Re: [PATCHv2 bpf 3/3] bpf: Force cookies array to follow symbols sorting

[Andrii Nakryiko]

On Wed, Jun 8, 2022 at 9:08 AM Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
>
> On Wed, 8 Jun 2022 08:59:50 -0700
> Andrii Nakryiko <andrii.nakryiko@xxxxxxxxx> wrote:
>
> > Would it be possible to preprocess ftrace_pages to remove such invalid
> > records (so that by the time we have to report
> > available_filter_functions there are no invalid records)? Or that data
> > is read-only when kernel is running?
>
> It's possible, but will be time consuming (slow down boot up) and racy. In
> other words, I didn't feel it was worth it.
>
> We can add it. How much of an issue is it to have these place holders for
> you? Currently, I only see it causes issues with tests. Is it really an
> issue for use cases?

I have the tool (retsnoop) that uses available_filter_functions, I'll
have to update it to ignore such entries. It's a small inconvenience,
once you know about this change, but multiply that for multiple users
that use available_filter_functions for some sort of generic tooling
doing kprobes/tracing, and it adds up. So while it's not horrible,
ideally user-visible data shouldn't have non-usable placeholders.

How much slowdown would you expect on start up? Not clear what would
be racy about this start up preprocessing, but I believe you.

So in summary, it's not the end of the world, but as a user I'd prefer
not to know about this quirk, of course :)

>
> -- Steve