On Sun, 2005-04-09 at 20:20 +0200, Patrick McHardy wrote:
> jamal wrote:
> > There is no need to check for NULL; if you got that far dev cannot be
> > NULL (refer to the first check for parm->ifindex). The device could not
> > have disapeared since we are protected by rtnl semaphore.
>
> It doesn't check for the new device, it checks whether an old device if
> needs to be released. This can't be done unconditionally since dev_put
> doesn't handle NULL pointers.
>
I am not sure i followed 100% what you said but i will take your word
for it.
> > You are right - did you mean NETDEV_UNREGISTER notification only,
> > correct?
>
> Correct.
[..]
> How do you propose to react? Kill the action or set the device
> to NULL and handle it in tcf_mirred()?
The setting to NULL and handle in tcf_mirred() seems simpler but would
require we put extra checks in tcf_mirred() - which is the fast path ...
Unless you can somehow squeeze it into that "if (!(dev->flags&IFF_UP) )
{" check ...
The "RightThing(tm)" would be to walk all instances of mirred and nuke
them right away if they are pointing to devices that dont exist - this
would be analogous behavior to the way fib routes are forced out if next
hop device disappears etc. But this would be tricky since there may be a
different device pointing to a mirred action in its list of actions
and nuking mirred would mean trouble.
So my thinking is: Whatever you find to be easy - any solution would be
better than nothing thats in there at the moment.
cheers,
jamal
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
