Date:   Thu, 26 Jan 2023 12:31:58 +0200
From:   Ilias Apalodimas <ilias.apalodimas@linaro.org>
To:     Alexander H Duyck <alexander.duyck@gmail.com>
Cc:     Felix Fietkau <nbd@nbd.name>, netdev@vger.kernel.org,
        Jesper Dangaard Brouer <hawk@kernel.org>,
        "David S. Miller" <davem@davemloft.net>,
        Eric Dumazet <edumazet@google.com>,
        Jakub Kicinski <kuba@kernel.org>,
        Paolo Abeni <pabeni@redhat.com>,
        Lorenzo Bianconi <lorenzo@kernel.org>,
        linux-kernel@vger.kernel.org, Yunsheng Lin <linyunsheng@huawei.com>
Subject: Re: [PATCH] net: page_pool: fix refcounting issues with fragmented
 allocation
Message-ID: <Y9JWniFQmcc7m5Ey@hera>
References: <20230124124300.94886-1-nbd@nbd.name>
 <CAC_iWjKAEgUB8Z3WNNVgUK8omXD+nwt_VPSVyFn1i4EQzJadog@mail.gmail.com>
 <f3d079ce930895475f307de3fdaed0b85b4f2671.camel@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <f3d079ce930895475f307de3fdaed0b85b4f2671.camel@gmail.com>
Precedence: bulk

Hi Alexander,

Sorry for being late to the party,  was overloaded...

On Tue, Jan 24, 2023 at 07:57:35AM -0800, Alexander H Duyck wrote:
> On Tue, 2023-01-24 at 16:11 +0200, Ilias Apalodimas wrote:
> > Hi Felix,
> >
> > ++cc Alexander and Yunsheng.
> >
> > Thanks for the report
> >
> > On Tue, 24 Jan 2023 at 14:43, Felix Fietkau <nbd@nbd.name> wrote:
> > >
> > > While testing fragmented page_pool allocation in the mt76 driver, I was able
> > > to reliably trigger page refcount underflow issues, which did not occur with
> > > full-page page_pool allocation.
> > > It appears to me, that handling refcounting in two separate counters
> > > (page->pp_frag_count and page refcount) is racy when page refcount gets
> > > incremented by code dealing with skb fragments directly, and
> > > page_pool_return_skb_page is called multiple times for the same fragment.
> > >
> > > Dropping page->pp_frag_count and relying entirely on the page refcount makes
> > > these underflow issues and crashes go away.
> > >
> >
> > This has been discussed here [1].  TL;DR changing this to page
> > refcount might blow up in other colorful ways.  Can we look closer and
> > figure out why the underflow happens?
> >
> > [1] https://lore.kernel.org/netdev/1625903002-31619-4-git-send-email-linyunsheng@huawei.com/
> >
> > Thanks
> > /Ilias
> >
> >
>
> The logic should be safe in terms of the page pool itself as it should
> be holding one reference to the page while the pp_frag_count is non-
> zero. That one reference is what keeps the two halfs in sync as the
> page shouldn't be able to be freed until we exhaust the pp_frag_count.

Do you remember why we decided to go with the fragment counter instead of
page references?

>
> To have an underflow there are two possible scenarios. One is that
> either put_page or free_page is being called somewhere that the
> page_pool freeing functions should be used.

Wouldn't that affect the non fragmented path as well? IOW the driver that
works with a full page would crash as well.

> The other possibility is
> that a pp_frag_count reference was taken somewhere a page reference
> should have.
>
> Do we have a backtrace for the spots that are showing this underrun? If
> nothing else we may want to look at tracking down the spots that are
> freeing the page pool pages via put_page or free_page to determine what
> paths these pages are taking.

Thanks
/Ilias