Sebastien Roy wrote:
On Tue, 2006-04-11 at 10:02 -0400, Peter Memishian wrote:
> I just wanted to point out; mac_alloc() always allocates a
> mac_register_t of size "sizeof (mac_register_t)", and that's compiled
> into the mac module. The mac module always knows the correct size, so I
> don't think this would be an issue. It's the driver that never really
> knows the true size of the structure.
Ah yes, my mistake. So not to dive too deeply into old discussions, but
let's say we later expand the mac_register_t, and someone compiles a
driver against the new version. They then put this new driver onto an
older system -- what prevents the driver from trashing memory when it
fills in the mac_register_t and writes off the end of the structure
allocated by mac_alloc()? By the time we discover the version number
mismatch in mac_register(), it's too late, no?
It would be too late, and bad things could happen. I can think of two
things to say about this:
1. I don't believe that putting binaries compiled on one version of
Solaris onto an older version of Solaris is supported or guaranteed to
work in any case.
You mean this just for the kernel, I hope, because we make a lot of
noise about ABI/API compatibility across releases to the extent that
a lot of people expect binaries from Solaris 9 to "just work" on
Solaris 10.
For kernel drivers, with a STREAMS based NIC driver, I've been
able to use 3rd party drivers for Solaris 2.5 on 2.4 - 2.7 because
the interfaces in use are common/stable. If this is just accidental
then it is a nice accident of a well written driver.
Darren
_______________________________________________
networking-discuss mailing list
[email protected]
