On Mon, 2005-25-07 at 13:54 -0700, David S. Miller wrote: > From: Thomas Graf <[EMAIL PROTECTED]> > Date: Mon, 25 Jul 2005 22:16:21 +0200 > > > * David S. Miller <[EMAIL PROTECTED]> 2005-07-25 13:14 > > > We removed INPUT_DEV from the meta ematch last week remember? > > > > Sure, but we can readd it once its clear how it will be used. > > What I meant is that the meta ematch doesn't care as it will > > provide access by name and by ifindex anyway. >
Most of the qdisc level code use ifindex and do byname checks first from user space to grab the ifindex. > I see. > > I was also just thinking about the semantic change that will > result if rules are specified by device name, and we start > doing match comparisons using ifindex. I don't think we can > legally do things that way. > the name comparison is useful in general if you want to do many devices per regexp (example eth* or ppp[0-11] etc). Its harder to do so since ifindex ranges cant be guaranteed to be contigous. > If a device goes down, and comes back up with the same name > but a different ifindex, we should still match it with existing > rules based upon name. The ifindex will change if you unload a module; If you try hard enough you will hit a point where things will wrap around; you just have to consume the 2^31 ifindices ;-> Name is not totaly sane either: I think one of the fscked s/ware i have seen is pppd that tries to recycle names immediately with the same name but different ifindex. It should probably use ppp<ifindex> for name. In theory, one should be able from user space to really configure the ifindex; we dont allow for it even for devices that are virtual. > So we'd end up with stuff like: > > dev = dev_get_by_index(ifindex); > match = 0; > if (!strcmp(p->name, dev->name)) > match = 1; > dev_put(dev); > > at which point we might as well just grab a device reference in > skb->input_dev since that's simpler and a lot cheaper and easier to > verify. BTW, I just glanced at your suggested patch. If we are not going to use ifindex we aint saving the 4 bytes on 64 bit machines. And we have to keep refcounts and release devices. 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
