linux-kernel - Re: [PATCH 5/5] writeback, memcg: Implement foreign dirty flushing

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: <20190821160037.GK2263813@devbig004.ftw2.facebook.com>
Date:   Wed, 21 Aug 2019 09:00:37 -0700
From:   Tejun Heo <tj@...nel.org>
To:     Jan Kara <jack@...e.cz>
Cc:     axboe@...nel.dk, hannes@...xchg.org, mhocko@...nel.org,
        vdavydov.dev@...il.com, cgroups@...r.kernel.org,
        linux-mm@...ck.org, linux-block@...r.kernel.org,
        linux-kernel@...r.kernel.org, kernel-team@...com, guro@...com,
        akpm@...ux-foundation.org
Subject: Re: [PATCH 5/5] writeback, memcg: Implement foreign dirty flushing

Hello,

On Fri, Aug 16, 2019 at 06:02:56PM +0200, Jan Kara wrote:
> 1) You ask to writeback LONG_MAX pages. That means that you give up any
> livelock avoidance for the flusher work and you can writeback almost
> forever if someone is busily dirtying pages in the wb. I think you need to
> pick something like amount of dirty pages in the given wb (that would have
> to be fetched after everything is looked up) or just some arbitrary
> reasonably small constant like 1024 (but then I guess there's no guarantee
> stuck memcg will make any progress and you've invalidated the frn entry
> here).

I see.  Yeah, I think the right thing to do would be feeding the
number of dirty pages or limiting it to one full sweep.  I'll look
into it.

> 2) When you invalidate frn entry here by writing 0 to 'at', it's likely to get
> reused soon. Possibly while the writeback is still running. And then you
> won't start any writeback for the new entry because of the
> atomic_read(&frn->done.cnt) == 1 check. This seems like it could happen
> pretty frequently?

Hmm... yeah, the clearing might not make sense.  I'll remove that.

Thanks.

-- 
tejun