Date:   Fri, 10 Dec 2021 16:09:31 +0000
From:   Aaron Tomlin <atomlin@redhat.com>
To:     Petr Mladek <pmladek@suse.com>
Cc:     Luis Chamberlain <mcgrof@kernel.org>,
        Christoph Lameter <cl@linux.com>,
        Miroslav Benes <mbenes@suse.cz>,
        Andrew Morton <akpm@linux-foundation.org>, jeyu@kernel.org,
        linux-kernel@vger.kernel.org, linux-modules@vger.kernel.org,
        atomlin@atomlin.com, ghalat@redhat.com
Subject: Re: [RFC PATCH] module: Introduce module unload taint tracking
Message-ID: <20211210160931.ftvxpulno73a2l7c@ava.usersys.com>
References: <20211124173327.3878177-1-atomlin@redhat.com>
 <YbEZ4HgSYQEPuRmS@bombadil.infradead.org>
 <20211209153131.a54fdfbci4qnyy6h@ava.usersys.com>
 <YbKUUJUtjBk/n913@bombadil.infradead.org>
 <YbMlVFwBiRujKdEX@alley>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <YbMlVFwBiRujKdEX@alley>
Precedence: bulk

On Fri 2021-12-10 11:00 +0100, Petr Mladek wrote:
> > If someone enables this feature I can't think of a reason why they
> > would want to limit this to some arbitrary number. So my preference
> > is to remove that limitation completely. I see no point to it.
> 
> I agree with Luis here. We could always add the limit later when
> people report some real life problems with too long list. It is
> always good to know that someone did some heavy lifting in
> the system.

Fair enough.

> It might be even interesting to remember timestamp of the removal
> to match it with another events reported in the system log.

I'm not so sure about this. We could gather such details already via Ftrace
(e.g. see load_module()). Personally, I'd prefer to maintain a simple list.

> > If you just bump the count then its not duplication, it just adds
> > more information that the same module name with the same taint flag
> > has been unloaded now more than once.
> 
> Please, do not remove records that a module was removed. IMHO, it
> might be useful to track all removed module, including the non-tainted
> ones. Module removal is always tricky and not much tested. The tain
> flags might be just shown as extra information in the output.

This is an interesting suggestion. Albeit, as per the subject, I prefer to
just keep track of any module that tainted the kernel. That being said,
Petr, if you'd prefer to track each module unload/or deletion event, then I
would suggest for instance to remove a module once it has been reintroduced
or maintain an unload count as suggested by Luis.

Please let me know your thoughts.


Kind regards,

-- 
Aaron Tomlin