are having with automount /mount. I haven't explained the problems in detail, but from reading the mailing list and other sources they seem like fairly common issues. I just want to make sure my proposed fixes make sense. Im guessing a lot of this has been discussed many times before on this list, but i just wanted some confirmation that i am on the right track (or not). Any advice would be greatly appreciated.

Setup:

A bunch of NFS servers running a mixture of RH Enterprise 4, Enterprise 3 and X/OS linux4 and a bunch of Fedora Core 3 clients running a non Fedora built 2.6.11.7 kernel.

The automount update i am looking at installing is the latest automount package from FC5 (4.1.4 Release 33) rebuilt for FC3 (which seems to work ok after some initial testing on a single box)

I want to avoid updating the kernel at this point.

Not sure about my strategy about updating 'mount', would you recommend getting the latest source or will an older package from FC3 do? In other words are there still serious issues with mount which mean i should be running the latest source or are the issues such as overriding RPC timeout values and locking issues with mtab only an issue with ancient versions of mount?

Issue 1:

nfs: mount failure chop:/export/home/rsp on /home/rsp

I believe this is due to automount using a mount retry value of 0 and dodgy versions of mount overriding the default RPC timeout from 60seconds to 0.7 seconds so the attempt times out too quickly

References:

http://nfsworld.blogspot.com/2005/03/automounter-tuning.html

http://www.mail-archive.com/[email protected]/msg05281.html

Possible Fix:

I want to fix this by using --retry=2 in the auto.master file, and update mount where i can (i want to do both as i can't update mount on every machine/ platform right now)

Issue 2:

When accessing an unmounted automount mount point the automounter will complain that the directory is already mounted (even though it isn't!)

You may get something like this in the log / console.

BUG: <dir> already mounted

I think this is due to a corrupt mtab file due to the race /locking issues with older versions of mount

Possible Fix:

I want to update mount where i can , and link /etc/mtab to /proc/mount . (Are there any issues with having mtab as a link to /proc/mounts?)

Issue 3:

Wierd errors in the logs about attempts to mount files:

Jan 20 18:45:36 glass automount[28315]: mount(nfs): nfs: mount failure 

hassan:/export/home/rsp/.sitecshrc on /home/rsp/.sitecshrc

The initial mount is failing due to the issues described above. However the lookup is continued for the entire path

I think i might ignore this issue until i can upgrade the kernel on the clients and do more testing as it seems to be related to issue 2.

thankyou if you have actually read this far:)

Cheers
Anthony

Signature · 4 lines
anthony menasse
systems administrator | [email protected]
rising sun pictures | www.rsp.com.au
direct line +61 2 9384 4572
maintainer
#1 lore raw
Quoted 44 lines · Anthony Menasse

Hi,

I was hoping for some feedback on wether the following actions are appropriate to fix the issues we are having with automount /mount. I haven't explained the problems in detail, but from reading the mailing list and other sources they seem like fairly common issues. I just want to make sure my proposed fixes make sense. Im guessing a lot of this has been discussed many times before on this list, but i just wanted some confirmation that i am on the right track (or not). Any advice would be greatly appreciated.

Setup:

A bunch of NFS servers running a mixture of RH Enterprise 4, Enterprise 3 and X/OS linux4 and a bunch of Fedora Core 3 clients running a non Fedora built 2.6.11.7 kernel.

The automount update i am looking at installing is the latest automount package from FC5 (4.1.4 Release 33) rebuilt for FC3 (which seems to work ok after some initial testing on a single box)

I want to avoid updating the kernel at this point.

Not sure about my strategy about updating 'mount', would you recommend getting the latest source or will an older package from FC3 do? In other words are there still serious issues with mount which mean i should be running the latest source or are the issues such as overriding RPC timeout values and locking issues with mtab only an issue with ancient versions of mount?

Issue 1:

nfs: mount failure chop:/export/home/rsp on /home/rsp

I believe this is due to automount using a mount retry value of 0 and dodgy versions of mount overriding the default RPC timeout from 60seconds to 0.7 seconds so the attempt times out too quickly

References:

http://nfsworld.blogspot.com/2005/03/automounter-tuning.html

Quoted 8 lines

http://www.mail-archive.com/[email protected]/msg05281.html

Possible Fix:

I want to fix this by using --retry=2 in the auto.master file, and update mount where i can (i want to do both as i can't update mount on every machine/ platform right now)

That would be "retry=2", the "--" is used for autofs options. Could be worth a try, yes.

Quoted 6 lines

Issue 2:

When accessing an unmounted automount mount point the automounter will complain that the directory is already mounted (even though it isn't!)

Quoted 6 lines

You may get something like this in the log / console.

BUG: <dir> already mounted

I think this is due to a corrupt mtab file due to the race /locking issues with older versions of mount

Could be, yes.

Quoted 5 lines

Possible Fix:

I want to update mount where i can , and link /etc/mtab to /proc/mount . (Are there any issues with having mtab as a link to /proc/mounts?)

Should be OK with version 4.

Quoted 6 lines

Issue 3:

Wierd errors in the logs about attempts to mount files:

Quoted 8 lines
Jan 20 18:45:36 glass automount[28315]: mount(nfs): nfs: mount failure 

hassan:/export/home/rsp/.sitecshrc on /home/rsp/.sitecshrc

The initial mount is failing due to the issues described above. However the lookup is continued for the entire path

I think i might ignore this issue until i can upgrade the kernel on the clients and do more testing as it seems to be related to issue 2.

Exactly.
Patches are included in the 5.0.1 tar for several kernels that help with this problem somewhat. They usually occur after a mount fail and are caused by the kernel continuing the lookup when it shouldn't.

Ian