On 3/20/20 10:30 PM, Jakub Kicinski wrote:

On Fri, 20 Mar 2020 21:40:46 +0100 Daniel Borkmann wrote:

quoted

On 3/20/20 9:30 PM, Daniel Borkmann wrote:

quoted

On 3/20/20 9:48 AM, Toke Høiland-Jørgensen wrote:

quoted

Jakub Kicinski [off-list ref] writes:

quoted

On Thu, 19 Mar 2020 14:13:13 +0100 Toke Høiland-Jørgensen wrote:

quoted

From: Toke Høiland-Jørgensen <redacted>

While it is currently possible for userspace to specify that an existing
XDP program should not be replaced when attaching to an interface, there is
no mechanism to safely replace a specific XDP program with another.

This patch adds a new netlink attribute, IFLA_XDP_EXPECTED_FD, which can be
set along with IFLA_XDP_FD. If set, the kernel will check that the program
currently loaded on the interface matches the expected one, and fail the
operation if it does not. This corresponds to a 'cmpxchg' memory operation.

A new companion flag, XDP_FLAGS_EXPECT_FD, is also added to explicitly
request checking of the EXPECTED_FD attribute. This is needed for userspace
to discover whether the kernel supports the new attribute.

Signed-off-by: Toke Høiland-Jørgensen <redacted>

I didn't know we wanted to go ahead with this...

Well, I'm aware of the bpf_link discussion, obviously. Not sure what's
happening with that, though. So since this is a straight-forward
extension of the existing API, that doesn't carry a high implementation
cost, I figured I'd just go ahead with this. Doesn't mean we can't have
something similar in bpf_link as well, of course.

Overall series looks okay, but before we go down that road, especially given there is
the new bpf_link object now, I would like us to first elaborate and figure out how XDP
fits into the bpf_link concept, where its limitations are, whether it even fits at all,
and how its semantics should look like realistically given bpf_link is to be generic to
all program types. Then we could extend the atomic replace there generically as well. I
think at the very minimum it might have similarities with what is proposed here, but
from a user experience I would like to avoid having something similar in XDP API and
then again in bpf_link which would just be confusing..

Another aspect that falls into this atomic replacement is also that the programs can
actually be atomically replaced at runtime. Last time I looked, some drivers still do
a down/up cycle on replacement and hence traffic would be interrupted. I would argue
that such /atomic/ swap operation on bpf_link would cover a guarantee of not having to
perform this as well (workaround today would be a simple tail call map as entry point).

I don't think that's the case. Drivers generally have a fast path
for the active-active replace.

Up/Down is only done to remap DMA buffers and change RX buffer
allocation scheme. That's when program is installed or removed,
not replaced.

I know; though it seems not all adhere to that scheme sadly. I don't have that HW so can
only judge on the code, but one example that looked suspicious enough to me is qede_xdp().
It calls qede_xdp_set(), which does a qede_reload() for /every/ prog update. The latter
basically does ...

     if (edev->state == QEDE_STATE_OPEN) {
         qede_unload(edev, QEDE_UNLOAD_NORMAL, true);
         if (args)
             args->func(edev, args);               <-- prog replace here
         qede_load(edev, QEDE_LOAD_RELOAD, true);
         [...]
     }

... now that is one driver. I haven't checked all the others (aside from i40e/ixgbe/mlx4/
mlx5/nfp), but in any case it's also fixable in the driver w/o the extra need for bpf_link.

Thanks,
Daniel