Re: [PATCH net v2] net: Clarify the difference between hard_header_len and needed_headroom

[Willem de Bruijn]

On Thu, Sep 10, 2020 at 7:44 AM Xie He <xie.he.0141@xxxxxxxxx> wrote:
>
> The difference between hard_header_len and needed_headroom has long been
> confusing to driver developers. Let's clarify it.
>
> The understanding on this issue in this patch is based on the following
> reasons:
>
> 1.
>
> In af_packet.c, the function packet_snd first reserves a headroom of
> length (dev->hard_header_len + dev->needed_headroom).
> Then if the socket is a SOCK_DGRAM socket, it calls dev_hard_header,
> which calls dev->header_ops->create, to create the link layer header.
> If the socket is a SOCK_RAW socket, it "un-reserves" a headroom of
> length (dev->hard_header_len), and checks if the user has provided a
> header of length (dev->hard_header_len) (in dev_validate_header).

I think if you want to clarify, we have to be exact: [up to]
dev->hard_header_len. For protocols with variable length link layer
headers, the length is at least dev->min_header_len.

> This shows the developers of af_packet.c expect hard_header_len to
> be consistent with header_ops.
>
> 2.
>
> In af_packet.c, the function packet_sendmsg_spkt has a FIXME comment.
> That comment states that prepending an LL header internally in a driver
> is considered a bug. I believe this bug can be fixed by setting
> hard_header_len to 0, making the internal header completely invisible
> to af_packet.c (and requesting the headroom in needed_headroom instead).
>
> 3.
>
> There is a commit for a WiFi driver:
> commit 9454f7a895b8 ("mwifiex: set needed_headroom, not hard_header_len")
> According to the discussion about it at:
>   https://patchwork.kernel.org/patch/11407493/
> The author tried to set the WiFi driver's hard_header_len to the Ethernet
> header length, and request additional header space internally needed by
> setting needed_headroom. This means this usage is already adopted by
> driver developers.
>
> Cc: Willem de Bruijn <willemdebruijn.kernel@xxxxxxxxx>
> Cc: Eric Dumazet <eric.dumazet@xxxxxxxxx>
> Cc: Brian Norris <briannorris@xxxxxxxxxxxx>
> Cc: Cong Wang <xiyou.wangcong@xxxxxxxxx>
> Signed-off-by: Xie He <xie.he.0141@xxxxxxxxx>

Thanks for trying to clarify the behavior.

This patch should target net-next.

> ---
>
> Change from v1:
> Small change to the commit message.
>
> ---
>  include/linux/netdevice.h |  4 ++--
>  net/packet/af_packet.c    | 19 +++++++++++++------
>  2 files changed, 15 insertions(+), 8 deletions(-)
>
> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> index 7bd4fcdd0738..3999b04e435d 100644
> --- a/include/linux/netdevice.h
> +++ b/include/linux/netdevice.h
> @@ -1691,8 +1691,8 @@ enum netdev_priv_flags {
>   *     @min_mtu:       Interface Minimum MTU value
>   *     @max_mtu:       Interface Maximum MTU value
>   *     @type:          Interface hardware type
> - *     @hard_header_len: Maximum hardware header length.
> - *     @min_header_len:  Minimum hardware header length
> + *     @hard_header_len: Maximum length of the headers created by header_ops
> + *     @min_header_len:  Minimum length of the headers created by header_ops

This does not help imho. The existing definitions were clear and more
exact: hardware (i.e., link layer) headers.

Even more explicit, the hardware headers here must (probably) match
those dictated by dev->type, such as ARPHRD_ETHER.

>   *
>   *     @needed_headroom: Extra headroom the hardware may need, but not in all
>   *                       cases can this be guaranteed
> diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
> index 2b33e977a905..0e324b08cb2e 100644
> --- a/net/packet/af_packet.c
> +++ b/net/packet/af_packet.c
> @@ -93,12 +93,15 @@
>
>  /*
>     Assumptions:
> -   - if device has no dev->hard_header routine, it adds and removes ll header
> -     inside itself. In this case ll header is invisible outside of device,
> -     but higher levels still should reserve dev->hard_header_len.
> -     Some devices are enough clever to reallocate skb, when header
> -     will not fit to reserved space (tunnel), another ones are silly
> -     (PPP).
> +   - If the device has no dev->header_ops, there is no LL header visible
> +     above the device. In this case, its hard_header_len should be 0.
> +     The device may prepend its own header internally. In this case, its
> +     needed_headroom should be set to the space needed for it to add its
> +     internal header.
> +     For example, a WiFi driver pretending to be an Ethernet driver should
> +     set its hard_header_len to be the Ethernet header length, and set its
> +     needed_headroom to be (the real WiFi header length - the fake Ethernet
> +     header length).
>     - packet socket receives packets with pulled ll header,
>       so that SOCK_RAW should push it back.
>
> @@ -2937,10 +2940,14 @@ static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
>         skb_reset_network_header(skb);
>
>         err = -EINVAL;
> +       if (!dev->header_ops)
> +               WARN_ON_ONCE(dev->hard_header_len != 0);

Please make clear in the commit message that this is not just a
comment clarification.


>         if (sock->type == SOCK_DGRAM) {
>                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
>                 if (unlikely(offset < 0))
>                         goto out_free;
> +               WARN_ON_ONCE(offset > dev->hard_header_len);
> +               WARN_ON_ONCE(offset < dev->min_header_len);

This is not necessary. If worthwhile, this would belong inside
dev_hard_header itself.