Re: [PATCH] net: ipv6: don't generate link-local address in any addr_gen_mode

From: Mark Smith <markzzzsmith@gmail.com>
To: David Ahern <dsahern@gmail.com>
Cc: "Lorenzo Colitti" <lorenzo@google.com>,
	"Rocco Yue" <rocco.yue@mediatek.com>,
	"David S . Miller" <davem@davemloft.net>,
	"Hideaki YOSHIFUJI" <yoshfuji@linux-ipv6.org>,
	"David Ahern" <dsahern@kernel.org>,
	"Jakub Kicinski" <kuba@kernel.org>,
	"Matthias Brugger" <matthias.bgg@gmail.com>,
	"Linux NetDev" <netdev@vger.kernel.org>,
	lkml <linux-kernel@vger.kernel.org>,
	linux-arm-kernel@lists.infradead.org,
	linux-mediatek@lists.infradead.org, wsd_upstream@mediatek.com,
	rocco.yue@gmail.com, chao.song@mediatek.com,
	"Kuohong Wang (王國鴻)" <kuohong.wang@mediatek.com>,
	"Zhuoliang Zhang (张卓亮)" <zhuoliang.zhang@mediatek.com>
Subject: Re: [PATCH] net: ipv6: don't generate link-local address in any addr_gen_mode
Date: Mon, 13 Sep 2021 01:47:30 +1000	[thread overview]
Message-ID: <CAO42Z2w-N6A4DmubhQsg6WbaApG+7sy2SVRRxMXtaLrTKYyieQ@mail.gmail.com> (raw)
In-Reply-To: <6a8f0e91-225a-e2a8-3745-12ff1710a8df@gmail.com>

This is all going in the wrong direction. Link-local addresses are not
optional on an interface, all IPv6 enabled interfaces are required to
have one:

RFC4291, "IP Version 6 Addressing Architecture"

"2.1.  Addressing Model

All interfaces are required to have at least one Link-Local unicast
   address (see Section 2.8 for additional required addresses)."

Regards,
Mark.

On Fri, 10 Sept 2021 at 05:13, David Ahern <dsahern@gmail.com> wrote:
>
> On 9/9/21 12:20 AM, Lorenzo Colitti wrote:
> >> I think another addr_gen_mode is better than a separate sysctl. It looks
> >> like IN6_ADDR_GEN_MODE_STABLE_PRIVACY and IN6_ADDR_GEN_MODE_RANDOM are
> >> the ones used for RAs, so add something like:
> >>
> >> IN6_ADDR_GEN_MODE_STABLE_PRIVACY_NO_LLA,
> >> IN6_ADDR_GEN_MODE_RANDOM_NO_LLA,
> >
> > I think the real requirement here (which wasn't clear in this thread)
> > is that the network needs to control the interface ID (i.e., the
> > bottom 64 bits) of the link-local address, but the device is free to
> > use whatever interface IDs to form global addresses. See:
> > https://www.etsi.org/deliver/etsi_ts/129000_129099/129061/15.03.00_60/ts_129061v150300p.pdf
> >
> > How do you think that would best be implemented?
>
> There is an established paradigm for configuring how an IPv6 address is
> created or whether it is created at all - the IFLA_INET6_ADDR_GEN_MODE
> attribute.
>
> >
> > 1. The actual interface ID could be passed in using IFLA_INET6_TOKEN,
> > but there is only one token, so that would cause all future addresses
> > to use the token, disabling things like privacy addresses (bad).
> > 2. We could add new IN6_ADDR_GEN_MODE_STABLE_PRIVACY_LL_TOKEN,
> > IN6_ADDR_GEN_MODE_RANDOM_LL_TOKEN, etc., but we'd need to add one such
> > mode for every new mode we add.
> > 3. We could add a separate sysctl for the link-local address, but you
> > said that per-device sysctls aren't free.
>
> per-device sysctl's are one of primary causes of per netdev memory usage.
>
> Besides that there is no reason to add complexity by having a link
> attribute and a sysctl for this feature.
>
> > 4. We could change the behaviour so that if the user configures a
> > token and then sets IN6_ADDR_GEN_MODE_*, then we use the token only
> > for the link-local address. But that would impact backwards
> > compatibility.
> >
> > Thoughts?
>
> We can have up to 255 ADDR_GEN_MODEs (GEN_MODE is a u8). There is
> established code for handling the attribute and changes to it. Let's
> reuse it.