| Message ID | 20211203101318.435618-1-atenart@kernel.org (mailing list archive) |
|---|---|
| State | Superseded |
| Commit | dde91ccfa25fd58f64c397d91b81a4b393100ffa |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [net] ethtool: do not perform operations on net devices being unregistered | expand |
On 03.12.21 12:13, Antoine Tenart wrote: > There is a short period between a net device starts to be unregistered > and when it is actually gone. In that time frame ethtool operations > could still be performed, which might end up in unwanted or undefined > behaviours[1]. > > Do not allow ethtool operations after a net device starts its > unregistration. This patch targets the netlink part as the ioctl one > isn't affected: the reference to the net device is taken and the > operation is executed within an rtnl lock section and the net device > won't be found after unregister. > > [1] For example adding Tx queues after unregister ends up in NULL > pointer exceptions and UaFs, such as: > > BUG: KASAN: use-after-free in kobject_get+0x14/0x90 > Read of size 1 at addr ffff88801961248c by task ethtool/755 > > CPU: 0 PID: 755 Comm: ethtool Not tainted 5.15.0-rc6+ #778 > Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.14.0-4.fc34 04/014 > Call Trace: > dump_stack_lvl+0x57/0x72 > print_address_description.constprop.0+0x1f/0x140 > kasan_report.cold+0x7f/0x11b > kobject_get+0x14/0x90 > kobject_add_internal+0x3d1/0x450 > kobject_init_and_add+0xba/0xf0 > netdev_queue_update_kobjects+0xcf/0x200 > netif_set_real_num_tx_queues+0xb4/0x310 > veth_set_channels+0x1c3/0x550 > ethnl_set_channels+0x524/0x610 > > Fixes: 041b1c5d4a53 ("ethtool: helper functions for netlink interface") > Suggested-by: Jakub Kicinski <kuba@kernel.org> > Signed-off-by: Antoine Tenart <atenart@kernel.org> > --- > > Following the discussions in those threads: > - https://lore.kernel.org/all/20211129154520.295823-1-atenart@kernel.org/T/ > - https://lore.kernel.org/all/20211122162007.303623-1-atenart@kernel.org/T/ > > net/ethtool/netlink.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/net/ethtool/netlink.c b/net/ethtool/netlink.c > index 38b44c0291b1..96f4180aabd2 100644 > --- a/net/ethtool/netlink.c > +++ b/net/ethtool/netlink.c > @@ -40,7 +40,8 @@ int ethnl_ops_begin(struct net_device *dev) > if (dev->dev.parent) > pm_runtime_get_sync(dev->dev.parent); > > - if (!netif_device_present(dev)) { > + if (!netif_device_present(dev) || > + dev->reg_state == NETREG_UNREGISTERING) { > ret = -ENODEV; > goto err; > } > Wondering if other places would also benefit from a netif_device_detach() in the unregistration sequence ...
On Mon, 6 Dec 2021 11:46:35 +0200 Julian Wiedmann wrote: > On 03.12.21 12:13, Antoine Tenart wrote: > > There is a short period between a net device starts to be unregistered > > and when it is actually gone. In that time frame ethtool operations > > could still be performed, which might end up in unwanted or undefined > > behaviours[1]. > > > > Do not allow ethtool operations after a net device starts its > > unregistration. This patch targets the netlink part as the ioctl one > > isn't affected: the reference to the net device is taken and the > > operation is executed within an rtnl lock section and the net device > > won't be found after unregister. > > [...] > > +++ b/net/ethtool/netlink.c > > @@ -40,7 +40,8 @@ int ethnl_ops_begin(struct net_device *dev) > > if (dev->dev.parent) > > pm_runtime_get_sync(dev->dev.parent); > > > > - if (!netif_device_present(dev)) { > > + if (!netif_device_present(dev) || > > + dev->reg_state == NETREG_UNREGISTERING) { > > ret = -ENODEV; > > goto err; > > } > > > > Wondering if other places would also benefit from a netif_device_detach() > in the unregistration sequence ... Sounds like a good idea but maybe as a follow up to net-next? The likelihood of that breaking things is low, but non-zero.
Hello: This patch was applied to netdev/net.git (master) by Jakub Kicinski <kuba@kernel.org>: On Fri, 3 Dec 2021 11:13:18 +0100 you wrote: > There is a short period between a net device starts to be unregistered > and when it is actually gone. In that time frame ethtool operations > could still be performed, which might end up in unwanted or undefined > behaviours[1]. > > Do not allow ethtool operations after a net device starts its > unregistration. This patch targets the netlink part as the ioctl one > isn't affected: the reference to the net device is taken and the > operation is executed within an rtnl lock section and the net device > won't be found after unregister. > > [...] Here is the summary with links: - [net] ethtool: do not perform operations on net devices being unregistered https://git.kernel.org/netdev/net/c/dde91ccfa25f You are awesome, thank you!
Quoting Jakub Kicinski (2021-12-06 16:15:20) > On Mon, 6 Dec 2021 11:46:35 +0200 Julian Wiedmann wrote: > > > > Wondering if other places would also benefit from a netif_device_detach() > > in the unregistration sequence ... > > Sounds like a good idea but maybe as a follow up to net-next? > The likelihood of that breaking things is low, but non-zero. Might be good to have a look at this yes. I'm wondering, there are multiple mechanisms to avoid using a net device after unregistration, including netif_device_detach and unlinking it from the device chain (unlist_netdevice). Currently it's possible to have a device unlinked but not detached. Haven't looked at it, but there might be something to do here. (One the other way, detaching a device without unlinking it is valid for PM). Antoine
On 06.12.21 17:15, Jakub Kicinski wrote: > On Mon, 6 Dec 2021 11:46:35 +0200 Julian Wiedmann wrote: >> On 03.12.21 12:13, Antoine Tenart wrote: >>> There is a short period between a net device starts to be unregistered >>> and when it is actually gone. In that time frame ethtool operations >>> could still be performed, which might end up in unwanted or undefined >>> behaviours[1]. >>> >>> Do not allow ethtool operations after a net device starts its >>> unregistration. This patch targets the netlink part as the ioctl one >>> isn't affected: the reference to the net device is taken and the >>> operation is executed within an rtnl lock section and the net device >>> won't be found after unregister. >>> [...] >>> +++ b/net/ethtool/netlink.c >>> @@ -40,7 +40,8 @@ int ethnl_ops_begin(struct net_device *dev) >>> if (dev->dev.parent) >>> pm_runtime_get_sync(dev->dev.parent); >>> >>> - if (!netif_device_present(dev)) { >>> + if (!netif_device_present(dev) || >>> + dev->reg_state == NETREG_UNREGISTERING) { >>> ret = -ENODEV; >>> goto err; >>> } >>> >> >> Wondering if other places would also benefit from a netif_device_detach() >> in the unregistration sequence ... > > Sounds like a good idea but maybe as a follow up to net-next? > The likelihood of that breaking things is low, but non-zero. > Oh absolutely, only via net-next.
diff --git a/net/ethtool/netlink.c b/net/ethtool/netlink.c index 38b44c0291b1..96f4180aabd2 100644 --- a/net/ethtool/netlink.c +++ b/net/ethtool/netlink.c @@ -40,7 +40,8 @@ int ethnl_ops_begin(struct net_device *dev) if (dev->dev.parent) pm_runtime_get_sync(dev->dev.parent); - if (!netif_device_present(dev)) { + if (!netif_device_present(dev) || + dev->reg_state == NETREG_UNREGISTERING) { ret = -ENODEV; goto err; }
There is a short period between a net device starts to be unregistered and when it is actually gone. In that time frame ethtool operations could still be performed, which might end up in unwanted or undefined behaviours[1]. Do not allow ethtool operations after a net device starts its unregistration. This patch targets the netlink part as the ioctl one isn't affected: the reference to the net device is taken and the operation is executed within an rtnl lock section and the net device won't be found after unregister. [1] For example adding Tx queues after unregister ends up in NULL pointer exceptions and UaFs, such as: BUG: KASAN: use-after-free in kobject_get+0x14/0x90 Read of size 1 at addr ffff88801961248c by task ethtool/755 CPU: 0 PID: 755 Comm: ethtool Not tainted 5.15.0-rc6+ #778 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.14.0-4.fc34 04/014 Call Trace: dump_stack_lvl+0x57/0x72 print_address_description.constprop.0+0x1f/0x140 kasan_report.cold+0x7f/0x11b kobject_get+0x14/0x90 kobject_add_internal+0x3d1/0x450 kobject_init_and_add+0xba/0xf0 netdev_queue_update_kobjects+0xcf/0x200 netif_set_real_num_tx_queues+0xb4/0x310 veth_set_channels+0x1c3/0x550 ethnl_set_channels+0x524/0x610 Fixes: 041b1c5d4a53 ("ethtool: helper functions for netlink interface") Suggested-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Antoine Tenart <atenart@kernel.org> --- Following the discussions in those threads: - https://lore.kernel.org/all/20211129154520.295823-1-atenart@kernel.org/T/ - https://lore.kernel.org/all/20211122162007.303623-1-atenart@kernel.org/T/ net/ethtool/netlink.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)