Jakub Kicinski <kuba@xxxxxxxxxx> writes: > On Tue, 24 Mar 2020 17:54:07 -0700 Andrii Nakryiko wrote: >> On Tue, Mar 24, 2020 at 11:13 AM Toke Høiland-Jørgensen <toke@xxxxxxxxxx> wrote: >> > >> > From: Toke Høiland-Jørgensen <toke@xxxxxxxxxx> >> > >> > 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_ID, 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. >> > Setting the new attribute with a negative value means that no program is >> > expected to be attached, which corresponds to setting the UPDATE_IF_NOEXIST >> > flag. >> > >> > A new companion flag, XDP_FLAGS_EXPECT_ID, is also added to explicitly >> > request checking of the EXPECTED_ID attribute. This is needed for userspace >> > to discover whether the kernel supports the new attribute. >> >> Doesn't it feel inconsistent in UAPI that FD is used to specify XDP >> program to be attached, but ID is used to specify expected XDP >> program? Especially that the same cgroup use case is using >> (consistently) prog FDs. Or is it another case where XDP needs its own >> special way? > > There was a comment during review of v1, I wish you spoke up then. > > The prog ID is what dump returns, so the consistency can go either way > (note that this API predates object IDs). Since XDP uses IDs internally > it's just simpler to take prog ID. > > But it's a detail, so if you feel strongly I don't really mind. Using an FD instead of an ID does make this more extensible (such as supporting bpf_link FDs in the future; see my other reply to Alexei). So I'll respin this, and switch it back to EXPECTED_FD. -Toke
