[OpenAFS] Possible Kernel Memory leak, OpenAFS 1.4.2+, RH3 i686/amd64

Kevin Hildebrand kevin@umd.edu
Tue, 20 Feb 2007 09:31:26 -0500 (EST)

On Mon, 19 Feb 2007, Marcus Watts wrote:

> Kevin Hildebrand <kevin@umd.edu> writes:
>> Hello, I've been banging my head on this for a while and thought I'd see
>> if anyone else has any clues.
>> Since upgrading to OpenAFS 1.4.2, both our amd64 and i386 RedHat 3 linux
>> platforms have been seeing significant kernel memory leaks.
>> Namely, the "size-256" slab grows by approximately 4K each time I stat
>> files in a certain directory in AFS.  Some of our more active machines
>> will run themselves out of memory in a matter of days.
>> I've tried 1.4.3-rc1 and it exhibits the same behavior.
>> I have not been able to nail down the exact characteristics of the files
>> that will cause this problem- but once I find the appropriate location,
>> simply statting files at or below the directory in question is sufficient
>> to cause the leak.
>> If anyone has any clues as to where else I should look, I'd appreciate it.
> At "or below"?  Sounds like some sort of directory issue, might
> not be the file itself at all.
> 2 parameters that might affect directory processing include:
> 	length of filename
> 	# of filenames in a directory
> mount points & symbolic links might also affect things.
> It would be interesting to know what's in the lost memory.

Ok, with some additional debugging, I have determined that the memory that 
is being leaked is coming from crget- each time I stat something in a 
particular directory tree (whether or not the file exists), one cred_t 
gets leaked.

I still haven't been able to determine what the characteristic is of the 
directory tree that makes it "poisoned"...