On 9/29/24 3:11 AM, Ido Schimmel wrote:
> On Sun, Sep 29, 2024 at 02:18:20AM -0400, Willem de Bruijn wrote:
>> From: Willem de Bruijn <willemb@xxxxxxxxxx>
>>
>> This reverts commit 504fc6f4f7f681d2a03aa5f68aad549d90eab853.
>>
>> dev_queue_xmit_nit is expected to be called with BH disabled.
>> __dev_queue_xmit has the following:
>>
>> /* Disable soft irqs for various locks below. Also
>> * stops preemption for RCU.
>> */
>> rcu_read_lock_bh();
>>
>> VRF must follow this invariant. The referenced commit removed this
>> protection. Which triggered a lockdep warning:
>
> [...]
>
>>
>> Fixes: 504fc6f4f7f6 ("vrf: Remove unnecessary RCU-bh critical section")
>> Link: https://lore.kernel.org/netdev/20240925185216.1990381-1-greearb@xxxxxxxxxxxxxxx/
>> Reported-by: Ben Greear <greearb@xxxxxxxxxxxxxxx>
>> Signed-off-by: Willem de Bruijn <willemb@xxxxxxxxxx>
>> Cc: stable@xxxxxxxxxxxxxxx
>
> Reviewed-by: Ido Schimmel <idosch@xxxxxxxxxx>
> Tested-by: Ido Schimmel <idosch@xxxxxxxxxx>
Reviewed-by: David Ahern <dsahern@xxxxxxxxxx>
>
> Thanks Willem!
>
> The reason my script from 504fc6f4f7f6 did not trigger the problem is
> that it was pinging the address inside the VRF, so vrf_finish_direct()
> was only called from the Rx path.
>
> If you ping the address outside of the VRF:
>
> ping -I vrf1 -i 0.1 -c 10 -q 192.0.2.1
>
> Then vrf_finish_direct() is called from process context and the lockdep
> warning is triggered. Tested that it does not trigger after applying the
> revert.
That case should be covered by the fcnal-test suite which does all
combinations of addresses.
