(KAME-snap 8792) Re: Weird memory exhaustion with FreeBSD 4.10-STABLE

[Pekka Savola]

On Sat, 25 Sep 2004, JINMEI Tatuya / [ISO-2022-JP] $B?@L at C#:H(B wrote:
> Apparently the most significant change is the memory consumption
> regarding routing table:
> 
> In "wrk"
> Memory statistics by bucket size
> Size   In Use   Free   Requests  HighWater  Couldfree
>   64     2196     44       5165     320          0
> Memory statistics by type                          Type  Kern
>         Type  InUse MemUse HighUse  Limit Requests Limit Limit Size(s)
>      routetbl   461    67K     67K 10148K      538    0     0  16,32,64,128,256,512
> 
> In "brk"
> Memory statistics by bucket size
> Size   In Use   Free   Requests  HighWater  Couldfree
>   64    37865   2583    5315047     320       1075
> Memory statistics by type                          Type  Kern
>         Type  InUse MemUse HighUse  Limit Requests Limit Limit Size(s)
>      routetbl 64959 10148K  10148K 10148K  2445306    0     0  16,32,64,128,256,512

OK..

> Some random thoughts, which may or may not help, at the moment:
> 
> 1. do you see massive number of entries with "netstat -rna"?

Yes.

# netstat -nra | wc -l
   32468
#

A couple of examples:

2002:41a:1e23::41a:1e23           2002:c058:6301::1741          UHW3       stf0
2002:41a:1eaa::41a:1eaa           2002:c058:6301::1741          UHW3       stf0
2002:41a:6b0e::41a:6b0e           2002:c058:6301::1741          UHW3       stf0
2002:41a:6f4b::41a:6f4b           2002:c058:6301::1741          UHW3       stf0
2002:41a:70f5::41a:70f5           2002:c058:6301::1741          UHW3       stf0   1433
2002:41a:7411::41a:7411           2002:c058:6301::1741          UHW3       stf0
2002:41a:8e81::41a:8e81           2002:c058:6301::1741          UHW3       stf0

That's basically about all the 2002:xxx addresses which have been 
relayed through.  I'd suspect the number is maxed and 
garbage-collected around 2^15 though, because the box is relaying w/ 
much many more addresses than that.

I guess this provides a hint at a very probable source of leakage.

> 2. if you specify the "link2" flag on the stf interface, what if you
>    do not specify it?  (if such operation is acceptable in your
>    environment)

The flag is not specified.

> 3. if you can replace the kernel with KAME snap versions, do you see
>    any difference in the memory consumption?

This would be a bit difficult, so I'd like to avoid it if possible.

-- 
Pekka Savola                 "You each name yourselves king, yet the
Netcore Oy                    kingdom bleeds."
Systems. Networks. Security. -- George R.R. Martin: A Clash of Kings