[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20240801125856.GC10274@breakpoint.cc>
Date: Thu, 1 Aug 2024 14:58:56 +0200
From: Florian Westphal <fw@...len.de>
To: Wojciech Gładysz <wojciech.gladysz@...ogain.com>
Cc: davem@...emloft.net, edumazet@...gle.com, kuba@...nel.org,
pabeni@...hat.com, jhs@...atatu.com, xiyou.wangcong@...il.com,
jiri@...nulli.us, netdev@...r.kernel.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH] kernel/net: missused TCQ_F_NOLOCK flag
Wojciech Gładysz <wojciech.gladysz@...ogain.com> wrote:
> TCQ_F_NOLOCK yields no locking option for a qdisc. At some places in the
> code the testing for the flag seems logically reverted. The change fixes
> the following lockdep issue.
>
> ======================================================
> WARNING: possible circular locking dependency detected
> 5.10.0-syzkaller #0 Not tainted
This kernel is over 3 years old. Please either only fuzz
on net or net-next and keep them up-to-date, or wait for
the fuzzer to get a reproducer and then verify that reproducer
still triggers on current kernels.
> static inline bool qdisc_is_running(struct Qdisc *qdisc)
> {
> - if (qdisc->flags & TCQ_F_NOLOCK)
> + if (!(qdisc->flags & TCQ_F_NOLOCK))
> return spin_is_locked(&qdisc->seqlock);
Are you absolutely sure? I find it hard to believe something
like this would go unnoticed for years.
Curious glance tells me seqlock is used to sync nolock qdiscs vs.
qdisc reset, i.e. exiting code is correct.
Powered by blists - more mailing lists