Re: [PATCH net-next] net: ethtool: request rtnl_lock for ETHTOOL_MSG_PORT_GET
From: Linkui Xiao
Date: Sun Nov 15 2026 - 21:55:49 EST
Hi Stanislav,
On 2026/11/13 01:24, Stanislav Fomichev wrote:
On 11/11, Linkui Xiao wrote:
From: Linkui Xiao <xiaolinkui@xxxxxxxxxx>
port_prepare_data() resolves the requested port through
phy_link_topo_get_port(), which requires rtnl_lock to be held before it
reads topo->ports:
/* ports in the topo are RTNL protected, but may be accessed under
* netdev_lock for ops-locked devices. For now keep them under rtnl
* protection, as no ops-locked devices have phy_port capabilities
* yet.
*/
ASSERT_RTNL();
The dump side honours this, port_dump_one_dev() takes rtnl_lock()
itself. The doit side does not. ETHTOOL_MSG_PORT_GET is served by
ethnl_default_doit(), which keeps rtnl_lock only when the device is not
ops-locked, or when ethtool_nl_msg_needs_rtnl() asks for it:
need_rtnl = !netdev_need_ops_lock(req_info->dev) ||
ethtool_nl_msg_needs_rtnl(req_info->dev, cmd);
ethtool_nl_msg_needs_rtnl() has no case for ETHTOOL_MSG_PORT_GET and
falls through to "return false", so on a device that opts into the
instance lock (request_ops_lock, queue_mgmt_ops or net_shaper_ops)
PORT_GET runs with netdev_lock only. Such a device that does have a
link topology trips the ASSERT_RTNL() warning and reads topo->ports
without the protection the xarray is documented to need.
Add the missing case. Like TSCONFIG_GET/SET, PORT_GET has no
corresponding ethtool ioctl, so it is Netlink-only and always needs
rtnl_lock.
[..]
The bug is latent as no device opts into the instance lock while also
carrying a link topology, so it cannot be observed yet and the change is
compile-tested only.
Then why fix it if there is no bug? And phy_link_topo_add_port has
a big comment explaining the expectations?
You are right, and I withdraw the patch. It fixes nothing, and the comment
you point at is why.
phy_link_topo_get_port() returns before the ASSERT_RTNL() when the netdev
has no topology:
struct phy_link_topology *topo = dev->link_topo;
if (!topo)
return NULL;
ASSERT_RTNL();
and dev->link_topo is only ever allocated from phy_link_topo_get_or_alloc(),
i.e. from phy_link_topo_add_port() and phy_link_topo_add_phy(). Both refuse
an ops-locked device before they get that far:
/* Ports can now be queried without rtnl for ops-locked devices, which
* we don't support now as port retrieval is done under rtnl.
*/
if (WARN_ON_ONCE(dev->netdev_ops && netdev_need_ops_lock(dev)))
return -EOPNOTSUPP;
(phy_link_topo_add_phy() has an equivalent check.) So an ops-locked device
never has a topology to begin with: port_prepare_data() gets NULL back from
phy_link_topo_get_port() and returns -ENODEV, well before the assert.
With my patch applied it would take rtnl_lock() and still return -ENODEV,
so the patch is a no-op that only adds a lock - not something I want to
send. For devices that are not ops-locked need_rtnl is already true, so
the assert is satisfied today.
My commit message was wrong on the point that mattered: it claimed such a
device trips the warning, when the guard at the producer makes that state
unrepresentable. I should have read the expected design out of that comment
first - retrieval stays rtnl-only until an ops-locked device actually grows
phy_ports, and whoever wires that up will have to move the reader to the
instance lock instead of threading rtnl_lock() through ethtool.
The other topology readers gate their assert on the topology existing as
well (ethnl_req_get_phydev() uses phy_link_topo_empty()), so they are not
separate instances of the same issue.
The patch is already marked rejected in patchwork. Thanks for the pointer.
Regards,
Linkui