On 10/11/2015 02:21 PM, Ido Schimmel wrote:

Wed, Sep 30, 2015 at 09:16:54PM IDT, razor@blackwall.org wrote:

quoted

From: Nikolay Aleksandrov <redacted>

When a new port is being added we need to make vlgrp available after
rhashtable has been initialized and when removing a port we need to
flush the vlans and free the resources after we're sure noone can use
the port, i.e. after it's removed from the port list and synchronize_rcu
is executed.

Hi Nikolay,

Changing the order of port deinit breaks symmetry with the init
sequence. It also introduces a problem for switchdev drivers. Flushing
the VLANs clears HW VLAN filters and then, when port is unlinked from
bridge and CHANGEUPPER is received, port is configured to direct traffic
to CPU (as it's not offloaded anymore). Doing the reverse (like in this
patch) renders the port unusable.

Regarding the reason for this change, are you afraid that vlgrp will be
accessed in bridge's rx handler or xmit function after it's freed? If
so, maybe we can access it using RCU primitives? That way, both the rx
handler and xmit function (executed under RCU lock) will either have a
valid copy or not. Looking at previous iterations of this code, I see
that was the case with the 'net_port_vlans' struct.

I can start working on a fix if you agree with the proposed solution.

Thanks.

Hi,
Ah, I didn't know about this, I feared that something might rely on the
particular order of the operations but didn't have a way to test this one in
particular. Anyway, your proposed solution sounds good to me.

Thank you,
 Nik