[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Strange XFS corruption...



Russell Cattelan wrote:
> On Mon, 2003-01-13 at 09:01, Michael Sinz wrote:
> 
>>Eric Sandeen wrote:
>>
>>>Yep, when we got multiple entries here, neither check nor repair complained.
>>>I guess it's not fundamentally inconsistent from a filesystem integrity
>>>perspective, but it does not make the operating system happy.
>>>
>>>We don't have a root cause for this one, we think it's another way
>>>that the shutdown problem exhibited itself.  If you can run the latest
>>>code (1.2pre5) the shutdown problem should be fixed, and if you see
>>>it again, let us know!
>>
>>Well, I am now running from CVS as of Jan 11, 2003.  I have not seen
>>this elsewhere yet.
>>
>>I wonder how the filesystem thought that this was "fine" since
>>all of the files (the multiple entries) showed exactly the same
>>date/size/etc.  (Albeit that may be due to the VFS layer - I am
>>not 100% clear on how those two interact)
> 
> Multiple entries in a directory shouldn't happen, and from a FS
> consistency point it really isn't a fatal problem either.
> It's basically a "lookup" problem.
> Checking for multiple entries of the same name is not something
> xfs_check/xfs_repair checks for and therefore will not report a problem.
> 
> It's certainly possible to add this check to xfs_repair although it's
> unlikely we will make it a high priority.

I can understand that - as it could significantly increase the time
needed to do a check and/or repair.  If it does not happen, it is not
high priority.  However, when it does happen, it is a PITA to deal with.

>>However, I would think that XFS should not like having multiple
>>file node entries in a directory with the same identifier (file name).
> 
> XFS doesn't really care just it just passes whatever it has to the upper
> layers for it to sort out. This is true of all the FS.

I guess this is why, when I try to delete one of them that things
go strange...

There still was something strange going on in the system since the
rm, while looking like it worked, caused problems (ls would then give
an error for each "duplicate" file that was there but not list the file
even once) and, after a reboot, they all came back)  (Sort of like
a bad dream with cats :-)

-- 
Michael Sinz -- Director, Systems Engineering -- Worldgate Communications
A master's secrets are only as good as
	the master's ability to explain them to others.