netdev - Re: [PATCH 2/2] netfilter: revert "conntrack: silent a memory leak warning"

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Thu, 17 Oct 2019 11:30:51 -0700
From:   Maciej Żenczykowski <zenczykowski@...il.com>
To:     Cong Wang <xiyou.wangcong@...il.com>
Cc:     Florian Westphal <fw@...len.de>,
        "David S . Miller" <davem@...emloft.net>,
        Linux NetDev <netdev@...r.kernel.org>,
        Eric Dumazet <edumazet@...gle.com>,
        Pablo Neira Ayuso <pablo@...filter.org>
Subject: Re: [PATCH 2/2] netfilter: revert "conntrack: silent a memory leak warning"

> So you conclude as it is not leak too? Then what are you trying to
> fix?

I conclude there is no easily *visible* leak.
At least not at first glance - not with single threaded code.

> I am becoming more confused after this. :-/

I think adding kmemleak_not_leak() is hiding the fact that there
actually is a leak.

I think the leak is far more subtle.  Possibly some sort of race
condition or something.
I don't see it though.

The rcu doesn't seem entirely kosher, but I know little about such things.

And I think the leak is *still* here.

After all kmemleak_not_leak is purely annotation.
It doesn't fix any leaks, it just makes us not warn about them.

> > Basically AFAICT our use of __krealloc() is exactly like krealloc()
> > except instead of kfree() we do kfree_rcu().
> >
> > And thus I don't understand the need for kmemleak_not_leak(old).
>
> kfree_rcu() is a callback deferred after a grace period, so if we
> allocate the memory again before that callback, it is reported to
> kmemleak as a memory leak unless we mark it as not, right?
>
> Or kfree_rcu() works nicely with kmemleak which I am not aware
> of?

We have kfree_rcu() all over the kernel, but there's very few
kmemleak_not_leak's.

I don't see how kfree_rcu() could not work nicely with kmemleak.
If it didn't we'd have it reporting leaks all over the place...