| Message ID | 751f88c0846df798a403643cefcaab53922ffe2f.1647255926.git.william.xuanziyang@huawei.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | net: ipvlan: fix potential UAF problem for phy_dev | expand |
On Mon, Mar 14, 2022 at 3:54 AM Ziyang Xuan <william.xuanziyang@huawei.com> wrote: > > Add the reference operation to phy_dev of ipvlan to avoid > the potential UAF problem under the following known scenario: > > Someone module puts the NETDEV_UNREGISTER event handler to a > work, and phy_dev is accessed in the work handler. But when > the work is excuted, phy_dev has been destroyed because upper > ipvlan did not get reference to phy_dev correctly. Can you name the module deferring NETDEV_UNREGISTER to a work queue ? This sounds like a bug to me. > > That likes as the scenario occurred by > commit 563bcbae3ba2 ("net: vlan: fix a UAF in vlan_dev_real_dev()"). Mentioning a commit that added a bug and many other commits trying to fix it is a bit unfortunate. Can you instead add a Fixes: tag ? Do you have a repro to trigger the bug ? > > Signed-off-by: Ziyang Xuan <william.xuanziyang@huawei.com> > --- > drivers/net/ipvlan/ipvlan_main.c | 13 +++++++++++++ > 1 file changed, 13 insertions(+) > > diff --git a/drivers/net/ipvlan/ipvlan_main.c b/drivers/net/ipvlan/ipvlan_main.c > index 696e245f6d00..dcdc01403f22 100644 > --- a/drivers/net/ipvlan/ipvlan_main.c > +++ b/drivers/net/ipvlan/ipvlan_main.c > @@ -158,6 +158,10 @@ static int ipvlan_init(struct net_device *dev) > } > port = ipvlan_port_get_rtnl(phy_dev); > port->count += 1; > + > + /* Get ipvlan's reference to phy_dev */ > + dev_hold(phy_dev); > + > return 0; > } > > @@ -665,6 +669,14 @@ void ipvlan_link_delete(struct net_device *dev, struct list_head *head) > } > EXPORT_SYMBOL_GPL(ipvlan_link_delete); > > +static void ipvlan_dev_free(struct net_device *dev) > +{ > + struct ipvl_dev *ipvlan = netdev_priv(dev); > + > + /* Get rid of the ipvlan's reference to phy_dev */ > + dev_put(ipvlan->phy_dev); > +} > + > void ipvlan_link_setup(struct net_device *dev) > { > ether_setup(dev); > @@ -674,6 +686,7 @@ void ipvlan_link_setup(struct net_device *dev) > dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE; > dev->netdev_ops = &ipvlan_netdev_ops; > dev->needs_free_netdev = true; > + dev->priv_destructor = ipvlan_dev_free; > dev->header_ops = &ipvlan_header_ops; > dev->ethtool_ops = &ipvlan_ethtool_ops; > } > -- > 2.25.1 >
> On Mon, Mar 14, 2022 at 3:54 AM Ziyang Xuan > <william.xuanziyang@huawei.com> wrote: >> >> Add the reference operation to phy_dev of ipvlan to avoid >> the potential UAF problem under the following known scenario: >> >> Someone module puts the NETDEV_UNREGISTER event handler to a >> work, and phy_dev is accessed in the work handler. But when >> the work is excuted, phy_dev has been destroyed because upper >> ipvlan did not get reference to phy_dev correctly. > > Can you name the module deferring NETDEV_UNREGISTER to a work queue ? The one I know is nb_netdevice netdevice notifier of roce_gid_mgmt. It will trigger vlan's real_dev UAF. It is a syzbot problem. See details: https://syzkaller.appspot.com/bug?extid=e4df4e1389e28972e955 > > This sounds like a bug to me. > >> >> That likes as the scenario occurred by >> commit 563bcbae3ba2 ("net: vlan: fix a UAF in vlan_dev_real_dev()"). > > Mentioning a commit that added a bug and many other commits trying to > fix it is a bit unfortunate. Yes, I am sorry for that. I will keep improving the quality of my code. The related problems have solved by the series patches of commit faab39f63c1f ("net: allow out-of-order netdev unregistration"). So I think it is the right time to fix other modules' potential problem. > > Can you instead add a Fixes: tag ? The potential problem exists since macvlan and ipvlan were added. > > Do you have a repro to trigger the bug ? For macvlan and ipvlan, there are not repros now. I think it is necessary to give a right logic to cope with the constant evolution of the kernel. > >> >> Signed-off-by: Ziyang Xuan <william.xuanziyang@huawei.com> >> --- >> drivers/net/ipvlan/ipvlan_main.c | 13 +++++++++++++ >> 1 file changed, 13 insertions(+) >> >> diff --git a/drivers/net/ipvlan/ipvlan_main.c b/drivers/net/ipvlan/ipvlan_main.c >> index 696e245f6d00..dcdc01403f22 100644 >> --- a/drivers/net/ipvlan/ipvlan_main.c >> +++ b/drivers/net/ipvlan/ipvlan_main.c >> @@ -158,6 +158,10 @@ static int ipvlan_init(struct net_device *dev) >> } >> port = ipvlan_port_get_rtnl(phy_dev); >> port->count += 1; >> + >> + /* Get ipvlan's reference to phy_dev */ >> + dev_hold(phy_dev); >> + >> return 0; >> } >> >> @@ -665,6 +669,14 @@ void ipvlan_link_delete(struct net_device *dev, struct list_head *head) >> } >> EXPORT_SYMBOL_GPL(ipvlan_link_delete); >> >> +static void ipvlan_dev_free(struct net_device *dev) >> +{ >> + struct ipvl_dev *ipvlan = netdev_priv(dev); >> + >> + /* Get rid of the ipvlan's reference to phy_dev */ >> + dev_put(ipvlan->phy_dev); >> +} >> + >> void ipvlan_link_setup(struct net_device *dev) >> { >> ether_setup(dev); >> @@ -674,6 +686,7 @@ void ipvlan_link_setup(struct net_device *dev) >> dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE; >> dev->netdev_ops = &ipvlan_netdev_ops; >> dev->needs_free_netdev = true; >> + dev->priv_destructor = ipvlan_dev_free; >> dev->header_ops = &ipvlan_header_ops; >> dev->ethtool_ops = &ipvlan_ethtool_ops; >> } >> -- >> 2.25.1 >> > . >
diff --git a/drivers/net/ipvlan/ipvlan_main.c b/drivers/net/ipvlan/ipvlan_main.c index 696e245f6d00..dcdc01403f22 100644 --- a/drivers/net/ipvlan/ipvlan_main.c +++ b/drivers/net/ipvlan/ipvlan_main.c @@ -158,6 +158,10 @@ static int ipvlan_init(struct net_device *dev) } port = ipvlan_port_get_rtnl(phy_dev); port->count += 1; + + /* Get ipvlan's reference to phy_dev */ + dev_hold(phy_dev); + return 0; } @@ -665,6 +669,14 @@ void ipvlan_link_delete(struct net_device *dev, struct list_head *head) } EXPORT_SYMBOL_GPL(ipvlan_link_delete); +static void ipvlan_dev_free(struct net_device *dev) +{ + struct ipvl_dev *ipvlan = netdev_priv(dev); + + /* Get rid of the ipvlan's reference to phy_dev */ + dev_put(ipvlan->phy_dev); +} + void ipvlan_link_setup(struct net_device *dev) { ether_setup(dev); @@ -674,6 +686,7 @@ void ipvlan_link_setup(struct net_device *dev) dev->priv_flags |= IFF_UNICAST_FLT | IFF_NO_QUEUE; dev->netdev_ops = &ipvlan_netdev_ops; dev->needs_free_netdev = true; + dev->priv_destructor = ipvlan_dev_free; dev->header_ops = &ipvlan_header_ops; dev->ethtool_ops = &ipvlan_ethtool_ops; }
Add the reference operation to phy_dev of ipvlan to avoid the potential UAF problem under the following known scenario: Someone module puts the NETDEV_UNREGISTER event handler to a work, and phy_dev is accessed in the work handler. But when the work is excuted, phy_dev has been destroyed because upper ipvlan did not get reference to phy_dev correctly. That likes as the scenario occurred by commit 563bcbae3ba2 ("net: vlan: fix a UAF in vlan_dev_real_dev()"). Signed-off-by: Ziyang Xuan <william.xuanziyang@huawei.com> --- drivers/net/ipvlan/ipvlan_main.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)