Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20200828070752.54444-1-xie.he.0141@gmail.com> <m3pn7b6opa.fsf@t19.piap.pl>
In-Reply-To: <m3pn7b6opa.fsf@t19.piap.pl>
From:   Xie He <xie.he.0141@gmail.com>
Date:   Fri, 28 Aug 2020 14:05:32 -0700
Message-ID: <CAJht_EOk2_L-77KDDEJTcfqhw48X0ZMA2PKdLG4+LXHAAtNtsw@mail.gmail.com>
Subject: Re: [PATCH net] drivers/net/wan/hdlc_cisco: Add hard_header_len
To:     =?UTF-8?Q?Krzysztof_Ha=C5=82asa?= <khalasa@piap.pl>
Cc:     "David S. Miller" <davem@davemloft.net>,
        Jakub Kicinski <kuba@kernel.org>,
        Linux Kernel Network Developers <netdev@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Martin Schiller <ms@dev.tdt.de>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Aug 28, 2020 at 3:37 AM Krzysztof Ha=C5=82asa <khalasa@piap.pl> wro=
te:
>
> OTOH hdlc_setup_dev() initializes hard_header_len to 16,
> but in this case I guess 4 bytes are better.
>
> Acked-by: Krzysztof Halasa <khc@pm.waw.pl>

Thank you, Krzysztof!

Actually I'm thinking about changing the default value of 16 in hdlc.c to 0=
.

I think a driver should always keep its hard_header_len consistent
with its header_ops functions. If a driver doesn't have header_ops,
its hard_header_len should be set to 0. This makes the driver able to
be correctly used with AF_PACKET sockets.

In net/packet/af_packet.c, in the function packet_snd, for
AF_PACKET/DGRAM sockets, it would reserve a headroom of
hard_header_len for the skb, and call dev_hard_header (which calls the
header_ops->create function) to fill in the headroom, but for
AF_PACKET/RAW sockets, it would not reserve the headroom of
hard_header_len, and will check (in function dev_validate_header)
whether the user has provided the header of length hard_header_len. So
I think hard_header_len should be kept consistent with header_ops to
make the driver able to work correctly with af_packet.c.

If the driver really needs to use additional header space outside of
the header_ops->create function, it should request that header space
in dev->needed_headroom instead of hard_header_len. This avoids the
complex header processing in af_packet.c but still gets the header
space reserved.

Currently for the 6 HDLC protocol drivers, hdlc_ppp sets
hard_header_len and the value is consistent with its header_ops,
hdlc_raw_eth sets both hard_header_len and header_ops correctly with
the ether_setup function, hdlc_x25 has been previously changed by me
to set hard_header_len to 0 because it doesn't have header_ops, and
this patch would make hdlc_cisco set its hard_header_len to the value
consistent with its header_ops. This leaves us hdlc_raw and hdlc_fr. I
see that both of these 2 drivers don't set hard_header_len when
attaching the protocol, so they will use the default value of 16. But
because both of these drivers don't have header_ops, I think their
hard_header_len should be set to 0. So I think maybe it's better to
change the default value in hdlc.c to 0 and let them take the default
value of 0.

What do you think?

Thanks!

Xie He