[PATCH] objtool,x86: Verify poke_int3_handler() is self contained

[PATCH] objtool,x86: Verify poke_int3_handler() is self contained

[peterz]

Abuse the SMAP rules to ensure poke_int3_handler() doesn't call out to
anything.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 tools/objtool/check.c |    8 ++++++++
 1 file changed, 8 insertions(+)

--- a/tools/objtool/check.c
+++ b/tools/objtool/check.c
@@ -551,6 +551,14 @@ static const char *uaccess_safe_builtin[
 	"__memcpy_mcsafe",
 	"mcsafe_handle_tail",
 	"ftrace_likely_update", /* CONFIG_TRACE_BRANCH_PROFILING */
+	/*
+	 * Abuse alert!
+	 *
+	 * poke_int3_handler() is not in fact related to uaccess at all, we
+	 * abuse the uaccess rules to ensure poke_int3_handler() is self
+	 * contained and doesn't CALL out to other code.
+	 */
+	"poke_int3_handler",
 	NULL
 };
 

Re: [PATCH] objtool,x86: Verify poke_int3_handler() is self contained

[Ingo Molnar]

* peterz@infradead.org <peterz@infradead.org> wrote:

> 
> Abuse the SMAP rules to ensure poke_int3_handler() doesn't call out to
> anything.
> 
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> ---
>  tools/objtool/check.c |    8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> --- a/tools/objtool/check.c
> +++ b/tools/objtool/check.c
> @@ -551,6 +551,14 @@ static const char *uaccess_safe_builtin[
>  	"__memcpy_mcsafe",
>  	"mcsafe_handle_tail",
>  	"ftrace_likely_update", /* CONFIG_TRACE_BRANCH_PROFILING */
> +	/*
> +	 * Abuse alert!
> +	 *
> +	 * poke_int3_handler() is not in fact related to uaccess at all, we
> +	 * abuse the uaccess rules to ensure poke_int3_handler() is self
> +	 * contained and doesn't CALL out to other code.
> +	 */
> +	"poke_int3_handler",

So ->uaccess_safe makes sure that we don't CALL into non-uaccess-safe 
functions, but it still allows CALLs into *other* uaccess-safe 
functions, right?

So unless I missed something in the logic, the comment should say 
something like "doesn't CALL out to other non-uaccess safe code" or 
so? Which is, arguably, like 99% of all functions - but still, a whole 
bunch are allowed, such as low level instrumentation and other utility 
functions.

Thanks,

	Ingo

Re: [PATCH] objtool,x86: Verify poke_int3_handler() is self contained

[peterz]

On Mon, Jul 27, 2020 at 01:21:44PM +0200, Ingo Molnar wrote:
> 
> * peterz@infradead.org <peterz@infradead.org> wrote:
> 
> > 
> > Abuse the SMAP rules to ensure poke_int3_handler() doesn't call out to
> > anything.
> > 
> > Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> > ---
> >  tools/objtool/check.c |    8 ++++++++
> >  1 file changed, 8 insertions(+)
> > 
> > --- a/tools/objtool/check.c
> > +++ b/tools/objtool/check.c
> > @@ -551,6 +551,14 @@ static const char *uaccess_safe_builtin[
> >  	"__memcpy_mcsafe",
> >  	"mcsafe_handle_tail",
> >  	"ftrace_likely_update", /* CONFIG_TRACE_BRANCH_PROFILING */
> > +	/*
> > +	 * Abuse alert!
> > +	 *
> > +	 * poke_int3_handler() is not in fact related to uaccess at all, we
> > +	 * abuse the uaccess rules to ensure poke_int3_handler() is self
> > +	 * contained and doesn't CALL out to other code.
> > +	 */
> > +	"poke_int3_handler",
> 
> So ->uaccess_safe makes sure that we don't CALL into non-uaccess-safe 
> functions, but it still allows CALLs into *other* uaccess-safe 
> functions, right?
> 
> So unless I missed something in the logic, the comment should say 
> something like "doesn't CALL out to other non-uaccess safe code" or 
> so? Which is, arguably, like 99% of all functions - but still, a whole 
> bunch are allowed, such as low level instrumentation and other utility 
> functions.

Right, so poke_int3_handler() is also noinstr and by that isn't allowed
to call out into !noinstr code. The intersection should be small.

But yeah, perhaps this is a bad idea and I should add another annotation
for this,.. dunno.

Re: [PATCH] objtool,x86: Verify poke_int3_handler() is self contained

[Ingo Molnar]

* peterz@infradead.org <peterz@infradead.org> wrote:

> On Mon, Jul 27, 2020 at 01:21:44PM +0200, Ingo Molnar wrote:
> > 
> > * peterz@infradead.org <peterz@infradead.org> wrote:
> > 
> > > 
> > > Abuse the SMAP rules to ensure poke_int3_handler() doesn't call out to
> > > anything.
> > > 
> > > Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> > > ---
> > >  tools/objtool/check.c |    8 ++++++++
> > >  1 file changed, 8 insertions(+)
> > > 
> > > --- a/tools/objtool/check.c
> > > +++ b/tools/objtool/check.c
> > > @@ -551,6 +551,14 @@ static const char *uaccess_safe_builtin[
> > >  	"__memcpy_mcsafe",
> > >  	"mcsafe_handle_tail",
> > >  	"ftrace_likely_update", /* CONFIG_TRACE_BRANCH_PROFILING */
> > > +	/*
> > > +	 * Abuse alert!
> > > +	 *
> > > +	 * poke_int3_handler() is not in fact related to uaccess at all, we
> > > +	 * abuse the uaccess rules to ensure poke_int3_handler() is self
> > > +	 * contained and doesn't CALL out to other code.
> > > +	 */
> > > +	"poke_int3_handler",
> > 
> > So ->uaccess_safe makes sure that we don't CALL into non-uaccess-safe 
> > functions, but it still allows CALLs into *other* uaccess-safe 
> > functions, right?
> > 
> > So unless I missed something in the logic, the comment should say 
> > something like "doesn't CALL out to other non-uaccess safe code" or 
> > so? Which is, arguably, like 99% of all functions - but still, a whole 
> > bunch are allowed, such as low level instrumentation and other utility 
> > functions.
> 
> Right, so poke_int3_handler() is also noinstr and by that isn't allowed
> to call out into !noinstr code. The intersection should be small.
> 
> But yeah, perhaps this is a bad idea and I should add another annotation
> for this,.. dunno.

'nocall' ? :-)

	Ingo

Re: [PATCH] objtool,x86: Verify poke_int3_handler() is self contained

[Thomas Gleixner]

peterz@infradead.org writes:
> Abuse the SMAP rules to ensure poke_int3_handler() doesn't call out to
> anything.

Yuck. Isn't that what noinstr is for or am I missing something?

Thanks,

        tglx

Re: [PATCH] objtool,x86: Verify poke_int3_handler() is self contained

[peterz]

On Mon, Jul 27, 2020 at 10:56:03PM +0200, Thomas Gleixner wrote:
> peterz@infradead.org writes:
> > Abuse the SMAP rules to ensure poke_int3_handler() doesn't call out to
> > anything.
> 
> Yuck. Isn't that what noinstr is for or am I missing something?

Well, we don't want poke_int3_handler() to call out to noinstr either.

So noinstr only allows calling noinstr, The above hack only allows it
calling uaccess-safe functions, and the group of functions that is both
noinstr and uaccess-safe is 'small'.

But like I said yesterday, perhaps this wants it own annotation. Then we
can also verify the lack of any dynamic code.

So yes, yuck and a bad idea, ignore this ;-)