Re: [PATCH bpf-next] selftests/bpf: fix potential premature unload in bpf_testmod

[Jiri Olsa <olsajiri@gmail.com>]

On Wed, Jan 10, 2024 at 09:14:51AM +0100, Artem Savkov wrote:
> On Tue, Jan 09, 2024 at 11:40:38AM -0800, Yonghong Song wrote:
> > 
> > On 1/9/24 8:43 AM, Artem Savkov wrote:
> > > It is possible for bpf_kfunc_call_test_release() to be called from
> > > bpf_map_free_deferred() when bpf_testmod is already unloaded and
> > > perf_test_stuct.cnt which it tries to decrease is no longer in memory.
> > > This patch tries to fix the issue by waiting for all references to be
> > > dropped in bpf_testmod_exit().
> > > 
> > > The issue can be triggered by running 'test_progs -t map_kptr' in 6.5,
> > > but is obscured in 6.6 by d119357d07435 ("rcu-tasks: Treat only
> > > synchronous grace periods urgently").
> > > 
> > > Fixes: 65eb006d85a2a ("bpf: Move kernel test kfuncs to bpf_testmod")
> > 
> > Please add your Signed-off-by tag.
> 
> Thanks for noticing. Will resend with signed-off-by and your ack.
> 
> > I think the root cause is that bpf_kfunc_call_test_acquire() kfunc
> > is defined in bpf_testmod and the kfunc returns some data in bpf_testmod.
> > But the release function bpf_kfunc_call_test_release() is in the kernel.
> > The release func tries to access some data in bpf_testmod which might
> > have been unloaded. The prog_test_ref_kfunc is defined in the kernel, so
> > no bpf_testmod btf reference is hold so bpf_testmod can be unloaded before
> > bpf_kfunc_call_test_release().
> > As you mentioned, we won't have this issue if bpf_kfunc_call_test_acquire()
> > is also in the kernel.
> > 
> > I think putting bpf_kfunc_call_test_acquire() in bpf_testmod and
> > bpf_kfunc_call_test_release() in kernel is not a good idea and confusing.
> > But since this is only for tests, I guess we can live with that. With that,
> 
> Correct. 65eb006d85a2a ("bpf: Move kernel test kfuncs to bpf_testmod")
> also mentions why bpf_kfunc_call_test_release() is not in the module and
> states that this is temporary. I'll add a comment in v2 so the wait can
> be removed once the functions are re-united.

I somehow recall it has to do with the fact you can't have trusted
pointer on module's object, so that's why those structs had to stay
in kernel.. but I might be wrong

jirka

>  
> > Acked-by: Yonghong Song <yonghong.song@linux.dev>
> > 
> > > ---
> > >   tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c | 4 ++++
> > >   1 file changed, 4 insertions(+)
> > > 
> > > diff --git a/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c b/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c
> > > index 91907b321f913..63f0dbd016703 100644
> > > --- a/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c
> > > +++ b/tools/testing/selftests/bpf/bpf_testmod/bpf_testmod.c
> > > @@ -2,6 +2,7 @@
> > >   /* Copyright (c) 2020 Facebook */
> > >   #include <linux/btf.h>
> > >   #include <linux/btf_ids.h>
> > > +#include <linux/delay.h>
> > >   #include <linux/error-injection.h>
> > >   #include <linux/init.h>
> > >   #include <linux/module.h>
> > > @@ -544,6 +545,9 @@ static int bpf_testmod_init(void)
> > >   static void bpf_testmod_exit(void)
> > >   {
> > > +	while (refcount_read(&prog_test_struct.cnt) > 1)
> > > +		msleep(20);
> > > +
> > >   	return sysfs_remove_bin_file(kernel_kobj, &bin_attr_bpf_testmod_file);
> > >   }
> > 
> 
> -- 
> Regards,
>   Artem
>