Very strange autofs failure on 2.6.30-rc2 with autofs-5.0.3-41 (fc10)

Very strange autofs failure on 2.6.30-rc2 with autofs-5.0.3-41 (fc10)

[H. Peter Anvin]

I have been using for a very long time a setup where /home is an autofs 
from a plain indirect NIS map, but my personal home directory 
(/home/hpa) is simply a bind mount from /export/home/hpa.  A 
straightforward use of autofs.

I just rebooted my system yesterday, however, and found that all the 
directory entries in my home directory had gotten replaced with ghost 
directories -- and even more confusingly, the date wasn't the current 
date, but was back in 2008.

The ghost directories were "sterile" in the sense that entering them 
wouldn't show the proper contents of those directories.  As a result, 
massive failure.

After suspecting filesystem corruption, and this, that and the other 
thing, I found that this was only when viewing though autofs 
(/home/hpa), and that the real filesystem (/export/home/hpa) was fully 
intact.  Somehow autofs had ended up ghosting pretty much my entire 
directory, and doing so in some incorrect fashion.

Replacing /home with a plain bind mount (no autofs) to /export/home 
resolved the issue.

	-hpa

Re: Very strange autofs failure on 2.6.30-rc2 with autofs-5.0.3-41 (fc10)

[Ian Kent]

H. Peter Anvin wrote:
> I have been using for a very long time a setup where /home is an autofs 
> from a plain indirect NIS map, but my personal home directory 
> (/home/hpa) is simply a bind mount from /export/home/hpa.  A 
> straightforward use of autofs.
> 
> I just rebooted my system yesterday, however, and found that all the 
> directory entries in my home directory had gotten replaced with ghost 
> directories -- and even more confusingly, the date wasn't the current 
> date, but was back in 2008.
> 
> The ghost directories were "sterile" in the sense that entering them 
> wouldn't show the proper contents of those directories.  As a result, 
> massive failure.
> 
> After suspecting filesystem corruption, and this, that and the other 
> thing, I found that this was only when viewing though autofs 
> (/home/hpa), and that the real filesystem (/export/home/hpa) was fully 
> intact.  Somehow autofs had ended up ghosting pretty much my entire 
> directory, and doing so in some incorrect fashion.
> 
> Replacing /home with a plain bind mount (no autofs) to /export/home 
> resolved the issue.

This failure sounds rather more spectacular than should result from a
known problem we have with a recent init scripts change. I'll investigate.

Have a look at https://bugzilla.redhat.com/show_bug.cgi?id=475002.

Ian

Re: Very strange autofs failure on 2.6.30-rc2 with autofs-5.0.3-41 (fc10)

[Ian Kent]

Ian Kent wrote:
> H. Peter Anvin wrote:
>> I have been using for a very long time a setup where /home is an autofs 
>> from a plain indirect NIS map, but my personal home directory 
>> (/home/hpa) is simply a bind mount from /export/home/hpa.  A 
>> straightforward use of autofs.
>>
>> I just rebooted my system yesterday, however, and found that all the 
>> directory entries in my home directory had gotten replaced with ghost 
>> directories -- and even more confusingly, the date wasn't the current 
>> date, but was back in 2008.
>>
>> The ghost directories were "sterile" in the sense that entering them 
>> wouldn't show the proper contents of those directories.  As a result, 
>> massive failure.
>>
>> After suspecting filesystem corruption, and this, that and the other 
>> thing, I found that this was only when viewing though autofs 
>> (/home/hpa), and that the real filesystem (/export/home/hpa) was fully 
>> intact.  Somehow autofs had ended up ghosting pretty much my entire 
>> directory, and doing so in some incorrect fashion.
>>
>> Replacing /home with a plain bind mount (no autofs) to /export/home 
>> resolved the issue.
> 
> This failure sounds rather more spectacular than should result from a
> known problem we have with a recent init scripts change. I'll investigate.
> 
> Have a look at https://bugzilla.redhat.com/show_bug.cgi?id=475002.

I built 2.6.30-rc2, installed it on F-10 and it worked OK for me with a
simple indirect map as you described. I also ran through the autofs
Connectathon test suite I use several times without seeing a problem.

Can you post the map your using and (although it's probabbly too late
now) a debug log of the event. Jeff Moyer has some information about
version 5 debug logging setup at http://people.redhat.com/jmoyer.

Ian