On Fri, Dec 16, 2022 at 9:43 AM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Fri, 16 Dec 2022 09:15:40 -0800 Jeff Xu <jeffxu@xxxxxxxxxx> wrote:
>
> > On Fri, Dec 16, 2022 at 7:47 AM Peter Xu <peterx@xxxxxxxxxx> wrote:
> > >
> > > Hi, Jeff,
> > >
> > > On Thu, Dec 08, 2022 at 02:55:45PM -0800, Jeff Xu wrote:
> > > > > > + if (!(flags & (MFD_EXEC | MFD_NOEXEC_SEAL))) {
> > >
> > > [...]
> > >
> > > > > > + pr_warn_ratelimited(
> > > > > > + "memfd_create() without MFD_EXEC nor MFD_NOEXEC_SEAL, pid=%d '%s'\n",
> > > > > > + task_pid_nr(current), get_task_comm(comm, current));
> > >
> > > This will be frequently dumped right now with mm-unstable. Is that what it
> > > wanted to achieve?
> > >
> > > [ 10.822575] memfd_create() without MFD_EXEC nor MFD_NOEXEC_SEAL, pid=491 'systemd'
> > > [ 10.824743] memfd_create() without MFD_EXEC nor MFD_NOEXEC_SEAL, pid=495 '(sd-executor)'
> > > ...
> > >
> > > If there's already a sane default value (and also knobs for the user to
> > > change the default) not sure whether it's saner to just keep it silent as
> > > before?
> > >
> > Thanks for your comments.
> >
> > The intention is it is a reminder to adjust API calls to explicitly
> > setting this bit.
>
> Do we need to warn more than once per boot? If not, use pr_warn_once()?
>
Once per boot seems too little, it would be nice if we can list all processes.
I agree ratelimited might be too much.
There is a feature gap here for logging.
Kees, what do you think ?
> > The sysctl vm.memfd_noexec = 0 1 is for transaction to the final
> > state, and 2 depends on API call setting this bit.
> >
> > The log is ratelimited, and there is a rate limit setting:
> > /proc/sys/kernel/printk_ratelimit
> > /proc/sys/kernel/printk_ratelimit_burst
> >
>
