On Fri, Sep 6, 2019 at 10:07 AM Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> Hmm. Maybe somebody can come up with a good legacy signaling solution
> (and "just use another pipe for error notification and OOB data" for
> the first one may _work_, but that sounds pretty hacky and just not
> very convenient).
... actually, maybe the trivial solution for at least some prototyping
cases is to make any user mode writers never drop messages. Don't use
a non-blocking fd for the write direction.
That's obviously *not* acceptable for a kernel writer, and it's not
acceptable for an actual system daemon writer (that you could block by
just not reading the notifications), but it's certainly acceptable for
the "let's prototype having kernel support for /proc/mounts
notifications using a local thread that just polls for it every few
seconds".
So at least for _some_ prototypes you can probably just ignore the
overflow issue. It won't get you full test coverage, but it will get
you a working legacy solution and a "look, if we have kernel support
for this, we can do better".
Linus
