Message-ID: <ZGz5VWan/nROHxhc@lincoln>
Date:   Tue, 23 May 2023 19:35:17 +0200
From:   Larysa Zaremba <larysa.zaremba@...el.com>
To:     Jesper Dangaard Brouer <jbrouer@...hat.com>
CC:     <brouer@...hat.com>, <bpf@...r.kernel.org>,
        Stanislav Fomichev <sdf@...gle.com>,
        Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Andrii Nakryiko <andrii@...nel.org>,
        Jakub Kicinski <kuba@...nel.org>,
        Martin KaFai Lau <martin.lau@...ux.dev>,
        Song Liu <song@...nel.org>, Yonghong Song <yhs@...com>,
        John Fastabend <john.fastabend@...il.com>,
        KP Singh <kpsingh@...nel.org>, Jiri Olsa <jolsa@...nel.org>,
        Jesse Brandeburg <jesse.brandeburg@...el.com>,
        "Tony Nguyen" <anthony.l.nguyen@...el.com>,
        Anatoly Burakov <anatoly.burakov@...el.com>,
        Alexander Lobakin <alexandr.lobakin@...el.com>,
        Magnus Karlsson <magnus.karlsson@...il.com>,
        Maryam Tahhan <mtahhan@...hat.com>,
        <xdp-hints@...-project.net>, <netdev@...r.kernel.org>,
        <intel-wired-lan@...ts.osuosl.org>, <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH RESEND bpf-next 09/15] xdp: Add VLAN tag hint

On Tue, May 23, 2023 at 12:16:46PM +0200, Jesper Dangaard Brouer wrote:
> 
> 
> On 22/05/2023 17.48, Larysa Zaremba wrote:
> > On Mon, May 22, 2023 at 10:37:33AM +0200, Jesper Dangaard Brouer wrote:
> > > 
> > > 
> > > On 15/05/2023 18.09, Larysa Zaremba wrote:
> > > > On Mon, May 15, 2023 at 05:36:12PM +0200, Jesper Dangaard Brouer wrote:
> > > > > 
> > > > > 
> > > > > On 12/05/2023 17.26, Larysa Zaremba wrote:
> > > > > > Implement functionality that enables drivers to expose VLAN tag
> > > > > > to XDP code.
> > > > > > 
> > > > > > Signed-off-by: Larysa Zaremba <larysa.zaremba@...el.com>
> > > > > > ---
> > > > > [...]
> > > > > 
> > > > > > diff --git a/net/core/xdp.c b/net/core/xdp.c
> > > > > > index 41e5ca8643ec..eff21501609f 100644
> > > > > > --- a/net/core/xdp.c
> > > > > > +++ b/net/core/xdp.c
> > > > > > @@ -738,6 +738,30 @@ __bpf_kfunc int bpf_xdp_metadata_rx_hash(const struct xdp_md *ctx, u32 *hash,
> > > > > >     	return -EOPNOTSUPP;
> > > > > >     }
> > > > > 
> > > > > Remember below becomes part of main documentation on HW metadata hints:
> > > > >    - https://kernel.org/doc/html/latest/networking/xdp-rx-metadata.html
> > > > > 
> > > > > Hint compiling locally I use:
> > > > >    make SPHINXDIRS="networking" htmldocs
> > > > > 
> > > > > > +/**
> > > > > > + * bpf_xdp_metadata_rx_ctag - Read XDP packet inner vlan tag.
> > > > > 
> > > > > Is bpf_xdp_metadata_rx_ctag a good function name for the inner vlan tag?
> > > > > Like wise below "stag".
> > > > > 
> > > > > I cannot remember if the C-tag or S-tag is the inner or outer vlan tag.
> > > > > 
> > > > > When reading BPF code that use these function names, then I would have
> > > > > to ask Google for help, or find-and-read this doc.
> > > > > 
> > > > > Can we come-up with a more intuitive name, that e.g. helps when reading
> > > > > the BPF-prog code?
> > > > 
> > > > Well, my reasoning for such naming is that if someone can configure s-tag
> > > > stripping in ethtool with 'rx-vlan-stag-hw-parse', they shouldn't have any
> > > > problem with understanding those function names.
> > > > 
> > > 
> > > Naming is hard.  My perspective is conveying the meaning without having
> > > to be knowledgeable about ethtool VLAN commands.  My perspective is a
> > > casual BPF-programmer that reads "bpf_xdp_metadata_rx_stag()".
> > > Hopefully we can choose a name that says "vlan" somewhere, such that the
> > > person reading this doesn't have to lookup and find the documentation to
> > > deduct this code is related to VLANs.
> > > 
> > > > One possible improvement that comes to mind is maybe (similarly ethtool) calling
> > > > c-tag just 'tag' and letting s-tag stay 'stag'. Because c-tag is this default
> > > > 802.1q tag, which is supported by various hardware, while s-tag is significantly
> > > > less widespread.
> > > > 
> > > > But there are many options, really.
> > > > 
> > > > What are your suggestions?
> > > > 
> > > 
> > > One suggestion is (the symmetrical):
> > >   * bpf_xdp_metadata_rx_vlan_inner_tag
> > >   * bpf_xdp_metadata_rx_vlan_outer_tag
> > > 
> > > As you say above the first "inner" VLAN tag is just the regular 802.1Q
> > > VLAN tag.  The concept of C-tag and S-tag is from 802.1ad that
> > > introduced the concept of double tagging.
> > > 
> > > Thus one could argue for shorter names like:
> > >   * bpf_xdp_metadata_rx_vlan_tag
> > >   * bpf_xdp_metadata_rx_vlan_outer_tag
> > > 
> > 
> > AFAIK, outer tag is a broader term, it's pretty often used for stacked 802.1Q
> > headers. I can't find what exactly is an expected behavior for rxvlan and
> > rx-vlan-stag-hw-parse in ethtool, but iavf documentation states that rxvlan
> > "enables outer or single 802.1Q VLAN stripping" and rx-vlan-stag-hw-parse
> > "enables outer or single 802.1ad VLAN stripping". This is in consistent with how
> > ice hardware behaves. More credible sources would be welcome.
> > 
> 
> It would be good to figure out how other hardware behaves.
> 
> The iavf doc sounds like very similar behavior from both functions, just
> 802.1Q vs 802.1ad.
> Sounds like both will just pop/strip the outer vlan tag.
> I have seen Ethertype 802.1Q being used (in practice) for double tagged
> packets, even-though 802.1ad should have been used to comply with the
> standard.
> 
> > What about:
> >    * bpf_xdp_metadata_rx_vlan_tag
> >    * bpf_xdp_metadata_rx_vlan_qinq_tag
> > 
> 
> This sounds good to me.
> 
> I do wonder if we really need two functions for this?
> Would one function be enough?
> 
> Given the (iavf) description, the functions basically does the same.
> Looking at your ice driver implementation, they could be merged into one
> function, as it is the same location in the descriptor.
>

This design was very debatable in the first place.
I looked at different in-tree driver implementations of NETIF_F_HW_VLAN_STAG_RX
feature once more. Among those I could comprehend, seems like none has c-tag and 
s-tag stored separately. Actually, there are 2 situations:

1. (ex. mlx4) HW always strips outer or single VLAN tag, without distinction 
between 802.1Q and 802.1ad. TPID in such case is deduced from descriptor. 
NETIF_F_HW_VLAN_STAG_RX and NETIF_F_HW_VLAN_CTAG_RX must be enabled together.

2. (ex. ice) HW strips outer or single VLAN tag with a configured TPID. In such 
case descriptor doesn't carry info about TPID, because it's the same for all 
stripped tags. C-tag and s-tag stripping are mutually exclusive.
Example:
 - 802.1Q double VLAN, with s-tag stripping enabled, packet arrives 
   untouched, with c-tag stripping outermost tag gets stripped.
 - 802.1ad+802.1Q, with s-tag stripping enabled, 802.1ad header gets stripped,
   with c-tag stripping, packet arrives untouched.

Obviously, I can be sure only about our hardware.

Long story short, probably re-inventing the wheel wasn't a good idea on my part. 
Now I am much more inclined to just copy the logic from skb, so function would 
look like this:

  bpf_xdp_metadata_rx_vlan_tag(const struct xdp_md *ctx, __u16 *vlan_tag,
			       __u16 *tpid);

Maybe some applications would make use of just:

  bpf_xdp_metadata_rx_vlan_tag(const struct xdp_md *ctx, __u16 *vlan_tag);

Both of the above functions would return information about outermost tag, if was 
stripped. Would have to think about the naming.

Comments are welcome!

> > > 
> > > > > 
> > > > > > + * @ctx: XDP context pointer.
> > > > > > + * @vlan_tag: Return value pointer.
> > > > > > + *
> > > > > 
> > > > > IMHO right here, there should be a description.
> > > > > 
> > > > > E.g. for what a VLAN "tag" means.  I assume a "tag" isn't the VLAN id,
> > > > > but the raw VLAN tag that also contains the prio numbers etc.
> > > > > 
> > > > > It this VLAN tag expected to be in network-byte-order ?
> > > > > IMHO this doc should define what is expected (and driver devel must
> > > > > follow this).
> > > > 
> > > > Will specify that.
> > > > 
> > > > > 
> > > > > > + * Returns 0 on success or ``-errno`` on error.
> > > > > > + */
> > > > > > +__bpf_kfunc int bpf_xdp_metadata_rx_ctag(const struct xdp_md *ctx, u16 *vlan_tag)
> > > > > > +{
> > > > > > +	return -EOPNOTSUPP;
> > > > > > +}
> > > > > > +
> > > > > > +/**
> > > > > > + * bpf_xdp_metadata_rx_stag - Read XDP packet outer vlan tag.
> > > > > > + * @ctx: XDP context pointer.
> > > > > > + * @vlan_tag: Return value pointer.
> > > > > > + *
> > > 
> > > (p.s. Googling I find multiple definitions of what the "S" in S-tag
> > > means. The most reliable or statistically consistent seems to be
> > > "Service tag", or "Service provider tag".)
> > > 
> > > The description for the renamed "bpf_xdp_metadata_rx_vlan_outer_tag"
> > > should IMHO explain that the outer VLAN tag is often refered to as the S-tag
> > > (or Service-tag) in Q-in-Q (802.1ad) terminology.  Perhaps we can even spell
> > > out that some hardware support (and must be configured via ethtool) to
> > > extract this stag.
> > > 
> > > A dump of the tool rx-vlan related commands:
> > > 
> > >    $ ethtool -k i40e2 | grep rx-vlan
> > >    rx-vlan-offload: on
> > >    rx-vlan-filter: on [fixed]
> > >    rx-vlan-stag-hw-parse: off [fixed]
> > >    rx-vlan-stag-filter: off [fixed]
> > > 
> [...]
> 

Powered by blists - more mailing lists