[PATCH] exit: use READ_ONCE() to read mm->owner

[PATCH] exit: use READ_ONCE() to read mm->owner

[Mani Tofigh via B4 Relay]

From: Mani Tofigh <mt4022@columbia.edu>

mm->owner is written with WRITE_ONCE() when ownership moves to another
task, but mm_update_next_owner() reads it with a plain load.  The race
is expected, so read it with READ_ONCE() to match the writer.

  [   14.915975] BUG: KCSAN: data-race in mm_update_next_owner / try_to_set_owner

  [   14.916154] write (marked) to 0xff11000004471b80 of 8 bytes by task 71 on cpu 0:
  [   14.916204]  try_to_set_owner+0x100/0x140
  [   14.916302]  mm_update_next_owner+0x177/0x300
  ...

  [   14.917246] read to 0xff11000004471b80 of 8 bytes by task 75 on cpu 1:
  [   14.917293]  mm_update_next_owner+0x2a/0x300
  [   14.917396]  do_exit+0x575/0x18b0
  ...

  [   14.923906] value changed: 0xff1100002ec04180 -> 0xff1100002ed3a0c0

Found with KCSAN while fuzzing under a deterministic hypervisor that's
currently under development.

Signed-off-by: Mani Tofigh <mt4022@columbia.edu>
---
 kernel/exit.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/exit.c b/kernel/exit.c
index d5e7bc2af804..cc78596db74d 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -494,7 +494,7 @@ void mm_update_next_owner(struct mm_struct *mm)
 	 * If the exiting or execing task is not the owner, it's
 	 * someone else's problem.
 	 */
-	if (mm->owner != p)
+	if (READ_ONCE(mm->owner) != p)
 		return;
 	/*
 	 * The current owner is exiting/execing and there are no other

---
base-commit: d135ab453a7400a34930bdda1ef4bdc8acfac80e
change-id: 20261027-kcsan-exit-109dce625d81

Best regards,
--  
Mani Tofigh <mt4022@columbia.edu>

Re: [PATCH] exit: use READ_ONCE() to read mm->owner

[Balbir Singh]

On Tue, Oct 27, 2026 at 04:09:23PM -0400, Mani Tofigh via B4 Relay wrote:
> From: Mani Tofigh <mt4022@columbia.edu>
> 
> mm->owner is written with WRITE_ONCE() when ownership moves to another
> task, but mm_update_next_owner() reads it with a plain load.  The race
> is expected, so read it with READ_ONCE() to match the writer.
> 
>   [   14.915975] BUG: KCSAN: data-race in mm_update_next_owner / try_to_set_owner
> 
>   [   14.916154] write (marked) to 0xff11000004471b80 of 8 bytes by task 71 on cpu 0:
>   [   14.916204]  try_to_set_owner+0x100/0x140
>   [   14.916302]  mm_update_next_owner+0x177/0x300
>   ...
> 
>   [   14.917246] read to 0xff11000004471b80 of 8 bytes by task 75 on cpu 1:
>   [   14.917293]  mm_update_next_owner+0x2a/0x300
>   [   14.917396]  do_exit+0x575/0x18b0
>   ...
> 
>   [   14.923906] value changed: 0xff1100002ec04180 -> 0xff1100002ed3a0c0
> 
> Found with KCSAN while fuzzing under a deterministic hypervisor that's
> currently under development.
> 
> Signed-off-by: Mani Tofigh <mt4022@columbia.edu>
> ---
>  kernel/exit.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/kernel/exit.c b/kernel/exit.c
> index d5e7bc2af804..cc78596db74d 100644
> --- a/kernel/exit.c
> +++ b/kernel/exit.c
> @@ -494,7 +494,7 @@ void mm_update_next_owner(struct mm_struct *mm)
>  	 * If the exiting or execing task is not the owner, it's
>  	 * someone else's problem.
>  	 */
> -	if (mm->owner != p)
> +	if (READ_ONCE(mm->owner) != p)
>  		return;
>  	/*
>  	 * The current owner is exiting/execing and there are no other
>

Acked-by: Balbir Singh <balbirs@nvidia.com>

Balbir

Re: [PATCH] exit: use READ_ONCE() to read mm->owner

[Oleg Nesterov]

On 10/27, Mani Tofigh via B4 Relay wrote:
>
> --- a/kernel/exit.c
> +++ b/kernel/exit.c
> @@ -494,7 +494,7 @@ void mm_update_next_owner(struct mm_struct *mm)
>  	 * If the exiting or execing task is not the owner, it's
>  	 * someone else's problem.
>  	 */
> -	if (mm->owner != p)
> +	if (READ_ONCE(mm->owner) != p)

Acked-by: Oleg Nesterov <oleg@redhat.com>

Re: [PATCH] exit: use READ_ONCE() to read mm->owner

[Andrew Morton]

On Tue, 27 Oct 2026 16:09:23 -0400 Mani Tofigh via B4 Relay <devnull+mt4022.columbia.edu@kernel.org> wrote:

> mm->owner is written with WRITE_ONCE() when ownership moves to another
> task, but mm_update_next_owner() reads it with a plain load.  The race
> is expected, so read it with READ_ONCE() to match the writer.
> 
>   [   14.915975] BUG: KCSAN: data-race in mm_update_next_owner / try_to_set_owner
> 
>   [   14.916154] write (marked) to 0xff11000004471b80 of 8 bytes by task 71 on cpu 0:
>   [   14.916204]  try_to_set_owner+0x100/0x140
>   [   14.916302]  mm_update_next_owner+0x177/0x300
>   ...
> 
>   [   14.917246] read to 0xff11000004471b80 of 8 bytes by task 75 on cpu 1:
>   [   14.917293]  mm_update_next_owner+0x2a/0x300
>   [   14.917396]  do_exit+0x575/0x18b0
>   ...
> 
>   [   14.923906] value changed: 0xff1100002ec04180 -> 0xff1100002ed3a0c0
> 
> Found with KCSAN while fuzzing under a deterministic hypervisor that's
> currently under development.

We don't want older kernels doing this either, so I suggest

Fixes: cf475ad28ac3 ("cgroups: add an owner to the mm_struct")
Cc: <stable$vger.kernel.org>