Re: [PATCH v2 5/6] panic: Introduce warn_limit

[Kees Cook <keescook@chromium.org>]

On Mon, Nov 14, 2022 at 10:48:38AM +0100, Marco Elver wrote:
> On Wed, 9 Nov 2022 at 21:00, Kees Cook <keescook@chromium.org> wrote:
> >
> > Like oops_limit, add warn_limit for limiting the number of warnings when
> > panic_on_warn is not set.
> >
> > Cc: Jonathan Corbet <corbet@lwn.net>
> > Cc: Andrew Morton <akpm@linux-foundation.org>
> > Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
> > Cc: "Jason A. Donenfeld" <Jason@zx2c4.com>
> > Cc: Eric Biggers <ebiggers@google.com>
> > Cc: Huang Ying <ying.huang@intel.com>
> > Cc: Petr Mladek <pmladek@suse.com>
> > Cc: tangmeng <tangmeng@uniontech.com>
> > Cc: "Guilherme G. Piccoli" <gpiccoli@igalia.com>
> > Cc: Tiezhu Yang <yangtiezhu@loongson.cn>
> > Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
> > Cc: linux-doc@vger.kernel.org
> > Signed-off-by: Kees Cook <keescook@chromium.org>
> > ---
> >  Documentation/admin-guide/sysctl/kernel.rst |  9 +++++++++
> >  kernel/panic.c                              | 13 +++++++++++++
> >  2 files changed, 22 insertions(+)
> >
> > diff --git a/Documentation/admin-guide/sysctl/kernel.rst b/Documentation/admin-guide/sysctl/kernel.rst
> > index 09f3fb2f8585..c385d5319cdf 100644
> > --- a/Documentation/admin-guide/sysctl/kernel.rst
> > +++ b/Documentation/admin-guide/sysctl/kernel.rst
> > @@ -1508,6 +1508,15 @@ entry will default to 2 instead of 0.
> >  2 Unprivileged calls to ``bpf()`` are disabled
> >  = =============================================================
> >
> > +
> > +warn_limit
> > +==========
> > +
> > +Number of kernel warnings after which the kernel should panic when
> > +``panic_on_warn`` is not set. Setting this to 0 or 1 has the same effect
> > +as setting ``panic_on_warn=1``.
> > +
> > +
> >  watchdog
> >  ========
> >
> > diff --git a/kernel/panic.c b/kernel/panic.c
> > index 3afd234767bc..b235fa4a6fc8 100644
> > --- a/kernel/panic.c
> > +++ b/kernel/panic.c
> > @@ -58,6 +58,7 @@ bool crash_kexec_post_notifiers;
> >  int panic_on_warn __read_mostly;
> >  unsigned long panic_on_taint;
> >  bool panic_on_taint_nousertaint = false;
> > +static unsigned int warn_limit __read_mostly = 10000;
> >
> >  int panic_timeout = CONFIG_PANIC_TIMEOUT;
> >  EXPORT_SYMBOL_GPL(panic_timeout);
> > @@ -88,6 +89,13 @@ static struct ctl_table kern_panic_table[] = {
> >                 .extra2         = SYSCTL_ONE,
> >         },
> >  #endif
> > +       {
> > +               .procname       = "warn_limit",
> > +               .data           = &warn_limit,
> > +               .maxlen         = sizeof(warn_limit),
> > +               .mode           = 0644,
> > +               .proc_handler   = proc_douintvec,
> > +       },
> >         { }
> >  };
> >
> > @@ -203,8 +211,13 @@ static void panic_print_sys_info(bool console_flush)
> >
> >  void check_panic_on_warn(const char *reason)
> >  {
> > +       static atomic_t warn_count = ATOMIC_INIT(0);
> > +
> >         if (panic_on_warn)
> >                 panic("%s: panic_on_warn set ...\n", reason);
> > +
> > +       if (atomic_inc_return(&warn_count) >= READ_ONCE(warn_limit))
> > +               panic("Warned too often (warn_limit is %d)", warn_limit);
> 
> Shouldn't this also include the "reason", like above? (Presumably a
> warning had just been generated to console so the reason is easy
> enough to infer from the log, although in that case "reason" also
> seems redundant above.)

Yeah, that makes sense. I had been thinking that since it was an action
due to repeated prior actions, the current "reason" didn't matter here.
But thinking about it more, I see what you mean. :)

-- 
Kees Cook