Re: [PATCH 4/9] ARM: syscall: always store thread_info->syscall

[Arnd Bergmann]

On Fri, Oct 30, 2020 at 5:53 PM Russell King - ARM Linux admin
<linux@xxxxxxxxxxxxxxx> wrote:
>
> On Fri, Oct 30, 2020 at 04:49:14PM +0100, Arnd Bergmann wrote:
> > From: Arnd Bergmann <arnd@xxxxxxxx>
> >
> > The system call number is used in a a couple of places, in particular
> > ptrace, seccomp and /proc/<pid>/syscall.
> >
> > The last one apparently never worked reliably on ARM for tasks
> > that are not currently getting traced.
> >
> > Storing the syscall number in the normal entry path makes it work,
> > as well as allowing us to see if the current system call is for
> > OABI compat mode, which is the next thing I want to hook into.
>
> I'm not sure this patch is correct.

I'm not following where you still see a mismatch, I was hoping I
had fixed them all after your previous review :(

The thread_info->syscall entry should now consistently contain
__NR_SYSCALL_BASE on an EABI kernel, and all users of
that should consistently mask it out.

> Tracing the existing code for OABI:
>
> asmlinkage int syscall_trace_enter(struct pt_regs *regs, int scno)
> {
>         current_thread_info()->syscall = scno;

This no longer stores to current_thread_info()->syscall but instead
reads the number from syscall_get_nr().

>         /* Legacy ABI only. */
> USER(   ldr     scno, [saved_pc, #-4]   )       @ get SWI instruction
>         bic     scno, scno, #0xff000000         @ mask off SWI op-code
>         eor     scno, scno, #__NR_SYSCALL_BASE  @ check OS number
>         tst     r10, #_TIF_SYSCALL_WORK         @ are we tracing syscalls?
>         bne     __sys_trace
>
> __sys_trace:
>         mov     r1, scno
>         add     r0, sp, #S_OFF
>         bl      syscall_trace_enter
>
> So, thread_info->syscall does not include __NR_SYSCALL_BASE. The
> reason for this is the code that makes use of that via syscall_get_nr().
> kernel/trace/trace_syscalls.c:

On both CONFIG_OABI_COMPAT and on !CONFIG_AEABI kernels,
I store the value before masking out __NR_SYSCALL_BASE
after my change. For EABI-only kernels there is no need for
the mask.

>         syscall_nr = trace_get_syscall_nr(current, regs);
>         if (syscall_nr < 0 || syscall_nr >= NR_syscalls)
>                 return;
>
> and NR_syscalls is the number of syscalls, which doesn't include the
> __NR_SYSCALL_BASE offset.
>
> So, I think this patch actually breaks OABI.

The value returned from trace_get_syscall_nr() is always in
the 0...NR_syscalls range without the __NR_SYSCALL_BASE
for a valid syscall. because of the added

 static inline int syscall_get_nr(struct task_struct *task,
                                 struct pt_regs *regs)
 {
-       return task_thread_info(task)->syscall;
+       return task_thread_info(task)->syscall & ~__NR_OABI_SYSCALL_BASE;
 }

(plus the corresponding logic for OABI_COMPAT.

Which of the above do you think I got wrong?

      Arnd