ietf
[Top] [All Lists]

Re: [v6ops] draft-ietf-6man-rfc4291bis prohibiting non-/64 subnets

2017-02-24 12:09:05


Le 24/02/2017 à 18:53, Christopher Morrow a écrit :


On Fri, Feb 24, 2017 at 12:09 PM, Alexandre Petrescu
<alexandre(_dot_)petrescu(_at_)gmail(_dot_)com 
<mailto:alexandre(_dot_)petrescu(_at_)gmail(_dot_)com>> wrote:



    Le 24/02/2017 à 17:27, Christopher Morrow a écrit :



        On Fri, Feb 24, 2017 at 10:48 AM, Alexandre Petrescu
        <alexandre(_dot_)petrescu(_at_)gmail(_dot_)com
        <mailto:alexandre(_dot_)petrescu(_at_)gmail(_dot_)com>
        <mailto:alexandre(_dot_)petrescu(_at_)gmail(_dot_)com
        <mailto:alexandre(_dot_)petrescu(_at_)gmail(_dot_)com>>>
        wrote:



        Le 24/02/2017 à 15:59, Christopher Morrow a écrit :



        On Fri, Feb 24, 2017 at 9:09 AM, Alexandre Petrescu
        <alexandre(_dot_)petrescu(_at_)gmail(_dot_)com
        <mailto:alexandre(_dot_)petrescu(_at_)gmail(_dot_)com>
        <mailto:alexandre(_dot_)petrescu(_at_)gmail(_dot_)com
        <mailto:alexandre(_dot_)petrescu(_at_)gmail(_dot_)com>>
        <mailto:alexandre(_dot_)petrescu(_at_)gmail(_dot_)com
        <mailto:alexandre(_dot_)petrescu(_at_)gmail(_dot_)com>
        <mailto:alexandre(_dot_)petrescu(_at_)gmail(_dot_)com
        <mailto:alexandre(_dot_)petrescu(_at_)gmail(_dot_)com>>>> wrote:


        A question to Windows is the following: what prefixlen does it set
        when the end user manually assigns an address on an interface
        without specifiying a prefixlen?


        I don't think this matters... 'end users' will in almost all cases
        just attach and get connectivity.


        Let me go next in cycle: what does linux do when one ifconfig add an
        IPv6 address without telling the plen?  Is it adding an entry in the
        rt table?  Which plen?  Is that plen normal?


        still don't think this matters. If a user messes up what they type,
        they messed up. if the instructions aren't complete, they aren't
        complete and there will be mistakes. An interface configuration
        requires all proper parameters be set, or problems will arise.
        Assumptions about current and future behavior are proven wrong time
        and again.

        non-deterministic behavior over time is the hallmark of this
        space... please do not rely on defaults for hand-managed/bespoke
        configurations if you expect things to work reliably and repeatably.


    That means the following: Windows please make the plen parameter
    mandatory.  Dont leave it optional only to subversively set it to 64.
    That's a software bug because nobody asks you to set the plen to 64 when
    the end user does not specify one.

    I guess the same bug is is in BSD, linux and what have you.


yes, I think so.. for linux:
$ sudo ip -6 addr add 2001:700:4::1 dev em1

gets me:
 inet6 2001:700:4::1/128 scope global

That address is good, but how about the routing table? Does it get an entry like 2001:700:4::/64?

so that actually seems correct.



        If they are in a place where someone says: "Hey, you should go
        if/ipconfig ...." then .. they are 'consenting adults' and can do
        whatever they please.


        I assume you assume that ip/ifconfig by consenting adults means the
        adults type a plen in the CLI, right?


        sure, or a script/program/etc does this, it's not important how the
        'ifconfig' happens, it's important that when it happens the right
        parameters are passed to the 'ifconfig' command.


    I agree.  If 'ifconfig' silently assumes 64 then that assumption is
    wrong.  The ifconfig programmer should take that as a software bug.
    It's not an RFC that requires them to put 64 there.


ok


        That makes it mandatory that the CLI _requires_ a plen, right?  That
        CLI should not allow silence for a plen parameter.


        sure, or you are at the mercy of the implementor of that command:
        Today I like /64! It's tomorrow and now I like /62!!

        don't rely on defaults.


    I agree.

        Because silent plen means 64.  And I dont think it's right to assume
        a by-default 64 plen.  Because many people think 64 is right and
        others think it's wrong, there does not seem to be a commonly agreed
        'by default' value for plen.


        correct, so.. be specific in your configuration effort please...
        which again, means that the 'what is the default?" conversation is
        moot.


    I am not asking what is the default.

    I am saying that apparently numerous implementations out there consider
    the default to be 64.  This consideration is wrong.


yes, I agree with you here.


    A 'default' value is something that everybody agrees with.  For example,
    one can leave out '::' from a command line adding a default route and
    just say 'default'.  There is an agreed standard that says that
    'default' is '::'.

    But there is by far no single standard that says the default prefix
    length is 64.

    That's why it's a bug.

    It's like boxes with pre-defined passwords admin/admin.  The local
    programmer imagined it a good 'default' but never asked around the
    validity of such assumption.  And that creates problems.


ok, cool, i think we agree on all of this at least :)

:-)

Alex





    Alex




        Alex



        again the proposed text (now 175+ messages back) really covers this
        already..




<Prev in Thread] Current Thread [Next in Thread>