[PATCH 0/1] 3rd resend: kill __task_cred()->task_is_dead()

[PATCH 0/1] 3rd resend: kill __task_cred()->task_is_dead()

[Oleg Nesterov]

Hello.

The patch was acked by David on 2011-09-20, see
http://marc.info/?l=linux-kernel&m=131653612415506

I didn't dare to keep this ack because the change in cred.h
was updated.

task_is_dead() must die ;) Another user, usbip, was already
fixed in the staging git.

Oleg.

[PATCH 1/1] creds: kill __task_cred()->task_is_dead() validation

[Oleg Nesterov]

commit 8f92054e:

    add the following validation condition:

        task->exit_state >= 0

    to permit the access if the target task is dead and therefore
    unable to change its own credentials.

OK, but afaics currently this can only help wait_task_zombie() which
calls __task_cred() without rcu lock.

Remove this validation and change wait_task_zombie() to use task_uid()
instead. This means we do rcu_read_lock() only to shut up the lockdep,
but we already do the same in, say, wait_task_stopped().

task_is_dead() should die, task->exit_state != 0 means that this task
has passed exit_notify(), only do_wait-like code paths should use this.

Unfortunately, we can't kill task_is_dead() right now, it has already
found the bugy users in drivers/staging/, the fix already exists.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---
 include/linux/cred.h |   10 +++-------
 kernel/exit.c        |    2 +-
 2 files changed, 4 insertions(+), 8 deletions(-)

diff --git a/include/linux/cred.h b/include/linux/cred.h
index adadf71..1b64c72 100644
--- a/include/linux/cred.h
+++ b/include/linux/cred.h
@@ -276,17 +276,13 @@ static inline void put_cred(const struct cred *_cred)
  * @task: The task to query
  *
  * Access the objective credentials of a task.  The caller must hold the RCU
- * readlock or the task must be dead and unable to change its own credentials.
+ * readlock.
  *
  * The result of this function should not be passed directly to get_cred();
  * rather get_task_cred() should be used instead.
  */
-#define __task_cred(task)						\
-	({								\
-		const struct task_struct *__t = (task);			\
-		rcu_dereference_check(__t->real_cred,			\
-				      task_is_dead(__t));		\
-	})
+#define __task_cred(task)	\
+	rcu_dereference((task)->real_cred)
 
 /**
  * get_current_cred - Get the current task's subjective credentials
diff --git a/kernel/exit.c b/kernel/exit.c
index 4b4042f..7b36288 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -1214,7 +1213,7 @@ static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
 	unsigned long state;
 	int retval, status, traced;
 	pid_t pid = task_pid_vnr(p);
-	uid_t uid = __task_cred(p)->uid;
+	uid_t uid = task_uid(p);
 	struct siginfo __user *infop;
 
 	if (!likely(wo->wo_flags & WEXITED))
-- 
1.5.5.1

Re: [PATCH 1/1] creds: kill __task_cred()->task_is_dead() validation

[Eric W. Biederman]

Oleg Nesterov <oleg@redhat.com> writes:

A small nit.  The subject should be:
"Remove task_is_dead from __task_cred() validation."
there is no method __task_cred()->task_is_dead().


> commit 8f92054e:
>
>     add the following validation condition:
>
>         task->exit_state >= 0
>
>     to permit the access if the target task is dead and therefore
>     unable to change its own credentials.
>
> OK, but afaics currently this can only help wait_task_zombie() which
> calls __task_cred() without rcu lock.
>
> Remove this validation and change wait_task_zombie() to use task_uid()
> instead. This means we do rcu_read_lock() only to shut up the lockdep,
> but we already do the same in, say, wait_task_stopped().
>
> task_is_dead() should die, task->exit_state != 0 means that this task
> has passed exit_notify(), only do_wait-like code paths should use this.
>
> Unfortunately, we can't kill task_is_dead() right now, it has already
> found the bugy users in drivers/staging/, the fix already exists.

I would say task_is_dead() has already acquired buggy users in
drivers/staging.

As for the patch itself, and the direction of removing task_is_dead().
It looks good from where I sit.

Reviewed-by: "Eric W. Biederman" <ebiederm@xmission.com>

Eric


> Signed-off-by: Oleg Nesterov <oleg@redhat.com>
> ---
>  include/linux/cred.h |   10 +++-------
>  kernel/exit.c        |    2 +-
>  2 files changed, 4 insertions(+), 8 deletions(-)
>
> diff --git a/include/linux/cred.h b/include/linux/cred.h
> index adadf71..1b64c72 100644
> --- a/include/linux/cred.h
> +++ b/include/linux/cred.h
> @@ -276,17 +276,13 @@ static inline void put_cred(const struct cred *_cred)
>   * @task: The task to query
>   *
>   * Access the objective credentials of a task.  The caller must hold the RCU
> - * readlock or the task must be dead and unable to change its own credentials.
> + * readlock.
>   *
>   * The result of this function should not be passed directly to get_cred();
>   * rather get_task_cred() should be used instead.
>   */
> -#define __task_cred(task)						\
> -	({								\
> -		const struct task_struct *__t = (task);			\
> -		rcu_dereference_check(__t->real_cred,			\
> -				      task_is_dead(__t));		\
> -	})
> +#define __task_cred(task)	\
> +	rcu_dereference((task)->real_cred)
>  
>  /**
>   * get_current_cred - Get the current task's subjective credentials
> diff --git a/kernel/exit.c b/kernel/exit.c
> index 4b4042f..7b36288 100644
> --- a/kernel/exit.c
> +++ b/kernel/exit.c
> @@ -1214,7 +1213,7 @@ static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
>  	unsigned long state;
>  	int retval, status, traced;
>  	pid_t pid = task_pid_vnr(p);
> -	uid_t uid = __task_cred(p)->uid;
> +	uid_t uid = task_uid(p);
>  	struct siginfo __user *infop;
>  
>  	if (!likely(wo->wo_flags & WEXITED))

Re: [PATCH 1/1] creds: kill __task_cred()->task_is_dead() validation

[Oleg Nesterov]

On 04/11, Eric W. Biederman wrote:
>
> Oleg Nesterov <oleg@redhat.com> writes:
>
> A small nit.  The subject should be:
> "Remove task_is_dead from __task_cred() validation."
> there is no method __task_cred()->task_is_dead().

I often use this notation to show the caller and the callee,
but I don't really mind.

> > Unfortunately, we can't kill task_is_dead() right now, it has already
> > found the bugy users in drivers/staging/, the fix already exists.
>
> I would say task_is_dead() has already acquired buggy users in
> drivers/staging.

Argh, it least I shouldn't have said "bugy".

> As for the patch itself, and the direction of removing task_is_dead().
> It looks good from where I sit.
>
> Reviewed-by: "Eric W. Biederman" <ebiederm@xmission.com>

Thanks! I'll fix the subject/changelog and resend with your and
David's acks.

Oleg.

Re: [PATCH 1/1] creds: kill __task_cred()->task_is_dead() validation

[David Howells]

Oleg Nesterov <oleg@redhat.com> wrote:

> commit 8f92054e:
> 
>     add the following validation condition:
> 
>         task->exit_state >= 0
> 
>     to permit the access if the target task is dead and therefore
>     unable to change its own credentials.
> 
> OK, but afaics currently this can only help wait_task_zombie() which
> calls __task_cred() without rcu lock.
> 
> Remove this validation and change wait_task_zombie() to use task_uid()
> instead. This means we do rcu_read_lock() only to shut up the lockdep,
> but we already do the same in, say, wait_task_stopped().
> 
> task_is_dead() should die, task->exit_state != 0 means that this task
> has passed exit_notify(), only do_wait-like code paths should use this.
> 
> Unfortunately, we can't kill task_is_dead() right now, it has already
> found the bugy users in drivers/staging/, the fix already exists.
>
> Signed-off-by: Oleg Nesterov <oleg@redhat.com>

Acked-by: David Howells <dhowells@redhat.com>