Re: [PATCH v1 1/2] mm: Implement memory-deny-write-execute as a prctl

[Joey Gouly]

Hi,

On Fri, Oct 28, 2022 at 11:51:00AM -0700, Kees Cook wrote:
> On Wed, Oct 26, 2022 at 04:04:56PM +0100, Joey Gouly wrote:
> > The aim of such policy is to prevent a user task from creating an
> > executable mapping that is also writeable.
> > 
> > An example of mmap() returning -EACCESS if the policy is enabled:
> > 
> > 	mmap(0, size, PROT_READ | PROT_WRITE | PROT_EXEC, flags, 0, 0);
> > 
> > Similarly, mprotect() would return -EACCESS below:
> > 
> > 	addr = mmap(0, size, PROT_READ | PROT_EXEC, flags, 0, 0);
> > 	mprotect(addr, size, PROT_READ | PROT_WRITE | PROT_EXEC);
> > 
> > The BPF filter that systemd MDWE uses is stateless, and disallows
> > mprotect() with PROT_EXEC completely. This new prctl allows PROT_EXEC to
> > be enabled if it was already PROT_EXEC, which allows the following case:
> > 
> > 	addr = mmap(0, size, PROT_READ | PROT_EXEC, flags, 0, 0);
> > 	mprotect(addr, size, PROT_READ | PROT_EXEC | PROT_BTI);
> > 
> > where PROT_BTI enables branch tracking identification on arm64.
> > 
> > Signed-off-by: Joey Gouly <joey.gouly@xxxxxxx>
> > Co-developed-by: Catalin Marinas <catalin.marinas@xxxxxxx>
> > Signed-off-by: Catalin Marinas <catalin.marinas@xxxxxxx>
> > Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> > ---
> >  include/linux/mman.h           | 15 +++++++++++++++
> >  include/linux/sched/coredump.h |  6 +++++-
> >  include/uapi/linux/prctl.h     |  6 ++++++
> >  kernel/sys.c                   | 18 ++++++++++++++++++
> >  mm/mmap.c                      |  3 +++
> >  mm/mprotect.c                  |  5 +++++
> >  6 files changed, 52 insertions(+), 1 deletion(-)
> > 
> > diff --git a/include/linux/mman.h b/include/linux/mman.h
> > index 58b3abd457a3..d84fdeab6b5e 100644
> > --- a/include/linux/mman.h
> > +++ b/include/linux/mman.h
> > @@ -156,4 +156,19 @@ calc_vm_flag_bits(unsigned long flags)
> >  }
> >  
> >  unsigned long vm_commit_limit(void);
> > +
> > +static inline bool map_deny_write_exec(struct vm_area_struct *vma,  unsigned long vm_flags)
> 
> Traditionally, it is easier to write these in the positive instead of
> needing to parse a double-negative.
> 
> static inline bool allow_write_exec(...)

This doesn't feel like a double negative to me, and I think it would be better
to keep the name of the function similar to the name of the 'feature'.
However I'm not too fussed either way.

> 
> > +{
> > +	if (!test_bit(MMF_HAS_MDWE, &current->mm->flags))
> > +		return false;
> > +
> > +	if ((vm_flags & VM_EXEC) && (vm_flags & VM_WRITE))
> > +		return true;
> > +
> > +	if (vma && !(vma->vm_flags & VM_EXEC) && (vm_flags & VM_EXEC))
> > +		return true;
> > +
> > +	return false;
> > +}
> 
> Since this is implementation "2" from the earlier discussion[1], I think
> some comments in here are good to have. (e.g. to explain to people
> reading this code why there is a vma test, etc.) Perhaps even explicit
> repeat the implementation expectations.
> 
> Restating from that thread:
> 
>   2. "is not already PROT_EXEC":
> 
>      a)	mmap(PROT_READ|PROT_WRITE|PROT_EXEC);	// fails
> 
>      b)	mmap(PROT_READ|PROT_EXEC);
> 	mprotect(PROT_READ|PROT_EXEC|PROT_BTI);	// passes
> 
>      c)	mmap(PROT_READ);
> 	mprotect(PROT_READ|PROT_EXEC);		// fails
> 
>      d)	mmap(PROT_READ|PROT_WRITE);
> 	mprotect(PROT_READ);
> 	mprotect(PROT_READ|PROT_EXEC);		// fails

Good idea, I will add a comment.

> 
> [1] https://lore.kernel.org/linux-arm-kernel/YmGjYYlcSVz38rOe@xxxxxxx/
> 
> >  #endif /* _LINUX_MMAN_H */
> > diff --git a/include/linux/sched/coredump.h b/include/linux/sched/coredump.h
> > index 8270ad7ae14c..0e17ae7fbfd3 100644
> > --- a/include/linux/sched/coredump.h
> > +++ b/include/linux/sched/coredump.h
> > @@ -81,9 +81,13 @@ static inline int get_dumpable(struct mm_struct *mm)
> >   * lifecycle of this mm, just for simplicity.
> >   */
> >  #define MMF_HAS_PINNED		27	/* FOLL_PIN has run, never cleared */
> > +
> > +#define MMF_HAS_MDWE		28
> > +#define MMF_HAS_MDWE_MASK	(1 << MMF_HAS_MDWE)
> > +
> >  #define MMF_DISABLE_THP_MASK	(1 << MMF_DISABLE_THP)
> >  
> >  #define MMF_INIT_MASK		(MMF_DUMPABLE_MASK | MMF_DUMP_FILTER_MASK |\
> > -				 MMF_DISABLE_THP_MASK)
> > +				 MMF_DISABLE_THP_MASK | MMF_HAS_MDWE_MASK)
> 
> Good, yes, new "live forever" bit here. Perhaps bikeshedding over the
> name, see below.
> 
> >  
> >  #endif /* _LINUX_SCHED_COREDUMP_H */
> > diff --git a/include/uapi/linux/prctl.h b/include/uapi/linux/prctl.h
> > index a5e06dcbba13..ab9db1e86230 100644
> > --- a/include/uapi/linux/prctl.h
> > +++ b/include/uapi/linux/prctl.h
> > @@ -281,6 +281,12 @@ struct prctl_mm_map {
> >  # define PR_SME_VL_LEN_MASK		0xffff
> >  # define PR_SME_VL_INHERIT		(1 << 17) /* inherit across exec */
> >  
> > +/* Memory deny write / execute */
> > +#define PR_SET_MDWE			65
> > +# define PR_MDWE_FLAG_MMAP		1
> > +
> > +#define PR_GET_MDWE			66
> > +
> >  #define PR_SET_VMA		0x53564d41
> >  # define PR_SET_VMA_ANON_NAME		0
> >  
> > diff --git a/kernel/sys.c b/kernel/sys.c
> > index 5fd54bf0e886..08e1dd6d2533 100644
> > --- a/kernel/sys.c
> > +++ b/kernel/sys.c
> > @@ -2348,6 +2348,18 @@ static int prctl_set_vma(unsigned long opt, unsigned long start,
> >  }
> >  #endif /* CONFIG_ANON_VMA_NAME */
> >  
> > +static inline int prctl_set_mdwe(void)
> > +{
> > +	set_bit(MMF_HAS_MDWE, &current->mm->flags);
> > +
> > +	return 0;
> > +}
> > +
> > +static inline int prctl_get_mdwe(void)
> > +{
> > +	return test_bit(MMF_HAS_MDWE, &current->mm->flags);
> > +}
> 
> These will need to change -- the aren't constructed for future expansion
> at all. At the very least, all the arguments need to passed to be
> checked that they are zero. e.g.:
> 
> int prctl_set_mdwe(unsigned long bits, unsigned long arg3,
> 		   unsigned long arg4, unsigned long arg5)
> {
> 	if (arg3 || arg4 || arg5)
> 		return -EINVAL;
> 
> 	...
> 
> 	return 0;
> }
> 
> Otherwise, there's no way to add arguments in the future because old
> userspace may have been sending arbitrary junk on the stack, etc.
> 
> And regardless, I think we'll need some explicit flag bits here, since
> we can see there has been a long history of various other desired
> features that may end up living in here. For now, a single bit is fine.
> The intended behavior is the inability to _add_ PROT_EXEC to an existing
> vma, and to deny the creating of a W+X vma to begin with, so perhaps
> this bit can be named MDWE_FLAG_REFUSE_EXEC_GAIN?
> 
> Then the above "..." becomes:
> 
> 	if (bits & ~(MDWE_FLAG_REFUSE_EXEC_GAIN))
> 		return -EINVAL;
> 
> 	if (bits & MDWE_FLAG_REFUSE_EXEC_GAIN)
> 		set_bit(MMF_HAS_MDWE, &current->mm->flags);
> 	else if (test_bit(MMF_HAS_MDWE, &current->mm->flags))
> 		return -EPERM; /* Cannot unset the flag */
> 
> And prctl_get_mdwe() becomes:
> 
> int prctl_get_mdwe(unsigned long arg2, unsigned long arg3,
> 		   unsigned long arg4, unsigned long arg5)
> {
> 	if (arg2 || arg3 || arg4 || arg5)
> 		return -EINVAL;
> 	return test_bit(MMF_HAS_MDWE, &current->mm->flags) ?
> 		MDWE_FLAG_REFUSE_EXEC_GAIN : 0;
> }

Thanks, makes sense, I have incorporated those changes.

> 
> > +
> >  SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
> >  		unsigned long, arg4, unsigned long, arg5)
> >  {
> > @@ -2623,6 +2635,12 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
> >  		error = sched_core_share_pid(arg2, arg3, arg4, arg5);
> >  		break;
> >  #endif
> > +	case PR_SET_MDWE:
> > +		error = prctl_set_mdwe();
> > +		break;
> > +	case PR_GET_MDWE:
> > +		error = prctl_get_mdwe();
> > +		break;
> >  	case PR_SET_VMA:
> >  		error = prctl_set_vma(arg2, arg3, arg4, arg5);
> >  		break;
> > diff --git a/mm/mmap.c b/mm/mmap.c
> > index 099468aee4d8..42eaf6683216 100644
> > --- a/mm/mmap.c
> > +++ b/mm/mmap.c
> > @@ -1409,6 +1409,9 @@ unsigned long do_mmap(struct file *file, unsigned long addr,
> >  			vm_flags |= VM_NORESERVE;
> >  	}
> >  
> > +	if (map_deny_write_exec(NULL, vm_flags))
> > +		return -EACCES;
> > +
> 
> This seems like the wrong place to do the check -- that the vma argument
> is a hard-coded "NULL" is evidence that something is wrong. Shouldn't
> it live in mmap_region()? What happens with MAP_FIXED, when there is
> an underlying vma? i.e. an MAP_FIXED will, I think, bypass the intended
> check. For example, we had "c" above:
> 
>      c)	mmap(PROT_READ);
> 	mprotect(PROT_READ|PROT_EXEC);		// fails
> 
> But this would allow another case:
> 
>      e)	addr = mmap(..., PROT_READ, ...);
> 	mmap(addr, ..., PROT_READ | PROT_EXEC, MAP_FIXED, ...);	// passes

I can move the check into mmap_region() but it won't fix the MAP_FIXED
example that you showed here.

mmap_region() calls do_mas_munmap(..) which will unmap overlapping regions.
However the `vma` for the 'old' region is not kept around, and a new vma will
be allocated later on "vma = vm_area_alloc(mm);", and the vm_flags are just set
to what is passed into mmap_region(), so map_deny_write_exec(vma, vm_flags)
will just be as good as passing NULL.

It's possible to save the vm_flags from the region that is unmapped, but Catalin
suggested it might be better if that is part of a later extension, what do you
think? 

> 
> 
> >  	addr = mmap_region(file, addr, len, vm_flags, pgoff, uf);
> >  	if (!IS_ERR_VALUE(addr) &&
> >  	    ((vm_flags & VM_LOCKED) ||
> > diff --git a/mm/mprotect.c b/mm/mprotect.c
> > index 8d770855b591..af71ef0788fd 100644
> > --- a/mm/mprotect.c
> > +++ b/mm/mprotect.c
> > @@ -766,6 +766,11 @@ static int do_mprotect_pkey(unsigned long start, size_t len,
> >  			break;
> >  		}
> >  
> > +		if (map_deny_write_exec(vma, newflags)) {
> > +			error = -EACCES;
> > +			goto out;
> > +		}
> > +
> 
> This looks like the right place. Any rationale for why it's before
> arch_validate_flags()?o

No big justification, it's just after the VM_ACCESS_FLAGS check and is more generic
than the architecture specific checks.

> 
> >  		/* Allow architectures to sanity-check the new flags */
> >  		if (!arch_validate_flags(newflags)) {
> >  			error = -EINVAL;
> 
> -Kees

Thanks for the review and for the rewritten test, I have replaced my commit with
the one that you sent.

Joey