Re: [PATCH v3 1/2] kernel/sys: add PR_GET_TASK_SIZE option to prctl(2)

[Rafael Aquini]

On Fri, May 03, 2019 at 02:08:31PM -0700, Yury Norov wrote:
> On Fri, May 03, 2019 at 02:10:20PM -0400, Joel Savitz wrote:
> > When PR_GET_TASK_SIZE is passed to prctl, the kernel will attempt to
> > copy the value of TASK_SIZE to the userspace address in arg2.
> > 
> > It is important that we account for the case of the userspace task
> > running in 32-bit compat mode on a 64-bit kernel. As such, we must be
> > careful to copy the correct number of bytes to userspace to avoid stack
> > corruption.
> > 
> > Suggested-by: Yuri Norov <yury.norov@xxxxxxxxx>
> 
> I actually didn't suggest that. If you _really_ need TASK_SIZE to
> be exposed, I would suggest to expose it in kernel headers. TASK_SIZE
> is a compile-time information, and it may available for userspace at
> compile time as well.
> 
> > Suggested-by: Alexey Dobriyan <adobriyan@xxxxxxxxx>
> > Signed-off-by: Joel Savitz <jsavitz@xxxxxxxxxx>
> > ---
> >  include/uapi/linux/prctl.h |  3 +++
> >  kernel/sys.c               | 23 +++++++++++++++++++++++
> >  2 files changed, 26 insertions(+)
> > 
> > diff --git a/include/uapi/linux/prctl.h b/include/uapi/linux/prctl.h
> > index 094bb03b9cc2..2c261c461952 100644
> > --- a/include/uapi/linux/prctl.h
> > +++ b/include/uapi/linux/prctl.h
> > @@ -229,4 +229,7 @@ struct prctl_mm_map {
> >  # define PR_PAC_APDBKEY                        (1UL << 3)
> >  # define PR_PAC_APGAKEY                        (1UL << 4)
> > 
> > +/* Get the process virtual memory size (i.e. the highest usable VM address) */
> > +#define PR_GET_TASK_SIZE               55
> > +
> >  #endif /* _LINUX_PRCTL_H */
> > diff --git a/kernel/sys.c b/kernel/sys.c
> > index 12df0e5434b8..709584400070 100644
> > --- a/kernel/sys.c
> > +++ b/kernel/sys.c
> > @@ -2252,6 +2252,26 @@ static int propagate_has_child_subreaper(struct task_struct *p, void *data)
> >         return 1;
> >  }
> > 
> > +static int prctl_get_tasksize(void __user *uaddr)
> > +{
> > +	unsigned long current_task_size, current_word_size;
> > +
> > +	current_task_size = TASK_SIZE;
> > +	current_word_size = sizeof(unsigned long);
> > +
> > +#ifdef CONFIG_64BIT
> > +	/* On 64-bit architecture, we must check whether the current thread
> > +	 * is running in 32-bit compat mode. If it is, we can simply cut
> > +	 * the size in half. This avoids corruption of the userspace stack.
> > +	 */
> > +	if (test_thread_flag(TIF_ADDR32))
> 
> It breaks build for all architectures except x86 since TIF_ADDR32 is
> defined for x86 only.

Or we could get TIF_32BIT also defined for x86 (same value of
 TIF_ADDR32) and check for it instead. i.e.

...
#if defined(CONFIG_64BIT) && defined(TIF_32BIT)
	if (test_thread_flag(TIF_32BIT))
... 

which is also uglier and keeps adding unecessary complexity to a very
simple task. At this point, I think we just should give up on trying
this via prctl(2) and do it via /proc/<pid>/status instead. 


> 
> In comment to v2 I suggested you to stick to fixed-size data type to
> avoid exactly this problem.
> 
> NACK
> 
> Yury
> 
> > +		current_word_size >>= 1;
> > +#endif
> > +
> > +	return copy_to_user(uaddr, &current_task_size, current_word_size) ? -EFAULT : 0;
> > +}
> > +
> >  int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
> >  {
> >         return -EINVAL;
> > @@ -2486,6 +2506,9 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
> >                         return -EINVAL;
> >                 error = PAC_RESET_KEYS(me, arg2);
> >                 break;
> > +	case PR_GET_TASK_SIZE:
> > +		error = prctl_get_tasksize((void *)arg2);
> > +		break;
> >         default:
> >                 error = -EINVAL;
> >                 break;
> > --
> > 2.18.1