netdev - Re: problem with rtnetlink 'reference' count

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

Message-ID: <20171023153200.GA12422@breakpoint.cc>
Date:   Mon, 23 Oct 2017 17:32:00 +0200
From:   Florian Westphal <fw@...len.de>
To:     Peter Zijlstra <peterz@...radead.org>
Cc:     David Miller <davem@...emloft.net>, fw@...len.de,
        netdev@...r.kernel.org
Subject: Re: problem with rtnetlink 'reference' count

Peter Zijlstra <peterz@...radead.org> wrote:
>   019a316992ee ("rtnetlink: add reference counting to prevent module unload while dump is in progress")
> 
> And that commit is _completely_ broken.
> 
>  1) it not in fact a refcount, so using refcount_t is silly

Your suggestion is...?

>  2) there is a distinct lack of memory barriers, so we can easily
>     observe the decrement while the msg_handler is still in progress.

I guess you mean it needs:

+	smp_mb__before_atomic();
	refcount_dec(&rtnl_msg_handlers_ref[family]);
?

However, this refcount_dec is misplaced anyway as it would need
to occur from nlcb->done() (the handler function gets stored in socket for
use by next recvmsg), so this change is indeed not helpful at all.

>  3) waiting with a schedule()/yield() loop is complete crap and subject
>     life-locks, imagine doing that rtnl_unregister_all() from a RT task.

Whats the alternative?

Only one I see is to pass THIS_MODULE to hook reg/unreg so we know what module registered the
family for purpose of module get/put but thats going to be messy.

Alternatively we can of course sleep instead of schedule() but that
doesn't appear too appealing either (albeit it is a lot less intrusive).

Any other idea?