On Fri, 08 Mar 2013 16:23:53 -0500 Vlad Yasevich <[email protected]> wrote:
> On 03/08/2013 10:17 AM, Vlad Yasevich wrote: > > On 03/08/2013 12:43 AM, Stephen Hemminger wrote: > >> On Thu, 7 Mar 2013 16:28:45 -0500 > >> Vlad Yasevich <[email protected]> wrote: > >> > >>> The series adds an ability to configure the bridge into a > >>> non-primiscuous > >>> mode. Instead, it provides the ability to identitfy some set of bridge > >>> ports as uplinks and allows for MAC addresses to be programmed onto > >>> those ports. In case the port hardware does not support mac filter, > >>> that port will be placed in promiscuous mode. > >>> > >>> Default bridge operation continues to remain as "promiscuous". The new > >>> functionality has to be enabled via sysfs (similar to other bridge > >>> extensions). > >>> > >>> The uplink mode is implemented as a flag on a bridge port. The api to > >>> change that flag follows the existing api to enable/disable other > >>> existing > >>> flags. > >>> > >>> All comments are welcome. > >>> > >> > >> Can we make this a one step process and less visible to the user. > >> If user defines an uplink device, and the uplink device is capable of > >> filtering > >> (and what ever other pre-conditions people can think of), then the > >> bridge will > >> transparently switch to uplink/non-promisc mode. This can also be > >> used to trigger > >> edge only mode in RSTP in the future. > >> > >> Less knobs. > >> > > > > Ok. Let me see what I can do. > > So I started working through this and realized that this complicates the > code significantly. > > * I have to re-introduce the uplink-list since now I need to track > "filter capable" uplinks in addition to non-capable ones. > * The really nice and simple sysfs interface to set a flag turns into > something that duplicates code. > * The bridge port removal can effect the promiscuity setting of the > bridge if the last uplink is removed. > * We lose the ability to run a promisc edge bridge with uplinks. > > I am really starting wonder if this is any better? The changes > are much bigger and more complex while the functional flexibility is > reduced. Is it really worth removing a configuration knob? > > I've attached an in-progress patch to demonstrate the above. > > -vlad > > -vlad > > > > Thanks > > -vlad > > > A two step process is fine (uplink and promisc as seperate step), as long as invalid combinations are prevented. More options may lead to more combinations that are invalid.
