Re: VLAN_MTU (was Re: cvs commit: src/sys/dev/fxp if_fxp.c if_fxpvar.h)

From: Brooks Davis (brooks_at_one-eyed-alien.net)
Date: 05/27/04

  • Next message: Gleb Smirnoff: "shutdown node VS disconnect all hooks" 
    Date: Thu, 27 May 2004 08:02:28 -0700
To: Yar Tikhiy <yar@comp.chem.msu.su>

    On Thu, May 27, 2004 at 06:10:55PM +0400, Yar Tikhiy wrote:
    > [moving the discussion from the cvs lists to -net]
    >
    > On Wed, May 26, 2004 at 10:41:52AM +0400, Gleb Smirnoff wrote:
    > >
    > > Y> ng_vlan(4) could send a control command to ng_ether(4) instructing
    > > Y> the latter to increment the VLAN counter on the Ethernet interface
    > > Y> and toggle VLAN_MTU on if the counter value became equal to 1.
    > >
    > > Two comments:
    > >
    > > 1) Just note that it should increment VLAN counter on creating of
    > > any new VLAN hook.
    > > 2) There may be some itermediate nodes between ng_ether and ng_vlan,
    > > e.g. ng_tee(4), ng_etf(4), any custom traffic shaping or accounting
    > > node.
    > >
    > > Two deal with second issue some new mechanism should be introduced in netgraph,
    > > e.g. "broadcast" messages, which go down a hook spreading across all nodes,
    > > and only nodes with appropriate cookie will see this message delivered.
    > >
    > > Y> Another way I see is to drop automatic fiddling with VLAN_MTU in
    > > Y> the first place and implement an option for ifconfig(8) so that a
    > > Y> user/admin can control the capability WRT a particular case, e.g.,
    > > Y> disable it if a NIC displays erroneous behaviour in long frame mode.
    > >
    > > From my point of view this is a good idea.
    >
    > The longer I've been thinking of the issue, the more I'm inclined
    > to take the latter path. I believe that it would conform to the
    > good tradition of Unix to offer a user as much control as possible
    > and avoid doing "automagic" tricks behind user's back.
    >
    > Therefore I'd like to ask the community: Would anybody mind if
    > vlan(4) gave up playing with VLAN_MTU on parent interfaces
    > while a new option to ifconfig(8), say `vlanmtu', was introduced
    > so that a user could control the feature manually?

    This seems OK as long as you enabled VLAN_MTU by default as I see you
    have just done with fxp.

    -- Brooks 

    -- 
Any statement of the form "X is the one, true Y" is FALSE.
PGP fingerprint 655D 519C 26A7 82E7 2529  9BF0 5D8E 8BE9 F238 1AD4

  • Next message: Gleb Smirnoff: "shutdown node VS disconnect all hooks"