On 2017/06/05 09:49, Reyk Floeter wrote: > > > Am 05.06.2017 um 09:35 schrieb Reyk Floeter <[email protected]>: > > > > > >> Am 05.06.2017 um 09:26 schrieb David Gwynne <[email protected]>: > >> > >> > >>> On 5 Jun 2017, at 17:05, Reyk Floeter <[email protected]> wrote: > >>> > >>> Well, not just muscle memory but the fact that some people including me > >>> had hostname.vlanX files without an explicit "vlan X" in it. > >> > >> hrm. yes. > >> > >> that means if you change the vnetid on such an interface at runtime, and > >> then re-run netstart later on to try and reset the stack, it wont > >> configure it back to the state it would be on boot. im not sure we support > >> running netstart after boot though, so maybe it doesnt matter. > >> > > > > We do and I run it all the time, for example > > > > # sh /etc/netstart vlan5 > > > > But I'd usually destroy a cloner before creating it again.
Yes me too. > Let me rephrase: > > I'm fine with the change as long as we never error out on vnetid/parent > transitions. The kernel should just cope with it, eg. running the following > commands in sequence should be fine and not throw EBUSY or similar: > > # ifconfig vlan5 create > # ifconfig vlan5 up > # ifconfig vlan5 vlandev em0 > # ifconfig vlan5 vlan 5 > # ifconfig vlan5 vlan 6 That is currently supported. > # ifconfig vlan5 vlandev em1 Support for changing vlandev was stopped last spring with the mpsafe vlan commit (it was a bit hairy in the first place, and removing it made the code much simpler).
