by Martin Schwidefsky

[permalink] [raw]

Subject: Re: [PATCH 06/11] compat_ioctl: remove /dev/random commands

On Wed, 12 Sep 2018 16:02:40 +0200
Arnd Bergmann <[email protected]> wrote:

> On Wed, Sep 12, 2018 at 7:29 AM Martin Schwidefsky
> <[email protected]> wrote:
> > On Tue, 11 Sep 2018 22:26:54 +0200 Arnd Bergmann <[email protected]> wrote:
> > > On Sun, Sep 9, 2018 at 6:12 AM Al Viro <[email protected]> wrote:
>
> > > Out of those, there are only a few that may get used on s390,
> > > in particular at most infiniband/uverbs, nvme, nvdimm,
> > > btrfs, ceph, fuse, fanotify and userfaultfd.
> > > [Note: there are three s390 drivers in the list, which use
> > > a different method: they check in_compat_syscall() from
> > > a shared handler to decide whether to do compat_ptr().
> >
> > Using in_compat_syscall() seems to be a good solution, no?
>
> It works fine for you, but wouldn't work on architecture-independent
> code, since 32-bit architectures generally don't provide
> a compat_ptr() implementation. This could of course
> be changed easily, but after this change it, your drivers
> work just as well with a couple few lines, and more consistent
> with other drivers:
>
> --- a/drivers/s390/char/sclp_ctl.c
> +++ b/drivers/s390/char/sclp_ctl.c
> @@ -93,12 +93,8 @@ static int sclp_ctl_ioctl_sccb(void __user *user_area)
> static long sclp_ctl_ioctl(struct file *filp, unsigned int cmd,
> unsigned long arg)
> {
> - void __user *argp;
> + void __user *argp = (void __user *)arg;
>
> - if (is_compat_task())
> - argp = compat_ptr(arg);
> - else
> - argp = (void __user *) arg;
> switch (cmd) {
> case SCLP_CTL_SCCB:
> return sclp_ctl_ioctl_sccb(argp);
> @@ -114,7 +110,7 @@ static const struct file_operations sclp_ctl_fops = {
> .owner = THIS_MODULE,
> .open = nonseekable_open,
> .unlocked_ioctl = sclp_ctl_ioctl,
> - .compat_ioctl = sclp_ctl_ioctl,
> + .compat_ioctl = generic_compat_ioctl_ptrarg,
> .llseek = no_llseek,
> };
>
> This should probably be separate from the change to using compat_ptr()
> in all other drivers, and I could easily drop this change if you prefer,
> it is meant only as a cosmetic change.

So generic_compat_ioctl_ptrarg will to the compat_ptr thing on the
"unsigned int cmd" argument? Should work just fine.

> > > According to my memory from when I last worked on this,
> > > the compat_ptr() is mainly a safeguard for legacy binaries
> > > that got created with ancient C compilers (or compilers for
> > > something other than C) and might leave the high bit set
> > > in a pointer, but modern C compilers (gcc-3+) won't ever
> > > do that.
> >
> > And compat_ptr clears the upper 32-bit of the register. If
> > the register is loaded to e.g. "lr" or "l" there will be
> > junk in the 4 upper bytes.
>
> I don't think we hit that problem anywhere: in the ioctl
> argument we pass an 'unsigned long' that has already
> been zero-extended by the compat_sys_ioctl() wrapper,
> while any other usage would get extended by the compiler
> when casting from compat_uptr_t to a 64-bit type.
> This would be different if you had a function call with the
> wrong prototype, i.e. calling a function declared as taking
> an compat_uptr_t, but defining it as taking a void __user*.
> (I suppose that is undefined behavior).

That is true. For the ioctls we have a compat "unsigned int"
or "unsigned long" and the system call wrapper must have cleared
the upper half already. There are a few places where we copy
a data structure from user space, then read a 32-bit pointer
from the structure. These get the compat_ptr treatment as well.
All of those structure definitions should use compat_uptr_t
though, the compiler has to do the zero extension at the time
the 32-bit value is cast to a pointer.

> Unless I'm missing something, compat_ptr() should
> always just clear bit 31. What I'd like to confirm is whether
> you have encountered any code that actually passes
> a pointer with that bit set by a user application in the
> past 15 years. As Al said, it's probably best to just always
> apply the compat_ptr() conversion in each case that s390
> needs it even for drivers that don't run on s390, but I'd also
> like to understand how much it matters in practice.
> (A separate question would be how long we expect to need
> 32 bit compat mode on arch/s390 at all, but for the moment
> I assume this is not up for debate at all).

I don't know if that is worth the risk, yes it should work if
compat_ptr just removes bit 31 and leaves the other bits alone.
But if you have to clear one bit, you can as well remove all
the other bits as well.

--
blue skies,
Martin.

"Reality continues to ruin my life." - Calvin.

2018-09-13 08:15:58

by Arnd Bergmann

[permalink] [raw]

Subject: Re: [PATCH 06/11] compat_ioctl: remove /dev/random commands

On Thu, Sep 13, 2018 at 8:42 AM Martin Schwidefsky
<[email protected]> wrote:
>
> On Wed, 12 Sep 2018 16:02:40 +0200
> Arnd Bergmann <[email protected]> wrote:
>
> > On Wed, Sep 12, 2018 at 7:29 AM Martin Schwidefsky
> > <[email protected]> wrote:
> > > On Tue, 11 Sep 2018 22:26:54 +0200 Arnd Bergmann <[email protected]> wrote:

> >
> > This should probably be separate from the change to using compat_ptr()
> > in all other drivers, and I could easily drop this change if you prefer,
> > it is meant only as a cosmetic change.
>
> So generic_compat_ioctl_ptrarg will to the compat_ptr thing on the
> "unsigned int cmd" argument? Should work just fine.

It will do it on the "unsigned long arg" argument, I assume that's
what you meant. The "cmd" argument is correctly zero-extended
by the COMPAT_SYSCALL_DEFINE() wrapper on architectures
that need that (IIRC s390 is in that category).

> > I don't think we hit that problem anywhere: in the ioctl
> > argument we pass an 'unsigned long' that has already
> > been zero-extended by the compat_sys_ioctl() wrapper,
> > while any other usage would get extended by the compiler
> > when casting from compat_uptr_t to a 64-bit type.
> > This would be different if you had a function call with the
> > wrong prototype, i.e. calling a function declared as taking
> > an compat_uptr_t, but defining it as taking a void __user*.
> > (I suppose that is undefined behavior).
>
> That is true. For the ioctls we have a compat "unsigned int"
> or "unsigned long" and the system call wrapper must have cleared
> the upper half already. There are a few places where we copy
> a data structure from user space, then read a 32-bit pointer
> from the structure. These get the compat_ptr treatment as well.
> All of those structure definitions should use compat_uptr_t
> though, the compiler has to do the zero extension at the time
> the 32-bit value is cast to a pointer.

There is actually one more case: A number of the newer
interfaces that have ioctl structures with indirect pointers
encoded as __u64, so the layout becomes
and we don't normally need a conversion handler.

An example of this would be the sys_rseq() system call
that passes a relatively complex structure in place
of a pointer:

struct rseq {
...
union {
__u64 ptr64;
#ifdef __LP64__
__u64 ptr;
#else
struct {
#if (defined(__BYTE_ORDER) && (__BYTE_ORDER == __BIG_ENDIAN)) ||
defined(__BIG_ENDIAN)
__u32 padding; /* Initialized to zero. */
__u32 ptr32;
#else /* LITTLE */
__u32 ptr32;
__u32 padding; /* Initialized to zero. */
#endif /* ENDIAN */
} ptr;
#endif
} rseq_cs;
__u32 flags;
};

We require user space to initialize the __padding field to
zero and then use the ptr64 field in the kernel as a pointer:

u64 ptr;
u32 __user *usig;
copy_from_user(&ptr, &t->rseq->rseq_cs.ptr64, sizeof(ptr));
urseq_cs = (struct rseq_cs __user *)(unsigned long)ptr;

but we don't ever clear bit 31 here. A similar pattern is used
in many device drivers (I could not find any that would apply to
s390 though). In theory, 32 bit user space might pass a pointer
with the high bit set in the ptr32 field, and that gets misinterpreted
by the kernel (resulting in -EFAULT).

It would be interesting to know whether there could be user space
that gets compiled from portable source code with a normal
C compiler but produces that high bit set, as opposed to someone
intentially settting the bit just to trigger the bug.

Arnd