netdev - Re: [PATCH net-next] ipv6: Add sysctl for RA default route table number

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Wed, 30 Jun 2021 07:34:48 +0200
From:   Antony Antony <antony.antony@...unet.com>
To:     David Ahern <dsahern@...il.com>
CC:     Eyal Birger <eyal.birger@...il.com>, <antony.antony@...unet.com>,
        "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        "Hideaki YOSHIFUJI" <yoshfuji@...ux-ipv6.org>,
        David Ahern <dsahern@...nel.org>, <netdev@...r.kernel.org>,
        Christian Perle <christian.perle@...unet.com>
Subject: Re: [PATCH net-next] ipv6: Add sysctl for RA default route table
 number

On Tue, Jun 29, 2021 at 09:34:22 -0600, David Ahern wrote:
> On 6/29/21 8:05 AM, Eyal Birger wrote:
> > Hi Antony,
> > 
> > On 29/06/2021 15:53, Antony Antony wrote:
> >> Hi David,
> >>
> >> On Fri, Jun 25, 2021 at 22:47:41 -0600, David Ahern wrote:
> >>> On 6/25/21 1:04 AM, Antony Antony wrote:
> >>>> From: Christian Perle <christian.perle@...unet.com>
> >>>>
> >>>> Default routes learned from router advertisements(RA) are always placed
> >>>> in main routing table. For policy based routing setups one may
> >>>> want a different table for default routes. This commit adds a sysctl
> >>>> to make table number for RA default routes configurable.
> >>>>
> >>>> examples:
> >>>> sysctl net.ipv6.route.defrtr_table
> >>>> sysctl -w net.ipv6.route.defrtr_table=42
> >>>> ip -6 route show table 42
> >>>
> >>> How are the routing tables managed? If the netdevs are connected to a
> >>> VRF this just works.

Ah! I figured it out what you were hinting at! Sorry, I didn't know about
IFLA_VRF_TABLE attribute of link type vrf.

I also found the Documentation/networking/vrf.rst and red the commits
including the iproute2 commits. Thanks for the hint.

It looks like the vrf->tb_id is used for both v4 and v6 routing?
We are only looking at V6 routing, because the table only have v6 routes.

> >>
> >> The main routing table has no default route. Our scripts add routing
> >> rules
> >> based on interfaces. These rules use the specific routing table where
> 
> That's the VRF use case -- routing rules based on interfaces. Connect
> those devices to VRFs and the RA does the right thing.
> 
> >> the RA
> >> (when using SLAAC) installs the default route. The rest just works.
> > 
> > Could this be a devconf property instead of a global property? seems

yes adding to ipv6_devconf.cnf.ra_defrtr_tble is interesting.

I may propose a general solution that can replaces 
vrf_fib_table(retrun vrf->tb_id).
Do we need two? one for v6(in ipv6_devconf) and one for v4(bit map entry
in ipv4_devconf.cnf.data)?

there will be two ways to configure the defrtr_table option.
Frst using current ip link vrf table, when creating the devince, and
sysctl options.

e.g
ip link add vrf-blue type vrf table 10
would set set both
	ipv4_devconf.cnf.data[IPV4_DEVCONF_DEFRTR_TABLE] 
	ipv6_devconf.cnf.ra_defrtr_tble

sysctl will add additional 4 options
 net.ipv4.conf.all.defrtr_table
 net.ipv4.conf.eth0.defrtr_table
 net.ipv6.conf.all.defrtr_tabe
 net.ipv6.conf.eth0.defrtr_table

I guess replacing l3mdev_fib_table_rcu() would be complicated. 
I will think about a generic l3mdev_fib_table(). Any tips how to add
defrtr_table support for various link types?

> > like the difference would be minor to your patch but the benefit is that
> > setups using different routing tables for different policies could
> > benefit (as mentioned when not using vrfs).
> 
> exactly. This is definitely not a global setting, but a per device
> setting if at all.

A per device setting would work for our use case.

thanks,
-antony