On 7/15/22 3:37 PM, Luis Chamberlain wrote:

On Fri, Jul 15, 2022 at 02:00:36PM -0600, Jens Axboe wrote:

quoted

I did author the basic framework of it, but Kanchan took over driving it
to completion and was the one doing the posting of it at that point.

And credit where due, that was a significant undertaking, and great
collaboration.

Definitely, the completion bit is usually the longest pole in the
endevaour.

quoted

It's not like I merge code I'm not aware of, we even discussed it at
LSFMM this year and nobody brought up the LSM oversight. Luis was there
too I believe.

I brought it up as a priority to Kanchan then. I cringed at not seeing it
addressed, but as with a lot of development, some things get punted for
'eventually'. What I think we need is more awareness of the importance of
addressing LSMs and making this a real top priority, not just, 'sure',
or 'eventually'. Without that wide awareness even those aware of its
importance cannot help make LSM considerations a tangible priority.

Not sure if this is a generic problem, or mostly on our side. uring_cmd
is a bit of an exception, since we don't really add a lot of non-syscall
accessible bits to begin with. But in general there's for sure more
action there than in other spots. I'm hopeful that this will be more on
top of our minds when the next time comes around.

For uring_cmd, extensions will most likely happen. At least I have some
in mind. We might want to make the control more finegrained at that
point, but let's deal with that when we get there.

We can do this with ksummit, or whatever that's called these days,
because just doing this at security conferences is just getting people
preaching to the choir.

Don't think anyone disagrees that it needs to get done, and there's not
much process to hash out here other than one subsystem being aware of
another ones needs. Hence don't think the kernel summit or maintainers
summit is doing to be useful in that regard. Just my 2 cents.

-- 
Jens Axboe