On Thu, Mar 11, 2021 at 12:08:11PM -0800, Andi Kleen wrote:
When a brute force attack is detected through the fork or execve system call, all the tasks involved in the attack will be killed with the exception of the init task (task with pid equal to zero). Now, and only if the init task is involved in the attack, block the fork system call from the init process during a user defined time (using a sysctl attribute). This way the brute force attack is mitigated and the system does not panic.
That means nobody can log in and fix the system during that time.
Would be better to have that policy in init. Perhaps add some way that someone doing wait*() can know the exit was due this mitigation (and not something way) Then they could disable respawning of that daemon.
Great. So, if we use wait*() to inform userspace that the exit of a process was due to a brute force attack then, the supervisors (not only init) can adopt the necessary policy in each case. This also allow us to deal with the respawned daemons.
As a summary of this useful discussion:
- When a brute force attack is detected through the fork or execve system call all the offending tasks involved in the attack will be killed. Due to the mitigation normally not reach init, do nothing special in this case -> the system will panic when we kill init.
- Use wait*() to inform userspace that every process killed by the mitigation has exited due to a brute force attack mitigation. So, each supervisor can adopt their own policy regarding respawned daemons.
I will work in that direction for the next version.
Thanks a lot for your time, proposals, guidance and solutions. John Wood