On 07/25/2014 12:01 AM, Andy Lutomirski wrote:

On Jul 23, 2014 10:57 PM, "AKASHI Takahiro" [off-list ref] wrote:

quoted

On 07/24/2014 12:54 PM, Andy Lutomirski wrote:

quoted

On 07/22/2014 02:14 AM, AKASHI Takahiro wrote:

quoted

Arm64 holds a syscall number in w8(x8) register. Ptrace tracer may change
its value either to:
    * any valid syscall number to alter a system call, or
    * -1 to skip a system call

This patch implements this behavior by reloading that value into syscallno
in struct pt_regs after tracehook_report_syscall_entry() or
secure_computing(). In case of '-1', a return value of system call can also
be changed by the tracer setting the value to x0 register, and so
sys_ni_nosyscall() should not be called.

See also:
      42309ab4, ARM: 8087/1: ptrace: reload syscall number after
           secure_computing() check

Signed-off-by: AKASHI Takahiro <redacted>
---
   arch/arm64/kernel/entry.S  |    2 ++
   arch/arm64/kernel/ptrace.c |   13 +++++++++++++
   2 files changed, 15 insertions(+)

diff --git a/arch/arm64/kernel/entry.S b/arch/arm64/kernel/entry.S
index 5141e79..de8bdbc 100644
--- a/arch/arm64/kernel/entry.S
+++ b/arch/arm64/kernel/entry.S

@@ -628,6 +628,8 @@ ENDPROC(el0_svc)
   __sys_trace:
       mov    x0, sp
       bl    syscall_trace_enter
+    cmp    w0, #-1                // skip syscall?
+    b.eq    ret_to_user

Does this mean that skipped syscalls will cause exit tracing to be skipped?

Yes. (and I guess yes on arm, too)

quoted

If so, then you risk (at least) introducing

a nice user-triggerable OOPS if audit is enabled.

Can you please elaborate this?
Since I didn't find any definition of audit's behavior when syscall is
rewritten to -1, I thought it is reasonable to skip "exit tracing" of
"skipped" syscall.
(otherwise, "fake" seems to be more appropriate :)

The audit entry hook will oops if you call it twice in a row without
calling the exit hook in between.

Thank you, I could reproduce this problem which hits BUG(in_syscall) in
audit_syscall_entry(). Really bad, and I fixed it in my next version and
now a "skipped" system call is also traced by audit.

I ran libseccomp test and Kees' test under auditd running with a rule,
	auditctl -a exit,always -S all
and all the tests seemed to pass.

   I can also imagine ptracers getting

confused if ptrace entry and exit don't line up.

FYI, on arm64, we can distinguish syscall enter/exit with x12 register.

What happens if user code directly issues syscall ~0?  Does the return
value register get set?  Is the behavior different between traced and
untraced syscalls?

Interesting cases. Let me think about it.

Thanks,
-Takahiro AKASHI

 The current approach seems a bit scary.

--Andy