On Fri, 2022-11-18 at 00:33 +0100, Paolo Abeni wrote:
> Recent changes in the veth driver caused a few regressions
> this series addresses a couple of them, causing oops.
>
> Paolo Abeni (2):
> veth: fix uninitialized napi disable
> veth: fix double napi enable
>
> drivers/net/veth.c | 6 ++++--
> 1 file changed, 4 insertions(+), 2 deletions(-)
@Xuan Zhuo: another option would be reverting 2e0de6366ac1 ("veth:
Avoid drop packets when xdp_redirect performs") and its follow-up
5e5dc33d5dac ("bpf: veth driver panics when xdp prog attached before
veth_open").
That option would be possibly safer, because I feel there are other
issues with 2e0de6366ac1, and would offer the opportunity to refactor
its logic a bit: the napi enable/disable condition is quite complex and
not used consistently mixing and alternating the gro/xdp/peer xdp check
with the napi ptr dereference.
Ideally it would be better to have an helper alike
napi_should_be_enabled(), use it everywhere, and pair the new code with
some selftests, extending the existing ones.
WDYT?
Side notes:
- Heng Qi address is bouncing
- the existing veth self-tests would have caught the bug addressed
here, if commit afef88e65554 ("selftests/bpf: Store BPF object files
with .bpf.o extension") would not have broken them meanwhile :(
/P
