Re: [RFC PATCH 0/4] kgdb: Honour the kprobe blacklist when setting breakpoints

[Peter Zijlstra]

On Fri, Jun 05, 2020 at 04:29:53PM +0200, Peter Zijlstra wrote:
> On Fri, Jun 05, 2020 at 02:21:26PM +0100, Daniel Thompson wrote:
> > kgdb has traditionally adopted a no safety rails approach to breakpoint
> > placement. If the debugger is commanded to place a breakpoint at an
> > address then it will do so even if that breakpoint results in kgdb
> > becoming inoperable.
> > 
> > A stop-the-world debugger with memory peek/poke does intrinsically
> > provide its operator with the means to hose their system in all manner
> > of exciting ways (not least because stopping-the-world is already a DoS
> > attack ;-) ) but the current no safety rail approach is not easy to
> > defend, especially given kprobes provides us with plenty of machinery to
> > mark parts of the kernel where breakpointing is discouraged.
> > 
> > This patchset introduces some safety rails by using the existing
> > kprobes infrastructure. It does not cover all locations where
> > breakpoints can cause trouble but it will definitely block off several
> > avenues, including the architecture specific parts that are handled by
> > arch_within_kprobe_blacklist().
> > 
> > This patch is an RFC because:
> > 
> > 1. My workstation is still chugging through the compile testing.
> > 
> > 2. Patch 4 needs more runtime testing.
> > 
> > 3. The code to extract the kprobe blacklist code (patch 4 again) needs
> >    more review especially for its impact on arch specific code.
> > 
> > To be clear I do plan to do the detailed review of the kprobe blacklist
> > stuff but would like to check the direction of travel first since the
> > change is already surprisingly big and maybe there's a better way to
> > organise things.
> 
> Thanks for doing these patches, esp 1-3 look very good to me.
> 
> I've taken the liberty to bounce the entire set to Masami-San, who is
> the kprobes maintainer for comments as well.

OK, after having had a second look, one thing we can perhaps address
with the last patch, or perhaps on top of that, is extending the
kprobes_blacklist() with data regions.

Because these patches only exclude kgdb from setting breakpoints on
code; data breakpoints do not match what we do with
arch_build_bp_info().