[PATCH 0/4] mm/mempolicy: kernel_migrate_pages: fix race between security checks and suid exec

[PATCH 0/4] mm/mempolicy: kernel_migrate_pages: fix race between security checks and suid exec

[Oleg Nesterov]

Hello,

Sorry, I have no idea how to test these changes. Please review.
Hopefully sashiko.dev will take a look too ;)

find_mm_struct() has the same problem, will send a patch tomorrow.
Unless you tell me this race doesn't need a fix.

Oleg.
---

 mm/mempolicy.c | 53 +++++++++++++++++++++++++----------------------------
 1 file changed, 25 insertions(+), 28 deletions(-)

[PATCH 1/4] mm/mempolicy: kernel_migrate_pages: simplify the usage of put_task_struct()

[Oleg Nesterov]

kernel_migrate_pages() calls put_task_struct() twice; this complicates the
code. Move put_task_struct() to the unified exit path. This also allows us
to move the "out_put" label up and avoid another goto.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---
 mm/mempolicy.c | 11 +++--------
 1 file changed, 3 insertions(+), 8 deletions(-)

diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 4e4421b22b59..4326dff16aa6 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -1922,25 +1922,20 @@ static int kernel_migrate_pages(pid_t pid, unsigned long maxnode,
 		goto out_put;
 
 	mm = get_task_mm(task);
-	put_task_struct(task);
-
 	if (!mm) {
 		err = -EINVAL;
-		goto out;
+		goto out_put;
 	}
 
 	err = do_migrate_pages(mm, old, new,
 		capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
 
 	mmput(mm);
+out_put:
+	put_task_struct(task);
 out:
 	NODEMASK_SCRATCH_FREE(scratch);
-
 	return err;
-
-out_put:
-	put_task_struct(task);
-	goto out;
 }
 
 SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
-- 
2.52.0

Re: [PATCH 1/4] mm/mempolicy: kernel_migrate_pages: simplify the usage of put_task_struct()

[Gregory Price]

On Sun, May 24, 2026 at 06:57:59PM +0200, Oleg Nesterov wrote:
> kernel_migrate_pages() calls put_task_struct() twice; this complicates the
> code. Move put_task_struct() to the unified exit path. This also allows us
> to move the "out_put" label up and avoid another goto.
> 
> Signed-off-by: Oleg Nesterov <oleg@redhat.com>

Reviewed-by: Gregory Price <gourry@gourry.net>

> ---
>  mm/mempolicy.c | 11 +++--------
>  1 file changed, 3 insertions(+), 8 deletions(-)
> 
> diff --git a/mm/mempolicy.c b/mm/mempolicy.c
> index 4e4421b22b59..4326dff16aa6 100644
> --- a/mm/mempolicy.c
> +++ b/mm/mempolicy.c
> @@ -1922,25 +1922,20 @@ static int kernel_migrate_pages(pid_t pid, unsigned long maxnode,
>  		goto out_put;
>  
>  	mm = get_task_mm(task);
> -	put_task_struct(task);
> -
>  	if (!mm) {
>  		err = -EINVAL;
> -		goto out;
> +		goto out_put;
>  	}
>  
>  	err = do_migrate_pages(mm, old, new,
>  		capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
>  
>  	mmput(mm);
> +out_put:
> +	put_task_struct(task);
>  out:
>  	NODEMASK_SCRATCH_FREE(scratch);
> -
>  	return err;
> -
> -out_put:
> -	put_task_struct(task);
> -	goto out;
>  }
>  
>  SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
> -- 
> 2.52.0
> 

Re: [PATCH 1/4] mm/mempolicy: kernel_migrate_pages: simplify the usage of put_task_struct()

[Gregory Price]

On Sun, May 24, 2026 at 06:57:59PM +0200, Oleg Nesterov wrote:
> kernel_migrate_pages() calls put_task_struct() twice; this complicates the
> code. Move put_task_struct() to the unified exit path. This also allows us
> to move the "out_put" label up and avoid another goto.
> 
> Signed-off-by: Oleg Nesterov <oleg@redhat.com>
> ---
>  mm/mempolicy.c | 11 +++--------
>  1 file changed, 3 insertions(+), 8 deletions(-)
> 
> diff --git a/mm/mempolicy.c b/mm/mempolicy.c
> index 4e4421b22b59..4326dff16aa6 100644
> --- a/mm/mempolicy.c
> +++ b/mm/mempolicy.c
> @@ -1922,25 +1922,20 @@ static int kernel_migrate_pages(pid_t pid, unsigned long maxnode,
>  		goto out_put;
>  
>  	mm = get_task_mm(task);
> -	put_task_struct(task);
> -

Hm, I was a little trigger-happy on that review, I just realized that
this increases the scope in which there is a task reference held (i.e.
we're now holding a reference across do_migrate_pages())

I don't *think* that's an issue, but it might we worth looking at the
changelog to see if this was done intentionally.

>  	if (!mm) {
>  		err = -EINVAL;
> -		goto out;
> +		goto out_put;
>  	}
>  
>  	err = do_migrate_pages(mm, old, new,
>  		capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
>  
>  	mmput(mm);
> +out_put:
> +	put_task_struct(task);
>  out:
>  	NODEMASK_SCRATCH_FREE(scratch);
> -
>  	return err;
> -
> -out_put:
> -	put_task_struct(task);
> -	goto out;
>  }
>  
>  SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
> -- 
> 2.52.0
> 

Re: [PATCH 1/4] mm/mempolicy: kernel_migrate_pages: simplify the usage of put_task_struct()

[Oleg Nesterov]

On 05/24, Gregory Price wrote:
>
> On Sun, May 24, 2026 at 06:57:59PM +0200, Oleg Nesterov wrote:
> >
> >  	mm = get_task_mm(task);
> > -	put_task_struct(task);
> > -
>
> Hm, I was a little trigger-happy on that review, I just realized that
> this increases the scope in which there is a task reference held (i.e.
> we're now holding a reference across do_migrate_pages())
>
> I don't *think* that's an issue, but it might we worth looking at the
> changelog to see if this was done intentionally.

I did this to simplify the code and the next changes. We pin this task's
memory across do_migrate_pages(), so I think that delaying put_task_struct()
doesn't add too much.

But of course, this is not necessary, I can rework if you don't like it.

Thanks for review!

Oleg.

Re: [PATCH 1/4] mm/mempolicy: kernel_migrate_pages: simplify the usage of put_task_struct()

[Gregory Price]

On Mon, May 25, 2026 at 10:27:06AM +0200, Oleg Nesterov wrote:
> On 05/24, Gregory Price wrote:
> >
> > On Sun, May 24, 2026 at 06:57:59PM +0200, Oleg Nesterov wrote:
> > >
> > >  	mm = get_task_mm(task);
> > > -	put_task_struct(task);
> > > -
> >
> > Hm, I was a little trigger-happy on that review, I just realized that
> > this increases the scope in which there is a task reference held (i.e.
> > we're now holding a reference across do_migrate_pages())
> >
> > I don't *think* that's an issue, but it might we worth looking at the
> > changelog to see if this was done intentionally.
> 
> I did this to simplify the code and the next changes. We pin this task's
> memory across do_migrate_pages(), so I think that delaying put_task_struct()
> doesn't add too much.
> 
> But of course, this is not necessary, I can rework if you don't like it.
>

Not necessary, just a curiosity whether holding the reference across the
larger scope has any functional difference.  Like I said, I don't think
that's an issue, but might be worth just checking the history on the
function to see if it was done that way on purpose.

> Thanks for review!
> 
> Oleg.
> 

[PATCH 2/4] mm/mempolicy: kernel_migrate_pages: use find_get_task_by_vpid()

[Oleg Nesterov]

kernel_migrate_pages() calls ptrace_may_access() under rcu_read_lock() for
no reason. This is a leftover from before the commit 313674661925 ("Unify
migrate_pages and move_pages access checks"), where rcu_read_lock() was
needed to protect __task_cred(task).

So we can drop the RCU lock right after get_task_struct(). Better yet, if
pid != 0, we can use find_get_task_by_vpid() which does get_task_struct()
itself, and avoid get/put_task_struct() otherwise.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---
 mm/mempolicy.c | 11 +++--------
 1 file changed, 3 insertions(+), 8 deletions(-)

diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 4326dff16aa6..2ec14001e4dc 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -1883,15 +1883,11 @@ static int kernel_migrate_pages(pid_t pid, unsigned long maxnode,
 	if (err)
 		goto out;
 
-	/* Find the mm_struct */
-	rcu_read_lock();
-	task = pid ? find_task_by_vpid(pid) : current;
+	task = pid ? find_get_task_by_vpid(pid) : current;
 	if (!task) {
-		rcu_read_unlock();
 		err = -ESRCH;
 		goto out;
 	}
-	get_task_struct(task);
 
 	err = -EINVAL;
 
@@ -1900,11 +1896,9 @@ static int kernel_migrate_pages(pid_t pid, unsigned long maxnode,
 	 * Use the regular "ptrace_may_access()" checks.
 	 */
 	if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) {
-		rcu_read_unlock();
 		err = -EPERM;
 		goto out_put;
 	}
-	rcu_read_unlock();
 
 	task_nodes = cpuset_mems_allowed(task);
 	/* Is the user allowed to access the target nodes? */
@@ -1932,7 +1926,8 @@ static int kernel_migrate_pages(pid_t pid, unsigned long maxnode,
 
 	mmput(mm);
 out_put:
-	put_task_struct(task);
+	if (pid)
+		put_task_struct(task);
 out:
 	NODEMASK_SCRATCH_FREE(scratch);
 	return err;
-- 
2.52.0

[PATCH 3/4] mm/mempolicy: kernel_migrate_pages: check ptrace_may_access() after nodes_and()

[Oleg Nesterov]

Move the ptrace_may_access() check down, after the nodemask validation.
This is just a preparation to simplify the next patch.

Note that because of this reordering, if the caller passes an invalid
nodemask and also lacks permissions, the syscall will now return -EINVAL
instead of -EPERM. Hopefully, nobody relies on the old error precedence.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---
 mm/mempolicy.c | 24 ++++++++++++------------
 1 file changed, 12 insertions(+), 12 deletions(-)

diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 2ec14001e4dc..c09ff9f9aa96 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -1889,17 +1889,6 @@ static int kernel_migrate_pages(pid_t pid, unsigned long maxnode,
 		goto out;
 	}
 
-	err = -EINVAL;
-
-	/*
-	 * Check if this process has the right to modify the specified process.
-	 * Use the regular "ptrace_may_access()" checks.
-	 */
-	if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) {
-		err = -EPERM;
-		goto out_put;
-	}
-
 	task_nodes = cpuset_mems_allowed(task);
 	/* Is the user allowed to access the target nodes? */
 	if (!nodes_subset(*new, task_nodes) && !capable(CAP_SYS_NICE)) {
@@ -1908,8 +1897,19 @@ static int kernel_migrate_pages(pid_t pid, unsigned long maxnode,
 	}
 
 	task_nodes = cpuset_mems_allowed(current);
-	if (!nodes_and(*new, *new, task_nodes))
+	if (!nodes_and(*new, *new, task_nodes)) {
+		err = -EINVAL;
 		goto out_put;
+	}
+
+	/*
+	 * Check if this process has the right to modify the specified process.
+	 * Use the regular "ptrace_may_access()" checks.
+	 */
+	if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) {
+		err = -EPERM;
+		goto out_put;
+	}
 
 	err = security_task_movememory(task);
 	if (err)
-- 
2.52.0

[PATCH 4/4] mm/mempolicy: kernel_migrate_pages: fix race between security checks and suid exec

[Oleg Nesterov]

The target task can execute a setuid binary between ptrace_may_access()
and get_task_mm(). Protect this critical section with exec_update_lock.

Sadly, we don't have DEFINE_LOCK_GUARD_1_COND(rwsem_read, _kill) yet.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---
 mm/mempolicy.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index c09ff9f9aa96..cbb45a876a93 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -1902,24 +1902,31 @@ static int kernel_migrate_pages(pid_t pid, unsigned long maxnode,
 		goto out_put;
 	}
 
+	err = down_read_killable(&task->signal->exec_update_lock);
+	if (err)
+		goto out_put;
 	/*
 	 * Check if this process has the right to modify the specified process.
 	 * Use the regular "ptrace_may_access()" checks.
 	 */
 	if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) {
 		err = -EPERM;
-		goto out_put;
+		goto unlock;
 	}
 
 	err = security_task_movememory(task);
 	if (err)
-		goto out_put;
+		goto unlock;
 
 	mm = get_task_mm(task);
 	if (!mm) {
 		err = -EINVAL;
-		goto out_put;
+		goto unlock;
 	}
+unlock:
+	up_read(&task->signal->exec_update_lock);
+	if (err)
+		goto out_put;
 
 	err = do_migrate_pages(mm, old, new,
 		capable(CAP_SYS_NICE) ? MPOL_MF_MOVE_ALL : MPOL_MF_MOVE);
-- 
2.52.0

Re: [PATCH 0/4] mm/mempolicy: kernel_migrate_pages: fix race between security checks and suid exec

[Oleg Nesterov]

On 05/24, Oleg Nesterov wrote:
>
> Hopefully sashiko.dev will take a look too ;)

See https://sashiko.dev/#/patchset/ahMt6xyUNnacZU8-%40redhat.com
Thanks Sashiko, let me reply.

As for reply to 3/4:

	I honestly think this is not a problem... but a) I will think
	again, and b) I rely on review from mm experts anyway.

	The main motivation for this patch is that we can create a common
	helper for find_mm_struct() and kernel_migrate_pages(), but it is
	not strictly necessary.

As for reply to 4/4:

	Yes! And I have already mentioned this in 0/4 and in the previous
	discussions.

	Plus (say) proc_coredump_filter_write() which relies on the open-time
	checks and don't use ptrace_may_access() at all. This looks wrong.

	But this need other changes/discussions, so I think we can ignore this
	comment.

Oleg.