To date, the worst that has happened are fork bombs. (For which I now have a solution.)
Is it safe to rely on LXC's security, provided that I keep the containers and hosts patches?
LXC security can doubtless be mis-configured; how much info can you share regarding your setup? From a few minutes research it sounds like you need user namespaces and seccomp for starters.
Redhat|OpenStack relies on SELinux much more than seccomp: https://news.ycombinator.com/item?id=14220503 http://rhelblog.redhat.com/2017/01/13/selinux-mitigates-cont...
Abusing Privileged and Unprivileged Linux Containers (2016) | https://www.nccgroup.trust/globalassets/our-research/us/whit...
You should get some free testing from your Show HN! https://news.ycombinator.com/item?id=14245447
There is only one user, tux, which is uid 1002 and tux does not exist in the host's /etc/passwd.
The users do not have root access in the containers.
Cgroups is used to impose limits on RAM, number of processes, visible CPUs, etc.
I made this site in evenings and weekends, and have put it up for the public to use. I want find out sooner, rather than later, if the whole idea is flawed.
https://dadario.com.br/preventing-docker-escaping-attacks/
> 1) Run the container as non-root user
> 2) [Use] a read-only file system
SELinux is probably your next step, which may be as simple as switching your host OS to RedHat/CentOS.
I'm not sure if you'll be able to use a VM on your current host, and using only one VM to host all your containers will not protect users from each other. (You could eventually use 1 VM for all free customers + VM per random batches of mid-tier paying customers + VM per premium customer.) VM escapes are much more rare than container escapes, and should be a solid line of defense separating your stuff from your users. The discussion I referenced recommended libvirt.