Re: [PATCH v2] uaccess: add noop untagged_addr definition

[Jason Gunthorpe]

On Tue, Jun 04, 2019 at 01:38:00PM +0100, Catalin Marinas wrote:
> On Tue, Jun 04, 2019 at 09:28:41AM -0300, Jason Gunthorpe wrote:
> > On Tue, Jun 04, 2019 at 02:04:47PM +0200, Andrey Konovalov wrote:
> > > Architectures that support memory tagging have a need to perform untagging
> > > (stripping the tag) in various parts of the kernel. This patch adds an
> > > untagged_addr() macro, which is defined as noop for architectures that do
> > > not support memory tagging. The oncoming patch series will define it at
> > > least for sparc64 and arm64.
> > > 
> > > Acked-by: Catalin Marinas <catalin.marinas@xxxxxxx>
> > > Reviewed-by: Khalid Aziz <khalid.aziz@xxxxxxxxxx>
> > > Signed-off-by: Andrey Konovalov <andreyknvl@xxxxxxxxxx>
> > >  include/linux/mm.h | 11 +++++++++++
> > >  1 file changed, 11 insertions(+)
> > > 
> > > diff --git a/include/linux/mm.h b/include/linux/mm.h
> > > index 0e8834ac32b7..dd0b5f4e1e45 100644
> > > +++ b/include/linux/mm.h
> > > @@ -99,6 +99,17 @@ extern int mmap_rnd_compat_bits __read_mostly;
> > >  #include <asm/pgtable.h>
> > >  #include <asm/processor.h>
> > >  
> > > +/*
> > > + * Architectures that support memory tagging (assigning tags to memory regions,
> > > + * embedding these tags into addresses that point to these memory regions, and
> > > + * checking that the memory and the pointer tags match on memory accesses)
> > > + * redefine this macro to strip tags from pointers.
> > > + * It's defined as noop for arcitectures that don't support memory tagging.
> > > + */
> > > +#ifndef untagged_addr
> > > +#define untagged_addr(addr) (addr)
> > 
> > Can you please make this a static inline instead of this macro? Then
> > we can actually know what the input/output types are supposed to be.
> > 
> > Is it
> > 
> > static inline unsigned long untagged_addr(void __user *ptr) {return ptr;}
> > 
> > ?
> > 
> > Which would sort of make sense to me.
> 
> This macro is used mostly on unsigned long since for __user ptr we can
> deference them in the kernel even if tagged. 

What does that mean? Do all kernel apis that accept 'void __user *'
already untag due to other patches?

> So if we are to use types here, I'd rather have:
> 
> static inline unsigned long untagged_addr(unsigned long addr);
> 
> In addition I'd like to avoid the explicit casting to (unsigned long)
> and use some userptr_to_ulong() or something. 

Personally I think it is a very bad habit we have in the kernel to
store a 'void __user *' as a u64 or an unsigned long all over the
place.

AFAIK a u64 passed in from userpace is supposed to be converted to the
'void __user *' via u64_to_user_ptr() before it can be used. (IIRC
Some arches require this..)

So, if I have a ioctl that takes a user pointer as a u64, and I want
to pass it to find_vma, then I do need to write:

    find_vma(untagged_addr(u64_to_user_ptr(ioctl_u64)))

Right?

So, IMHO, not accepting a 'void __user *' is just encouraging drivers
to skip the needed u64_to_user_ptr() step.

At the very worst we should have at least a 2nd function, but, IMHO,
it would be better to do a bit more work on adding missing
u64_to_user_ptr() calls to get the 'void __user *', and maybe a bit
more work on swapping unsigned long for 'void __user *' in various
places.

Jason