lkml.org 
[lkml]   [2022]   [Feb]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [PATCH RFC bpf-next v2 5/5] selftests/bpf: test for pinning for cgroup_view link
On Mon, Feb 14, 2022 at 11:25 AM Alexei Starovoitov
<alexei.starovoitov@gmail.com> wrote:
>
> On Mon, Feb 14, 2022 at 10:29 AM Hao Luo <haoluo@google.com> wrote:
> > Hi Alexei,
> >
> > Actually, I found this almost worked, except that the tracepoints
> > cgroup_mkdir and cgroup_rmdir are not sleepable. They are inside a
> > spinlock's critical section with irq off. I guess one solution is to
> > offload the sleepable part of the bpf prog into a thread context. We
> > may create a dedicated kernel thread or use workqueue for this. Do you
> > have any advice?
>
> Are you referring to spin_lock in TRACE_CGROUP_PATH
> that protects global trace_cgroup_path[] buffer?

Yes, that's the spin_lock I am talking about.

> That is fixable.
> Do you actually need the string path returned by cgroup_path() in bpf prog?
> Maybe prog can call cgroup_path() by itself when necessary.
> Parsing strings isn't great anyway. The bpf prog probably needs the last
> part of the dir only. So cgrp->kn->name would do it?
> The TRACE_CGROUP_PATH wasn't designed to be turned on 24/7.
> That global spin_lock is not great for production use.
> No need to delegate sleepable bpf to thread context.
> Let's refactor that tracepoint a bit.

No, we don't need cgroup_path(). We are going to name the directories
by cgrp->kn->id. I can add a fast version for cgroup_xxx tracepoints,
which don't require the full path and can be turned on 24/7.

\
 
 \ /
  Last update: 2022-02-14 21:58    [W:0.140 / U:1.168 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site