Skip to contentSkip to conversation

mm: abstract shadow stack vma behind arch_is_shadow_stack_vma

PATCH In review Deepak Gupta 4 participants 9 messages Dec 22, 2023 – Jan 2, 2024

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mm: abstract shadow stack vma behind arch_is_shadow_stack_vma

x86 has used VM_SHADOW_STACK (alias to VM_HIGH_ARCH_5) to encode shadow stack VMA. VM_SHADOW_STACK is thus not possible on 32bit. Some arches may need a way to encode shadow stack on 32bit and 64bit both and they may encode this information differently in VMAs.

This patch changes checks of VM_SHADOW_STACK flag in generic code to call to a function arch_is_shadow_stack_vma which will return true if arch supports shadow stack and vma is shadow stack else stub returns false.

---
 include/linux/mm.h | 15 ++++++++++++++-
 mm/gup.c           |  2 +-
 mm/internal.h      |  2 +-
 3 files changed, 16 insertions(+), 3 deletions(-)
include/linux/mm.h +14 -1
index 418d26608ece..9586e7bbd2aa 100644
@@ -352,8 +352,21 @@ extern unsigned int kobjsize(const void *objp);
  * for more details on the guard size.
  */
 # define VM_SHADOW_STACK	VM_HIGH_ARCH_5
+
+static inline bool arch_is_shadow_stack_vma(vm_flags_t vm_flags)
+{
+	return (vm_flags & VM_SHADOW_STACK) ? true : false;
+}
+
 #else
+
 # define VM_SHADOW_STACK	VM_NONE
+
+static inline bool arch_is_shadow_stack_vma(vm_flags_t vm_flags)
+{
+	return false;
+}
+
 #endif
 
 #if defined(CONFIG_X86)
@@ -3452,7 +3465,7 @@ static inline unsigned long stack_guard_start_gap(struct vm_area_struct *vma)
 		return stack_guard_gap;
 
 	/* See reasoning around the VM_SHADOW_STACK definition */
-	if (vma->vm_flags & VM_SHADOW_STACK)
+	if (arch_is_shadow_stack_vma(vma->vm_flags))
 		return PAGE_SIZE;
 
 	return 0;
mm/gup.c +1 -1
index 231711efa390..dcc2aa079163 100644
@@ -1051,7 +1051,7 @@ static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags)
 		    !writable_file_mapping_allowed(vma, gup_flags))
 			return -EFAULT;
 
-		if (!(vm_flags & VM_WRITE) || (vm_flags & VM_SHADOW_STACK)) {
+		if (!(vm_flags & VM_WRITE) || arch_is_shadow_stack_vma(vm_flags)) {
 			if (!(gup_flags & FOLL_FORCE))
 				return -EFAULT;
 			/* hugetlb does not support FOLL_FORCE|FOLL_WRITE. */
mm/internal.h +1 -1
index b61034bd50f5..05a6b47c3ca1 100644
@@ -572,7 +572,7 @@ static inline bool is_exec_mapping(vm_flags_t flags)
  */
 static inline bool is_stack_mapping(vm_flags_t flags)
 {
-	return ((flags & VM_STACK) == VM_STACK) || (flags & VM_SHADOW_STACK);
+	return ((flags & VM_STACK) == VM_STACK) || arch_is_shadow_stack_vma(flags);
 }
 
 /*
Signature · 1 line
2.43.0
maintainer
#1 lore raw
Quoted 4 lines · On Fri, 22 Dec 2023 15:51:04 -0800 Deepak Gupta <[email protected]> wrote:

x86 has used VM_SHADOW_STACK (alias to VM_HIGH_ARCH_5) to encode shadow stack VMA. VM_SHADOW_STACK is thus not possible on 32bit. Some arches may need a way to encode shadow stack on 32bit and 64bit both and they may encode this information differently in VMAs.

Is such a patch in the pipeline? Otherwise we're making a change that serves no purpose.

Quoted 9 lines

This patch changes checks of VM_SHADOW_STACK flag in generic code to call to a function arch_is_shadow_stack_vma which will return true if arch supports shadow stack and vma is shadow stack else stub returns false.

...

include/linux/mm.h extern unsigned int kobjsize(const void *objp); · line 359

Quoted 8 lines
include/linux/mm.h +5 -0
@@ -352,8 +352,21 @@ extern unsigned int kobjsize(const void *objp);
  * for more details on the guard size.
  */
 # define VM_SHADOW_STACK	VM_HIGH_ARCH_5
+
+static inline bool arch_is_shadow_stack_vma(vm_flags_t vm_flags)
+{
+	return (vm_flags & VM_SHADOW_STACK) ? true : false;
+}

The naming seems a little wrong. I'd expect it to take a vma* arg. Maybe just drop the "_vma"?

#2 lore raw
Quoted 10 lines · Andrew Morton
Quoted 4 lines · On Fri, 22 Dec 2023 15:51:04 -0800 Deepak Gupta <[email protected]> wrote:

x86 has used VM_SHADOW_STACK (alias to VM_HIGH_ARCH_5) to encode shadow stack VMA. VM_SHADOW_STACK is thus not possible on 32bit. Some arches may need a way to encode shadow stack on 32bit and 64bit both and they may encode this information differently in VMAs.

Is such a patch in the pipeline? Otherwise we're making a change that serves no purpose.

Yes I do have patches in the pipeline for riscv.
On riscv, presence of only VM_WRITE (i.e. (flags & (VM_READ | VM_WRITE | VM_EXEC))
== VM_WRITE) would mean a shadow stack.
And yes there would be relevant patches to ensure that existing consumers using PROT_WRITE gets translated to (VM_WRITE | VM_READ)

Quoted 7 lines
Quoted 7 lines

This patch changes checks of VM_SHADOW_STACK flag in generic code to call to a function arch_is_shadow_stack_vma which will return true if arch supports shadow stack and vma is shadow stack else stub returns false.

...

Quoted 14 lines

include/linux/mm.h extern unsigned int kobjsize(const void *objp); · line 359

Quoted 11 lines
include/linux/mm.h +5 -0
@@ -352,8 +352,21 @@ extern unsigned int kobjsize(const void *objp);
  * for more details on the guard size.
  */
 # define VM_SHADOW_STACK     VM_HIGH_ARCH_5
+
+static inline bool arch_is_shadow_stack_vma(vm_flags_t vm_flags)
+{
+     return (vm_flags & VM_SHADOW_STACK) ? true : false;
+}

The naming seems a little wrong. I'd expect it to take a vma* arg. Maybe just drop the "_vma"?

Well I did start with taking vma* argument but then realized that is_stack_mapping
only takes vma flags. And in order to change that I would have to change vm_stat_account
and every place it's called.

In the next version I'll either do that or drop _vma from the proposed function name.

Quoted 1 line
maintainer
#3 lore raw
Quoted 12 lines · On Wed, 27 Dec 2023 14:20:36 -0800 Deepak Gupta <[email protected]> wrote:
Quoted 10 lines · Andrew Morton
Quoted 4 lines · On Fri, 22 Dec 2023 15:51:04 -0800 Deepak Gupta <[email protected]> wrote:

x86 has used VM_SHADOW_STACK (alias to VM_HIGH_ARCH_5) to encode shadow stack VMA. VM_SHADOW_STACK is thus not possible on 32bit. Some arches may need a way to encode shadow stack on 32bit and 64bit both and they may encode this information differently in VMAs.

Is such a patch in the pipeline? Otherwise we're making a change that serves no purpose.

Quoted 6 lines

Yes I do have patches in the pipeline for riscv.
On riscv, presence of only VM_WRITE (i.e. (flags & (VM_READ | VM_WRITE | VM_EXEC))
== VM_WRITE) would mean a shadow stack.
And yes there would be relevant patches to ensure that existing consumers using PROT_WRITE gets translated to (VM_WRITE | VM_READ)

OK, please plan to carry this patch in whatever tree contains the above.

#4 lore raw
Quoted 15 lines · Andrew Morton
Quoted 12 lines · On Wed, 27 Dec 2023 14:20:36 -0800 Deepak Gupta <[email protected]> wrote:
Quoted 10 lines · Andrew Morton
Quoted 4 lines · On Fri, 22 Dec 2023 15:51:04 -0800 Deepak Gupta <[email protected]> wrote:

x86 has used VM_SHADOW_STACK (alias to VM_HIGH_ARCH_5) to encode shadow stack VMA. VM_SHADOW_STACK is thus not possible on 32bit. Some arches may need a way to encode shadow stack on 32bit and 64bit both and they may encode this information differently in VMAs.

Is such a patch in the pipeline? Otherwise we're making a change that serves no purpose.

Quoted 10 lines
Quoted 6 lines

Yes I do have patches in the pipeline for riscv.
On riscv, presence of only VM_WRITE (i.e. (flags & (VM_READ | VM_WRITE | VM_EXEC))
== VM_WRITE) would mean a shadow stack.
And yes there would be relevant patches to ensure that existing consumers using PROT_WRITE gets translated to (VM_WRITE | VM_READ)

OK, please plan to carry this patch in whatever tree contains the above.

ACK. Thanks

reviewer
#5 lore raw
In reply to Andrew Morton
Quoted 16 lines · Andrew Morton
Quoted 4 lines · On Fri, 22 Dec 2023 15:51:04 -0800 Deepak Gupta <[email protected]> wrote:

x86 has used VM_SHADOW_STACK (alias to VM_HIGH_ARCH_5) to encode shadow stack VMA. VM_SHADOW_STACK is thus not possible on 32bit. Some arches may need a way to encode shadow stack on 32bit and 64bit both and they may encode this information differently in VMAs.

Is such a patch in the pipeline? Otherwise we're making a change that serves no purpose.

Quoted 6 lines

This patch changes checks of VM_SHADOW_STACK flag in generic code to call to a function arch_is_shadow_stack_vma which will return true if arch supports shadow stack and vma is shadow stack else stub returns false.

...

Quoted 14 lines

include/linux/mm.h extern unsigned int kobjsize(const void *objp); · line 359

Quoted 11 lines
include/linux/mm.h +5 -0
@@ -352,8 +352,21 @@ extern unsigned int kobjsize(const void *objp);
  * for more details on the guard size.
  */
 # define VM_SHADOW_STACK	VM_HIGH_ARCH_5
+
+static inline bool arch_is_shadow_stack_vma(vm_flags_t vm_flags)
+{
+	return (vm_flags & VM_SHADOW_STACK) ? true : false;
+}

The naming seems a little wrong. I'd expect it to take a vma* arg. Maybe just drop the "_vma"?

I'd suggest to use vma_is_shadow_stack() to make it inline with other vma_is_*() tests.

Signature · 2 lines
Sincerely yours,
Mike.
#6 lore raw
Quoted 33 lines · Mike Rapoport
Quoted 16 lines · Andrew Morton
Quoted 4 lines · On Fri, 22 Dec 2023 15:51:04 -0800 Deepak Gupta <[email protected]> wrote:

x86 has used VM_SHADOW_STACK (alias to VM_HIGH_ARCH_5) to encode shadow stack VMA. VM_SHADOW_STACK is thus not possible on 32bit. Some arches may need a way to encode shadow stack on 32bit and 64bit both and they may encode this information differently in VMAs.

Is such a patch in the pipeline? Otherwise we're making a change that serves no purpose.

Quoted 6 lines

This patch changes checks of VM_SHADOW_STACK flag in generic code to call to a function arch_is_shadow_stack_vma which will return true if arch supports shadow stack and vma is shadow stack else stub returns false.

...

Quoted 14 lines

include/linux/mm.h extern unsigned int kobjsize(const void *objp); · line 359

Quoted 11 lines
include/linux/mm.h +5 -0
@@ -352,8 +352,21 @@ extern unsigned int kobjsize(const void *objp);
  * for more details on the guard size.
  */
 # define VM_SHADOW_STACK   VM_HIGH_ARCH_5
+
+static inline bool arch_is_shadow_stack_vma(vm_flags_t vm_flags)
+{
+   return (vm_flags & VM_SHADOW_STACK) ? true : false;
+}

The naming seems a little wrong. I'd expect it to take a vma* arg. Maybe just drop the "_vma"?

Quoted 2 lines

I'd suggest to use vma_is_shadow_stack() to make it inline with other vma_is_*() tests.

Noted. Thanks

Quoted 4 lines

--
Sincerely yours,
Mike.

#7 lore raw
In reply to Deepak Gupta

include/linux/mm.h extern unsigned int kobjsize(const void *objp); · line 360

Quoted 6 lines · Deepak Gupta
include/linux/mm.h +6 -0
+
+static inline bool arch_is_shadow_stack_vma(vm_flags_t vm_flags)
+{
+       return (vm_flags & VM_SHADOW_STACK) ? true : false;
+}
+

The bit after the "?" should be unneeded. I would think that patterns like:

bool res = val ? true : false;

...should never be needed for the kernel's current bool typedef. Is there some special arch define consideration or something, I'm unaware of?

https://www.kernel.org/doc/html/latest/process/coding-style.html#using-bool

#8 lore raw
Quoted 12 lines · Rick P

include/linux/mm.h extern unsigned int kobjsize(const void *objp); · line 360

Quoted 6 lines · Deepak Gupta
include/linux/mm.h +6 -0
+
+static inline bool arch_is_shadow_stack_vma(vm_flags_t vm_flags)
+{
+       return (vm_flags & VM_SHADOW_STACK) ? true : false;
+}
+

The bit after the "?" should be unneeded. I would think that patterns like:

Quoted 7 lines
   bool res = val ? true : false;

...should never be needed for the kernel's current bool typedef. Is there some special arch define consideration or something, I'm unaware of?

https://www.kernel.org/doc/html/latest/process/coding-style.html#using-bool

Thanks. Just checked out the link you sent.
Yes it's not needed. Will remove it.