svn commit: r205024 - head/sys/net

[Robert N. M. Watson]

On Mar 12, 2010, at 8:11 AM, Qing Li wrote:

> I like Julian's suggestion because it is simple and very low risk.
> And there isn't a need to check for interface type any more.
> Here is why:

Re-reading this e-mail: perhaps you mean Juli, not Julian?

Robert

> 
> 1. The interfaces that are popular and modern are already supporting
>    link_state. So for these drivers, and there are just a few, I will go set
>    its if_flags to include "can change link_state".
> 
> 2. For the existing dated drivers, because that flag bit is never set,
>    no check is done.
> 
> 3. In the mean time, we try to convert the drivers progressively.
> 
> 4. If one wants to do ECMP and not having packets go into a black
>    hole when the physical link is down, that person can ping the ML
>    and ask for driver compatibility list. If we haven't converted that
>    particular driver by then, we will update the driver if it's capable
>    at that time.
> 
> -- Qing
> 
> 
> On Fri, Mar 12, 2010 at 12:00 AM, Robert N. M. Watson
> <rwatson at freebsd.org> wrote:
>> 
>> On Mar 12, 2010, at 7:52 AM, Qing Li wrote:
>> 
>>>> Is there any way we can pick up via an assertion that an interface driver has failed to implement this functionality? This has never been a historic requirement, so I suspect there are a lot of drivers floating around that fail to meet the requirement. Also, is this for IFT_ETHER only, or also other link types?
>>> 
>>> Not sure if I get the assertion suggestion. How would an assertion help here ?
>> 
>> I think my proposal is similar to what Juli is suggesting:
>> 
>> - Define a new interface capability for link state detection.
>> - If a packet is sent or received on the interface, the capability is set, but the link state hasn't been set, panic.
>> - If a packet is sent received on the interface, the capability isn't set, and the link state has been set, panic.
>> 
>> That way the system blows up nicely and immediately, rather than dhclient simply never working, etc. Also, that way, testing for link state support is done at a point when we know the interface is live (a packet is sent or received).
>> 
>> Finally, it means that code interested in link state isn't testing for one of (n) IFT_ types it thinks should have link state, but instead testing specifically whether the driver declares link state support. Of course, then you have to decide how to behave if a configured interface ECMP is running on doesn't support link state: the answer there is probably to assume it is always up, which would make this work for all those drivers that current fail to implement it. And if the hardware can't support link state, which some historic (and perhaps future) link types can't, things still work.
>> 
>> Robert