Re: [PATCH 18/27] bpf: Restrict kernel image access functions when the kernel is locked down

Hi Alexei,

Thanks for your review! 

On Thu, Oct 19, 2017 at 03:18:30PM -0700, Alexei Starovoitov wrote:

On Thu, Oct 19, 2017 at 03:52:49PM +0100, David Howells wrote:

quoted

From: Chun-Yi Lee <jlee@suse.com>

There are some bpf functions can be used to read kernel memory:
bpf_probe_read, bpf_probe_write_user and bpf_trace_printk.  These allow
private keys in kernel memory (e.g. the hibernation image signing key) to
be read by an eBPF program.  Prohibit those functions when the kernel is
locked down.

Signed-off-by: Chun-Yi Lee <jlee@suse.com>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: netdev@vger.kernel.org
---

 kernel/trace/bpf_trace.c |   11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
index dc498b605d5d..35e85a3fdb37 100644
--- a/kernel/trace/bpf_trace.c
+++ b/kernel/trace/bpf_trace.c

@@ -65,6 +65,11 @@ BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr)
 {
 	int ret;
 
+	if (kernel_is_locked_down("BPF")) {
+		memset(dst, 0, size);
+		return -EPERM;
+	}

That doesn't help the lockdown purpose.
If you don't trust the root the only way to prevent bpf read
memory is to disable the whole thing.

Not totally untrust root, I don't want that root reads
arbitrary memory address through bpf.

Is it not enough to lock down bpf_probe_read, bpf_probe_write_user
and bpf_trace_printk?

Have a single check in sys_bpf() to disallow everything if kernel_is_locked_down()
and don't add overhead to critical path like bpf_probe_read().

Yes, it give overhead to bpf_probe_read but it prevents arbitrary
memory read.

Another idea is signing bpf bytecode then verifying signture when
loading to kernel. 

Thanks a lot!
Joey Lee