Re: [PATCH] fix mem-leak in netfilter

[Patrick McHardy]

Stephen Frost wrote:
> * Patrick McHardy (kaber@xxxxxxxxx) wrote:
> 
>>Anyway, here goes the first shot at a replacement, it should be fully
>>compatible. Comments and testing welcome.
> 
> 
> This patch didn't apply cleanly against 2.6.16; I didn't think there had
> been other changes since then.  As it was an entire replacement I just
> pulled out the '[+ ]' lines from the patch.  Hopefully this doesn't lead
> to problems in my review.


That should be fine. That patch applies on top of Jespers patch which
started this thread, which I plan to push to Dave today.

> It probably would have been better to integrate it with ipset, as I've
> mentioned previously.  Other comments:


Unfortunately we need to provide compatibility.

> recent_entry_init() appears to just look for something to delete when
> the maximum number of entries has been reached, starting from the hash
> position of the address.  The original ipt_recent, quite intentionally,
> looked for the *oldest* address to replace.  This meant that the list
> only had to be large enough to cover the number of addresses seen in a
> given time-period.  This change would mean that the list would need to
> be large enough to hold all addresses seen always, to be able to enforce
> the time-based rules ipt_recent was written for.
> 
> ie: List of 100 addresses.  Highest timeout value in the ruleset is 60
> seconds.  Average of 100 individual addresses in a 60-second timeframe.
> The old ipt_recent would correctly enforce the 60-second requirement in
> the ruleset.  With the new version, as soon as the list was full the
> next address could replace any address in the list, even if that address
> was only 15 seconds old.
> 
> One way to handle this would be to track the highest time value in the
> rulesets but as the ruleset is dynamic you could end up throwing away an
> address which would have been caught by a rule that was about to be
> added.  The old module was written with the expectation of the list
> always being full and that it would only be less-than-full shortly after
> booting.  By then only removing the oldest entry in the table for each
> new address seen the maximum amount of time possible for the given table
> size and distinct addresses seen is achieved.


I wasn't sure whether eviction was happening intentional in the old code
at all - still not able to locate the code where this happens, just
noticed that it does do eviction when I manually tried to trigger
a table overflow by adding entries through /proc. Anyway, it should
be easy to fix by keeping an additional lru list. I'll post
an updated patch soon.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/