Skip to contentSkip to conversation

A race condition is suspected.

report? Hyunwoo Kim 1 participant 2 messages Jul 4, 2023 – Jul 28, 2023

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Dear all,

From the code, it looks like a use-after-free due to a race condition could be caused by the lack of a proper lock between the ncsi_dev_work() worker and the vlan ioctl:

```
                  cpu0                                          cpu1

         ncsi_dev_work()
           ncsi_configure_channel()
             set_one_vid()
               rcu_read_lock()
                                                            sock_ioctl()
                                                              vlan_ioctl_handler()
                                                                unregister_vlan_dev()
                                                                  vlan_vid_del()
                                                                    __vlan_vid_del()
                                                                      vlan_kill_rx_filter_info()
                                                                        ncsi_vlan_rx_kill_vid()
                                                                          list_del_rcu(&vlan->list);
                                                                          kfree(vlan);
               vid = vlan->vid;    // use-after-free?
               rcu_read_unlock()
```

However, I do not have a device that uses the ncsi_dev_work() worker, so I have not been able to debug this.

Can a race condition like this actually happen, and if so, it seems like we need to change kfree(vlan); to kfree_rcu(vlan), or add an appropriate lock to the worker.

Regards,
Hyunwoo Kim

#1 lore raw
Quoted 28 lines · Hyunwoo Kim

Dear all,

From the code, it looks like a use-after-free due to a race condition could be caused by the lack of a proper lock between the ncsi_dev_work() worker and the vlan ioctl:

```
                  cpu0                                          cpu1

         ncsi_dev_work()
           ncsi_configure_channel()
             set_one_vid()
               rcu_read_lock()
                                                            sock_ioctl()
                                                              vlan_ioctl_handler()
                                                                unregister_vlan_dev()
                                                                  vlan_vid_del()
                                                                    __vlan_vid_del()
                                                                      vlan_kill_rx_filter_info()
                                                                        ncsi_vlan_rx_kill_vid()
                                                                          list_del_rcu(&vlan->list);
                                                                          kfree(vlan);
               vid = vlan->vid;    // use-after-free?
               rcu_read_unlock()
```

However, I do not have a device that uses the ncsi_dev_work() worker, so I have not been able to debug this.

Quoted 7 lines

Can a race condition like this actually happen, and if so, it seems like we need to change kfree(vlan); to kfree_rcu(vlan), or add an appropriate lock to the worker.

Regards,
Hyunwoo Kim

Hi,

Can I get a review of this report?
Should I submit a patch to replace kfree_rcu()?

Regards,
Hyunwoo Kim