Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220907015618.2140679-1-william.xuanziyang@huawei.com>
In-Reply-To: <20220907015618.2140679-1-william.xuanziyang@huawei.com>
From:   Eric Dumazet <edumazet@google.com>
Date:   Fri, 9 Sep 2022 09:36:00 -0700
Message-ID: <CANn89iKPmDXvPzw9tYpiqHH7LegAgTb14fAiAqH8vAxZ3KsryA@mail.gmail.com>
Subject: Re: [PATCH net] net: tun: limit first seg size to avoid oversized linearization
To:     Ziyang Xuan <william.xuanziyang@huawei.com>
Cc:     David Miller <davem@davemloft.net>,
        Jakub Kicinski <kuba@kernel.org>,
        Paolo Abeni <pabeni@redhat.com>,
        netdev <netdev@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Alexei Starovoitov <ast@kernel.org>,
        Daniel Borkmann <daniel@iogearbox.net>,
        Jesper Dangaard Brouer <hawk@kernel.org>,
        John Fastabend <john.fastabend@gmail.com>,
        Petar Penkov <peterpenkov96@gmail.com>,
        Mahesh Bandewar <maheshb@google.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Sep 6, 2022 at 6:56 PM Ziyang Xuan
<william.xuanziyang@huawei.com> wrote:
>
> Recently, we found a syzkaller problem as following:
>
> ========================================================
> WARNING: CPU: 1 PID: 17965 at mm/page_alloc.c:5295 __alloc_pages+0x1308/0x16c4 mm/page_alloc.c:5295
> ...
> Call trace:
>  __alloc_pages+0x1308/0x16c4 mm/page_alloc.c:5295
>  __alloc_pages_node include/linux/gfp.h:550 [inline]
>  alloc_pages_node include/linux/gfp.h:564 [inline]
>  kmalloc_large_node+0x94/0x350 mm/slub.c:4038
>  __kmalloc_node_track_caller+0x620/0x8e4 mm/slub.c:4545
>  __kmalloc_reserve.constprop.0+0x1e4/0x2b0 net/core/skbuff.c:151
>  pskb_expand_head+0x130/0x8b0 net/core/skbuff.c:1654
>  __skb_grow include/linux/skbuff.h:2779 [inline]
>  tun_napi_alloc_frags+0x144/0x610 drivers/net/tun.c:1477
>  tun_get_user+0x31c/0x2010 drivers/net/tun.c:1835
>  tun_chr_write_iter+0x98/0x100 drivers/net/tun.c:2036
>
> It is because the first seg size of the iov_iter from user space is
> very big, it is 2147479538 which is bigger than the threshold value
> for bail out early in __alloc_pages(). And skb->pfmemalloc is true,
> __kmalloc_reserve() would use pfmemalloc reserves without __GFP_NOWARN
> flag. Thus we got a warning.
>
> I noticed that non-first segs size are required less than PAGE_SIZE in
> tun_napi_alloc_frags(). The first seg should not be a special case, and
> oversized linearization is also unreasonable. Limit the first seg size to
> PAGE_SIZE to avoid oversized linearization.
>
> Fixes: 90e33d459407 ("tun: enable napi_gro_frags() for TUN/TAP driver")
> Signed-off-by: Ziyang Xuan <william.xuanziyang@huawei.com>
> ---
>  drivers/net/tun.c | 5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/net/tun.c b/drivers/net/tun.c
> index 259b2b84b2b3..7db515f94667 100644
> --- a/drivers/net/tun.c
> +++ b/drivers/net/tun.c
> @@ -1454,12 +1454,12 @@ static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
>                                             size_t len,
>                                             const struct iov_iter *it)
>  {
> +       size_t linear = iov_iter_single_seg_count(it);
>         struct sk_buff *skb;
> -       size_t linear;
>         int err;
>         int i;
>
> -       if (it->nr_segs > MAX_SKB_FRAGS + 1)
> +       if (it->nr_segs > MAX_SKB_FRAGS + 1 || linear > PAGE_SIZE)
>                 return ERR_PTR(-EMSGSIZE);
>

This does not look good to me.

Some drivers allocate 9KB+ for 9000 MTU, in a single allocation,
because the hardware is not SG capable in RX.

>         local_bh_disable();
> @@ -1468,7 +1468,6 @@ static struct sk_buff *tun_napi_alloc_frags(struct tun_file *tfile,
>         if (!skb)
>                 return ERR_PTR(-ENOMEM);
>
> -       linear = iov_iter_single_seg_count(it);
>         err = __skb_grow(skb, linear);
>         if (err)
>                 goto free;
> --
> 2.25.1
>