[PATCH -next] fork.c: Move check of clone NEWIPC and SYSVSEM to copy_process

[PATCH -next] fork.c: Move check of clone NEWIPC and SYSVSEM to copy_process

[Marcos Paulo de Souza]

Currently this check for CLONE_NEWIPC with CLONE_SYSVSEM is done inside
copy_namespaces, resulting in a handful of error paths being executed if
these flags were used together. So, move this check to the beginning of
copy_process, exiting earlier if the condition is true.

This move is safe because copy_namespaces is called just from
copy_process function.

Signed-off-by: Marcos Paulo de Souza <marcos.souza.org@gmail.com>
---
 kernel/fork.c    | 11 +++++++++++
 kernel/nsproxy.c | 11 -----------
 2 files changed, 11 insertions(+), 11 deletions(-)

diff --git a/kernel/fork.c b/kernel/fork.c
index 2113e252cb9d..691f9ba135fc 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1600,6 +1600,17 @@ static __latent_entropy struct task_struct *copy_process(
 		return ERR_PTR(-EINVAL);
 
 	/*
+	 * CLONE_NEWIPC must detach from the undolist: after switching
+	 * to a new ipc namespace, the semaphore arrays from the old
+	 * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
+	 * means share undolist with parent, so we must forbid using
+	 * it along with CLONE_NEWIPC.
+	 */
+	if ((clone_flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
+		(CLONE_NEWIPC | CLONE_SYSVSEM))
+		return ERR_PTR(-EINVAL);
+
+	/*
 	 * Thread groups must share signals as well, and detached threads
 	 * can only be started up within the thread group.
 	 */
diff --git a/kernel/nsproxy.c b/kernel/nsproxy.c
index f6c5d330059a..30882727dff5 100644
--- a/kernel/nsproxy.c
+++ b/kernel/nsproxy.c
@@ -151,17 +151,6 @@ int copy_namespaces(unsigned long flags, struct task_struct *tsk)
 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
 		return -EPERM;
 
-	/*
-	 * CLONE_NEWIPC must detach from the undolist: after switching
-	 * to a new ipc namespace, the semaphore arrays from the old
-	 * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
-	 * means share undolist with parent, so we must forbid using
-	 * it along with CLONE_NEWIPC.
-	 */
-	if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
-		(CLONE_NEWIPC | CLONE_SYSVSEM)) 
-		return -EINVAL;
-
 	new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs);
 	if (IS_ERR(new_ns))
 		return  PTR_ERR(new_ns);
-- 
2.13.6

Re: [PATCH -next] fork.c: Move check of clone NEWIPC and SYSVSEM to copy_process

[Michal Hocko]

CC Eric

On Sun 26-11-17 14:06:52, Marcos Paulo de Souza wrote:
> Currently this check for CLONE_NEWIPC with CLONE_SYSVSEM is done inside
> copy_namespaces, resulting in a handful of error paths being executed if
> these flags were used together. So, move this check to the beginning of
> copy_process, exiting earlier if the condition is true.
> 
> This move is safe because copy_namespaces is called just from
> copy_process function.

I am not familiar with the code all that much but the justification is
not clear to me. Thesea re namespace related flags so why should we pull
them out of copy_namespaces. I do not see any simplifications in the
error code paths or something like that.

> Signed-off-by: Marcos Paulo de Souza <marcos.souza.org@gmail.com>
> ---
>  kernel/fork.c    | 11 +++++++++++
>  kernel/nsproxy.c | 11 -----------
>  2 files changed, 11 insertions(+), 11 deletions(-)
> 
> diff --git a/kernel/fork.c b/kernel/fork.c
> index 2113e252cb9d..691f9ba135fc 100644
> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -1600,6 +1600,17 @@ static __latent_entropy struct task_struct *copy_process(
>  		return ERR_PTR(-EINVAL);
>  
>  	/*
> +	 * CLONE_NEWIPC must detach from the undolist: after switching
> +	 * to a new ipc namespace, the semaphore arrays from the old
> +	 * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
> +	 * means share undolist with parent, so we must forbid using
> +	 * it along with CLONE_NEWIPC.
> +	 */
> +	if ((clone_flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
> +		(CLONE_NEWIPC | CLONE_SYSVSEM))
> +		return ERR_PTR(-EINVAL);
> +
> +	/*
>  	 * Thread groups must share signals as well, and detached threads
>  	 * can only be started up within the thread group.
>  	 */
> diff --git a/kernel/nsproxy.c b/kernel/nsproxy.c
> index f6c5d330059a..30882727dff5 100644
> --- a/kernel/nsproxy.c
> +++ b/kernel/nsproxy.c
> @@ -151,17 +151,6 @@ int copy_namespaces(unsigned long flags, struct task_struct *tsk)
>  	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
>  		return -EPERM;
>  
> -	/*
> -	 * CLONE_NEWIPC must detach from the undolist: after switching
> -	 * to a new ipc namespace, the semaphore arrays from the old
> -	 * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
> -	 * means share undolist with parent, so we must forbid using
> -	 * it along with CLONE_NEWIPC.
> -	 */
> -	if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
> -		(CLONE_NEWIPC | CLONE_SYSVSEM)) 
> -		return -EINVAL;
> -
>  	new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs);
>  	if (IS_ERR(new_ns))
>  		return  PTR_ERR(new_ns);
> -- 
> 2.13.6
> 

-- 
Michal Hocko
SUSE Labs

Re: [PATCH -next] fork.c: Move check of clone NEWIPC and SYSVSEM to copy_process

[Marcos Paulo de Souza]

On Thu, Nov 30, 2017 at 11:04:06AM +0100, Michal Hocko wrote:
> CC Eric
> 
> On Sun 26-11-17 14:06:52, Marcos Paulo de Souza wrote:
> > Currently this check for CLONE_NEWIPC with CLONE_SYSVSEM is done inside
> > copy_namespaces, resulting in a handful of error paths being executed if
> > these flags were used together. So, move this check to the beginning of
> > copy_process, exiting earlier if the condition is true.
> > 
> > This move is safe because copy_namespaces is called just from
> > copy_process function.

This change is introduced right below the point where clone_flags is already
checking for inconsistencies in namespace flags[1], and returns EINVAL when
conflicting flags are informed together.

In this case, it's easier to return early when conflicting flags are informed at
the beginning, so moving a namespace check to where namespaces are already being
sanitized makes sense. If the code stays where it is now, and a user calls clone
syscalls informing CLONE_NEWIPC | CLONE_SYSVSEM, the code will need to undo a
lot of work before returning the same EINVAL[2].

[1] https://elixir.free-electrons.com/linux/latest/source/kernel/fork.c#L1552
[2] https://elixir.free-electrons.com/linux/latest/source/kernel/fork.c#L1953

> 
> I am not familiar with the code all that much but the justification is
> not clear to me. Thesea re namespace related flags so why should we pull
> them out of copy_namespaces. I do not see any simplifications in the
> error code paths or something like that.
> 
> > Signed-off-by: Marcos Paulo de Souza <marcos.souza.org@gmail.com>
> > ---
> >  kernel/fork.c    | 11 +++++++++++
> >  kernel/nsproxy.c | 11 -----------
> >  2 files changed, 11 insertions(+), 11 deletions(-)
> > 
> > diff --git a/kernel/fork.c b/kernel/fork.c
> > index 2113e252cb9d..691f9ba135fc 100644
> > --- a/kernel/fork.c
> > +++ b/kernel/fork.c
> > @@ -1600,6 +1600,17 @@ static __latent_entropy struct task_struct *copy_process(
> >  		return ERR_PTR(-EINVAL);
> >  
> >  	/*
> > +	 * CLONE_NEWIPC must detach from the undolist: after switching
> > +	 * to a new ipc namespace, the semaphore arrays from the old
> > +	 * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
> > +	 * means share undolist with parent, so we must forbid using
> > +	 * it along with CLONE_NEWIPC.
> > +	 */
> > +	if ((clone_flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
> > +		(CLONE_NEWIPC | CLONE_SYSVSEM))
> > +		return ERR_PTR(-EINVAL);
> > +
> > +	/*
> >  	 * Thread groups must share signals as well, and detached threads
> >  	 * can only be started up within the thread group.
> >  	 */
> > diff --git a/kernel/nsproxy.c b/kernel/nsproxy.c
> > index f6c5d330059a..30882727dff5 100644
> > --- a/kernel/nsproxy.c
> > +++ b/kernel/nsproxy.c
> > @@ -151,17 +151,6 @@ int copy_namespaces(unsigned long flags, struct task_struct *tsk)
> >  	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
> >  		return -EPERM;
> >  
> > -	/*
> > -	 * CLONE_NEWIPC must detach from the undolist: after switching
> > -	 * to a new ipc namespace, the semaphore arrays from the old
> > -	 * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
> > -	 * means share undolist with parent, so we must forbid using
> > -	 * it along with CLONE_NEWIPC.
> > -	 */
> > -	if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
> > -		(CLONE_NEWIPC | CLONE_SYSVSEM)) 
> > -		return -EINVAL;
> > -
> >  	new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs);
> >  	if (IS_ERR(new_ns))
> >  		return  PTR_ERR(new_ns);
> > -- 
> > 2.13.6
> > 
> 
> -- 
> Michal Hocko
> SUSE Labs

-- 
Thanks,
	Marcos

Re: [PATCH -next] fork.c: Move check of clone NEWIPC and SYSVSEM to copy_process

[Michal Hocko]

On Wed 29-11-17 22:33:43, Marcos Paulo de Souza wrote:
> On Thu, Nov 30, 2017 at 11:04:06AM +0100, Michal Hocko wrote:
> > CC Eric
> > 
> > On Sun 26-11-17 14:06:52, Marcos Paulo de Souza wrote:
> > > Currently this check for CLONE_NEWIPC with CLONE_SYSVSEM is done inside
> > > copy_namespaces, resulting in a handful of error paths being executed if
> > > these flags were used together. So, move this check to the beginning of
> > > copy_process, exiting earlier if the condition is true.
> > > 
> > > This move is safe because copy_namespaces is called just from
> > > copy_process function.
> 
> This change is introduced right below the point where clone_flags is already
> checking for inconsistencies in namespace flags[1], and returns EINVAL when
> conflicting flags are informed together.
> 
> In this case, it's easier to return early when conflicting flags are informed at
> the beginning, so moving a namespace check to where namespaces are already being
> sanitized makes sense.

What doe easier mean in this context?

> If the code stays where it is now, and a user calls clone
> syscalls informing CLONE_NEWIPC | CLONE_SYSVSEM, the code will need to undo a
> lot of work before returning the same EINVAL[2].

And why do we care about that error path? Does it trigger that often?

Do not take me wrong. I am not saying the patch is incorrect or wrong.
The changelog lacks the justification and explanation _why_ do we care
at all. A rarely failing path reorganization doesn't sound too
interesting to me. I cannot judge the cleanup aspect of this though but
from a very naive diffstat POV it doesn't look like a huge win to me.
 
> [1] https://elixir.free-electrons.com/linux/latest/source/kernel/fork.c#L1552
> [2] https://elixir.free-electrons.com/linux/latest/source/kernel/fork.c#L1953
> 
> > 
> > I am not familiar with the code all that much but the justification is
> > not clear to me. Thesea re namespace related flags so why should we pull
> > them out of copy_namespaces. I do not see any simplifications in the
> > error code paths or something like that.
> > 
> > > Signed-off-by: Marcos Paulo de Souza <marcos.souza.org@gmail.com>
> > > ---
> > >  kernel/fork.c    | 11 +++++++++++
> > >  kernel/nsproxy.c | 11 -----------
> > >  2 files changed, 11 insertions(+), 11 deletions(-)
> > > 
> > > diff --git a/kernel/fork.c b/kernel/fork.c
> > > index 2113e252cb9d..691f9ba135fc 100644
> > > --- a/kernel/fork.c
> > > +++ b/kernel/fork.c
> > > @@ -1600,6 +1600,17 @@ static __latent_entropy struct task_struct *copy_process(
> > >  		return ERR_PTR(-EINVAL);
> > >  
> > >  	/*
> > > +	 * CLONE_NEWIPC must detach from the undolist: after switching
> > > +	 * to a new ipc namespace, the semaphore arrays from the old
> > > +	 * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
> > > +	 * means share undolist with parent, so we must forbid using
> > > +	 * it along with CLONE_NEWIPC.
> > > +	 */
> > > +	if ((clone_flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
> > > +		(CLONE_NEWIPC | CLONE_SYSVSEM))
> > > +		return ERR_PTR(-EINVAL);
> > > +
> > > +	/*
> > >  	 * Thread groups must share signals as well, and detached threads
> > >  	 * can only be started up within the thread group.
> > >  	 */
> > > diff --git a/kernel/nsproxy.c b/kernel/nsproxy.c
> > > index f6c5d330059a..30882727dff5 100644
> > > --- a/kernel/nsproxy.c
> > > +++ b/kernel/nsproxy.c
> > > @@ -151,17 +151,6 @@ int copy_namespaces(unsigned long flags, struct task_struct *tsk)
> > >  	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
> > >  		return -EPERM;
> > >  
> > > -	/*
> > > -	 * CLONE_NEWIPC must detach from the undolist: after switching
> > > -	 * to a new ipc namespace, the semaphore arrays from the old
> > > -	 * namespace are unreachable.  In clone parlance, CLONE_SYSVSEM
> > > -	 * means share undolist with parent, so we must forbid using
> > > -	 * it along with CLONE_NEWIPC.
> > > -	 */
> > > -	if ((flags & (CLONE_NEWIPC | CLONE_SYSVSEM)) ==
> > > -		(CLONE_NEWIPC | CLONE_SYSVSEM)) 
> > > -		return -EINVAL;
> > > -
> > >  	new_ns = create_new_namespaces(flags, tsk, user_ns, tsk->fs);
> > >  	if (IS_ERR(new_ns))
> > >  		return  PTR_ERR(new_ns);
> > > -- 
> > > 2.13.6
> > > 
> > 
> > -- 
> > Michal Hocko
> > SUSE Labs
> 
> -- 
> Thanks,
> 	Marcos

-- 
Michal Hocko
SUSE Labs