Repository navigation
RFE: BPF programs on cgroups #10227
Description
Activity
- changed the title
[-]RFI: BPF programs on cgroups[/-][+]RFE: BPF programs on cgroups[/+]on Oct 1, 2018 Here is a previous discussion:
#6764 (comment)
where I suggest to use eBPF programs on the BPF filesystem as an alternative to/proc/$PID/fd/<appropriate FD>. We also need to be able to speify different eBPF programs for egress and ingress.- added a commit that references this issue
on Jun 25, 2019 - added a commit that references this issue
on Sep 23, 2019
At ASG2018 hack we discussed a path to allow users to ask Systemd to install BPF programs into cgroups. This issue is to document that design as an RFE.
The motivation is to allow users to associate a BPF program with a unit file so that systemd can install them after the cgroup is setup.
The design we came up with was:
The user would be responsible for creating any maps, loading the program and associating any maps with the program. At this point the user simply needs to pass the FD for the BPF program to systemd. The suggested name for this was BPFProgramPath. In this case, the suggested way of doing this was via
/proc/$PID/fds/<appropriate FD>.An option to
systemd-nspawncould then be made to expose this as well.