netdev - Re: ipv6 ::1 and lo dev

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 for Android: free password hash cracker in your pocket

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

Message-ID: <20210819091652.GA22188@legohost>
Date:   Thu, 19 Aug 2021 12:16:52 +0300
From:   Oleg <lego12239@...dex.ru>
To:     Thorsten Glaser <t.glaser@...ent.de>
Cc:     netdev@...r.kernel.org
Subject: Re: ipv6 ::1 and lo dev

On Wed, Aug 18, 2021 at 07:47:37PM +0200, Thorsten Glaser wrote:
> On Wed, 18 Aug 2021, Oleg wrote:
> 
> > I try to replace ::1/128 ipv6 address on lo dev with ::1/112 to
> > access more than 1 address(like with ipv4 127.0.0.1/8). But i get
> 
> AIUI this is not possible in IPv6, only :: and ::1 are reserved,
> the rest of ::/96 is IPv4-compatible IPv6 addresses.

This is a big mistake of standards and, i think, we shouldn't conform to
standards in this area. Because standards conflicts with real life practice
(and needs) here.

I think, we can safely use ::/104 as analog of 127.0.0.1/8, because, AFAIK,
0.0.0.0/8 isn't used now in practice and thus there are no problems with
ipv4-mapped ipv6 addresses.

In any case, we should allow users to get an expected behaviour for lo dev
address - as with 127.0.0.1/8. I.e. when i remove ::1/128 and set ::1/104
i expect that ping ::2, ::3 and etc will work(may be only if some parameter for
"ip a add" is specified).

Unfortunately i can't suggest any patch, because my kernel programming level
is about 0 :-). May be anybody can do it?

> I never understood why you’d want more than one address for loopback
> anyway (in my experience, the more addresses a host has, the more
> confused it’ll get about which ones to use for what).

Besides already mentioned cases i say you about my 2 cases:

1. We have a service which serve many tunnels to different machines
  (think of it as many ssh -R X:127.0.0.N:Y to our service servers). Each
  machine is mapped to address:port(thus it constant between sessions).
  And we use many addresses from 127.0.0.1/8 :-).
2. We have many qemu VMs on several hardware hosts. We assign an address
  to every VM from 127.0.0.1/8 for comfort use of telnet/vnc. E.g. we
  have in /etc/hosts something like this(where third column is for
  host machine index and fourth column is for VM index):

  ...
  127.0.1.2       www1.vm
  127.0.1.3       www2.vm
  127.0.1.4       www3.vm
  127.0.1.5       dns1.vm
  127.0.1.5       dns2.vm
  127.0.1.6       mail1.vm
  ...

  and run qemu with:

  -serial telnet:dns1.vm:23,server,nowait -vnc dns1.vm:0

  This /etc/hosts syncronized between hardware hosts and if one if it fail
  we can migrate all VMs from it to another one and their addresses aren't
  intermixed.

-- 
Олег Неманов (Oleg Nemanov)