Re: [PATCH bpf-next v9 7/8] bpf: lsm: Add selftests for BPF_PROG_TYPE_LSM

From: Alexei Starovoitov <alexei.starovoitov@gmail.com>
To: KP Singh <kpsingh@chromium.org>
Cc: bpf <bpf@vger.kernel.org>, Andrii Nakryiko <andriin@fb.com>,
	Brendan Jackman <jackmanb@google.com>,
	Florent Revest <revest@google.com>,
	Thomas Garnier <thgarnie@google.com>,
	Daniel Borkmann <daniel@iogearbox.net>,
	Kees Cook <keescook@chromium.org>, Paul Turner <pjt@google.com>,
	Jann Horn <jannh@google.com>,
	Florent Revest <revest@chromium.org>,
	Brendan Jackman <jackmanb@chromium.org>
Subject: Re: [PATCH bpf-next v9 7/8] bpf: lsm: Add selftests for BPF_PROG_TYPE_LSM
Date: Wed, 1 Apr 2020 21:30:37 -0700	[thread overview]
Message-ID: <20200402043037.ltgyptxsf7jaaudu@ast-mbp> (raw)
In-Reply-To: <20200402040357.GA217889@google.com>

On Thu, Apr 02, 2020 at 06:03:57AM +0200, KP Singh wrote:
> On 01-Apr 17:09, Alexei Starovoitov wrote:
> > On Sat, Mar 28, 2020 at 5:44 PM KP Singh <kpsingh@chromium.org> wrote:
> > > +int BPF_PROG(test_int_hook, struct vm_area_struct *vma,
> > > +            unsigned long reqprot, unsigned long prot, int ret)
> > > +{
> > > +       if (ret != 0)
> > > +               return ret;
> > > +
> > > +       __u32 pid = bpf_get_current_pid_tgid() >> 32;
> > > +       int is_heap = 0;
> > > +
> > > +       is_heap = (vma->vm_start >= vma->vm_mm->start_brk &&
> > > +                  vma->vm_end <= vma->vm_mm->brk);
> > 
> > This test fails for me.
> 
> Trying this from bpf/master:
> 
>   b9258a2cece4 ("slcan: Don't transmit uninitialized stack data in padding")
> 
> also from bpf-next/master:
> 
>  1a323ea5356e ("x86: get rid of 'errret' argument to __get_user_xyz() macross")
> 
> and I am unable to reproduce the failure (the output when using bpf/master):
..
> 
> Also, I am wondering if this happens just in the BPF program or also
> in the kernel as the other variable I can think of is the compiled
> bpf program itself which might be reading a different value thinking
> it's vm->vma_start, possible something to do with BTF / CO RE due to a
> compiler bug:

I don't think it's anything to do with clang/btf or core.
I think that condition is simply incorrect.
I've added:

diff --git a/mm/mprotect.c b/mm/mprotect.c
index 311c0dadf71c..16ae0ada34ba 100644
--- a/mm/mprotect.c
+++ b/mm/mprotect.c
@@ -577,6 +577,7 @@ static int do_mprotect_pkey(unsigned long start, size_t len,
                        goto out;
                }

+               printk("start %llx %llx\n", vma->vm_start, vma->vm_mm->start_brk);
                error = security_file_mprotect(vma, reqprot, prot);

and see exactly the same values as bpf side (at least it was nice to see
that all CO-RE logic is working as expected :))

[   24.787442] start 523000 39b9000

I think it has something to do with the way test_progs is linked.
But the problem is in condition itself.
I suspect you copy-pasted it from selinux_file_mprotect() ?
I think it's incorrect there as well.
Did we just discover a way to side step selinux protection?
Try objdump -h test_progs|grep bss
the number I see in vma->vm_start is the beginning of .bss rounded to page boundary.
I wonder where your 55dc6e8df000 is coming from.
