Richard L. Hamilton wrote:
Are at least most supplied (even on the companion CD) or known freeware
clients smart enough to handle style 1 or 2 as needed? (I'm thinking ethereal
as a likely example, since for many purposes it's _much_ nicer than snoop;
but I haven't looked at the code of either.)
AFAIK, a client can issue a DL_INFO_REQ to find out, among
other things, whether the device it opened corresponds to a style 1 or style 2
provider; but it wouldn't surprise me if some programs simply take one or
the other for granted.
I've been through this run around once before, about 6 years ago. (When
I first pointed out that DLPI style 2 was a "bug", not a feature.)
At the time there were a few pieces left in Solaris that couldn't handle
style 2 properly (I think snoop was one of them), but I think they are
all fixed now.
Since Sun's network drivers have historically been mostly style 2, I trust the
conversion to style 1 as drivers are nemo-ified will be well publicized,
so that application developers would be reminded to ensure that their
code checks for style 1 vs style 2 providers and either doesn't or does issue
an attach, respectively.
I think this is well known.
In any case, Nemo exports both style 1 and style 2 nodes, just like
GLDv2. I certainly wish style 2 would go away, but converting to Nemo
apparently doesn't do this. (Though it can do so in the future by just
changing Nemo to stop creating them.)
On systems with lots of interfaces, will there be any problem with increased
consumption of major (?) device numbers with a style 1 provider?
Huh? No additional major numbers are consumed. Its still one major per
driver.
-- Garrett
_______________________________________________
networking-discuss mailing list
[email protected]
