On Thu, Mar 28, 2024 at 2:01 PM Tejun Heo tj@kernel.org wrote:
Hello,
On Thu, Mar 28, 2024 at 01:45:56PM -0700, Alexei Starovoitov wrote:
On Thu, Mar 28, 2024 at 1:02 PM Tejun Heo tj@kernel.org wrote:
There's also cgroup.kill which would be useful for similar use cases. We can add interface for both but idk. Let's say we have something like the following (pardon the bad naming):
bpf_cgroup_knob_write(struct cgroup *cgrp, char *filename, char *buf)
Would that work? I'm not necessarily in love with the idea or against adding separate helpers but the duplication still bothers me a bit.
I liked it. So filename will be one of cgroup_base_files[].name ? We probably don't want psi or cgroup1_base_files in there.
Would it matter?
Few weak reasons: . cgroup_psi_files have show/write/poll/release which doesn't map to this bpf_cgroup_knob_write/read ? . cgroup1_base_files probably needs to a separate kfunc bpf_cgroup1_...
If the user has root perm, they can do whatever with the files anyway, so I'm not sure why we'd restrict any specific knob. Maybe we wanna make sure @filename doesn't include '/'? Or is it that you don't want to go through the usual file name look up?
yeah. why do a file lookup? The names are there in the array. cgroup pointer gives that "relative path" and knob name is the last part of such "path". Easy to search in that array(s).
From the verifier pov 2nd arg can be "char *knob__str" and the verifier will make sure it's a constant NULL terminated string, so at runtime it will be easier to search cgroup_base_files array. And 'buf' can be: void *mem, int mem__sz with kfunc doing run-time validation that there a null there.
That all sound good.
Thanks.
-- tejun