Re: [PATCH] powerpc/ptrace: Mitigate potential Spectre v1

Hi Breno,

On 1/29/19 10:38 AM, Breno Leitao wrote:

Hi Gustavo,

On 1/24/19 3:25 PM, Gustavo A. R. Silva wrote:

quoted

On 1/24/19 8:01 AM, Breno Leitao wrote:

quoted

'regno' is directly controlled by user space, hence leading to a potential
exploitation of the Spectre variant 1 vulnerability.

On PTRACE_SETREGS and PTRACE_GETREGS requests, user space passes the
register number that would be read or written. This register number is
called 'regno' which is part of the 'addr' syscall parameter.

This 'regno' value is checked against the maximum pt_regs structure size,
and then used to dereference it, which matches the initial part of a
Spectre v1 (and Spectre v1.1) attack. The dereferenced value, then,
is returned to userspace in the GETREGS case.

Was this reported by any tool?

It was not. I was writing an userspace tool, which required me to read the
Ptrace subsystem carefully, then, I just found this case.

Great. Good catch.

quoted

If so, it might be worth mentioning it.

quoted

This patch sanitizes 'regno' before using it to dereference pt_reg.

Notice that given that speculation windows are large, the policy is
to kill the speculation on the first load and not worry if it can be
completed with a dependent load/store [1].

[1] https://marc.info/?l=linux-kernel&m=152449131114778&w=2

Signed-off-by: Breno Leitao <leitao@debian.org>
---
 arch/powerpc/kernel/ptrace.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/arch/powerpc/kernel/ptrace.c b/arch/powerpc/kernel/ptrace.c
index cdd5d1d3ae41..3eac38a29863 100644
--- a/arch/powerpc/kernel/ptrace.c
+++ b/arch/powerpc/kernel/ptrace.c

@@ -33,6 +33,7 @@
 #include <linux/hw_breakpoint.h>
 #include <linux/perf_event.h>
 #include <linux/context_tracking.h>
+#include <linux/nospec.h>
 
 #include <linux/uaccess.h>
 #include <linux/pkeys.h>

@@ -298,6 +299,9 @@ int ptrace_get_reg(struct task_struct *task, int regno, unsigned long *data)
 #endif
 
 	if (regno < (sizeof(struct user_pt_regs) / sizeof(unsigned long))) {

I would use a variable to store sizeof(struct user_pt_regs) / sizeof(unsigned long).

Right.

quoted

quoted

+		regno = array_index_nospec(regno,
+				(sizeof(struct user_pt_regs) /
+				 sizeof(unsigned long)));

See the rest of my comments below.

quoted

 		*data = ((unsigned long *)task->thread.regs)[regno];
 		return 0;
 	}

@@ -321,6 +325,7 @@ int ptrace_put_reg(struct task_struct *task, int regno, unsigned long data)
 		return set_user_dscr(task, data);
 
 	if (regno <= PT_MAX_PUT_REG) {
+		regno = array_index_nospec(regno, PT_MAX_PUT_REG);

This is wrong.  array_index_nospec() will return PT_MAX_PUT_REG - 1 in case regno is equal to
PT_MAX_PUT_REG, and this is not what you want.

Right, this is really wrong. It would be correct if the comparison was
regno < PT_MAX_PUT_REG. Since it is PT_MAX_PUT_REGS is a valid array entry,
then I need to care about this case also. Doing something as:

	regno = array_index_nospec(regno, PT_MAX_PUT_REG + 1);

This is the right way.

Other than that, I think that for the regno = PT_MAX_PUT_REG base,
array_index_nospec() will redefine regno to 0, not PT_MAX_PUT_REG - 1.

quoted

Similar reasoning applies to the case above.

I understand that the case above does not seem to have the same problem,
since it does not address over the array as done in the case above. Does it
make sense?

Yeah. I was wrong about that.  If regno is out of bounds, array_index_nospec()
will return 0.

Anyway, this is how I am thinking the v2 of the patch:

V2 looks good.

Thanks
--
Gustavo