linux-kernel - Re: [RT v5.11-rt7] WARNING at include/linux/seqlock.h:271 nft_counter

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]

Message-ID: <20210223142118.u22yhw3ku4ttx5hq@linutronix.de>
Date:   Tue, 23 Feb 2021 15:21:18 +0100
From:   Sebastian Andrzej Siewior <bigeasy@...utronix.de>
To:     Juri Lelli <juri.lelli@...hat.com>
Cc:     linux-rt-users <linux-rt-users@...r.kernel.org>,
        LKML <linux-kernel@...r.kernel.org>,
        "Ahmed S. Darwish" <a.darwish@...utronix.de>,
        Peter Zijlstra <peterz@...radead.org>,
        Thomas Gleixner <tglx@...utronix.de>
Subject: Re: [RT v5.11-rt7] WARNING at include/linux/seqlock.h:271
 nft_counter_eval

On 2021-02-23 14:53:40 [+0100], Juri Lelli wrote:
> 
> So, I'm a bit confused and I'm very likely missing details (still
> digesting the seqprop_ magic), but write_seqcount_being() has
> 
>  if (seqprop_preemptible(s))
>      preempt_disable();
> 
> which in this case (no lock associated) is defined to return false, 
> while it should return true on RT (or in some occasions)? Or maybe this
> is what you are saying already.

write_seqcount_begin() has seqprop_assert() at the very beginning which
ends in __seqprop_assert() in your case (seqcount_t). Your warning.

> Also, the check for preemption been disabled happens before we can
> actually potentially disable it, no?

That seqprop_preemptible() is true for !RT for mutex/ww_mutex locks. On
RT it is always false since it does lock()+unlock() of the lock that is
part of the seqcount.

But back to the original issue: at write_seqcount_begin() preemption is
disabled !RT implicit by local_bh_disable(). Therefore no warning.
On RT local_bh_disable() disables BH on the CPUs so locking wise (since
it is a per-CPU seqcount it should work. Preemption remains enabled so
we have a warning.

I have no idea what annotation would be best here. Having a
local_bh_disable() type of a lock and the seqcount is not part of the data
structure it protects is less than ideal.
However, if I understand this correct then this nft_counter_percpu_priv
exists once per nft rule. The seqcount exists once per-CPU since it is
unlikely to modify two counters at once on a single CPU :) So there is
that.

While looking at it, there is nft_counter_reset() which modifies the
values without a seqcount write lock. This might be okay.

> Thanks for the quick reply!
> 
> Best,
> Juri

Sebastian