[PATCH] KEYS: fix refcount_inc() on zero

[PATCH] KEYS: fix refcount_inc() on zero

[Mark Rutland]

If a key's refcount is dropped to zero between key_lookup() peeking at
the refcount and subsequently attempting to increment it, refcount_inc()
will see a zero refcount. Here, refcount_inc() will WARN_ONCE(), and
will *not* increment the refcount, which will remain zero.

Once key_lookup() drops key_serial_lock, it is possible for the key to
be freed behind our back.

This patch uses refcount_inc_not_zero() to perform the peek and
increment atomically. A helper with lockdep annotation is added to
document why this is safe.

Fixes: fff292914d3a2f1e ("security, keys: convert key.usage from atomic_t to refcount_t")
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: David Howells <dhowells@redhat.com>
Cc: David Windsor <dwindsor@gmail.com>
Cc: Elena Reshetova <elena.reshetova@intel.com>
Cc: Hans Liljestrand <ishkamiel@gmail.com>
Cc: James Morris <james.l.morris@oracle.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Peter Zijlstra <peterz@infradead.org>
---
 security/keys/key.c | 18 +++++++++++-------
 1 file changed, 11 insertions(+), 7 deletions(-)

diff --git a/security/keys/key.c b/security/keys/key.c
index 455c04d..150f51d 100644
--- a/security/keys/key.c
+++ b/security/keys/key.c
@@ -632,6 +632,12 @@ void key_put(struct key *key)
 }
 EXPORT_SYMBOL(key_put);
 
+static bool key_get_not_free(struct key *key)
+{
+	lockdep_assert_held(&key_serial_lock);
+	return refcount_inc_not_zero(&key->usage);
+}
+
 /*
  * Find a key by its serial number.
  */
@@ -660,14 +666,12 @@ struct key *key_lookup(key_serial_t id)
 	goto error;
 
 found:
-	/* pretend it doesn't exist if it is awaiting deletion */
-	if (refcount_read(&key->usage) == 0)
-		goto not_found;
-
-	/* this races with key_put(), but that doesn't matter since key_put()
-	 * doesn't actually change the key
+	/*
+	 * Pretend it doesn't exist if it is awaiting deletion. This races with
+	 * key_put(), but we can peek at the key until we drop key_serial_lock.
 	 */
-	__key_get(key);
+	if (!key_get_not_free(key))
+		goto not_found;
 
 error:
 	spin_unlock(&key_serial_lock);
-- 
1.9.1

Re: [PATCH] KEYS: fix refcount_inc() on zero

[Kees Cook]

On Fri, May 26, 2017 at 10:37 AM, Mark Rutland <mark.rutland@arm.com> wrote:
> If a key's refcount is dropped to zero between key_lookup() peeking at
> the refcount and subsequently attempting to increment it, refcount_inc()
> will see a zero refcount. Here, refcount_inc() will WARN_ONCE(), and
> will *not* increment the refcount, which will remain zero.
>
> Once key_lookup() drops key_serial_lock, it is possible for the key to
> be freed behind our back.
>
> This patch uses refcount_inc_not_zero() to perform the peek and
> increment atomically. A helper with lockdep annotation is added to
> document why this is safe.
>
> Fixes: fff292914d3a2f1e ("security, keys: convert key.usage from atomic_t to refcount_t")
> Signed-off-by: Mark Rutland <mark.rutland@arm.com>
> Cc: David Howells <dhowells@redhat.com>
> Cc: David Windsor <dwindsor@gmail.com>
> Cc: Elena Reshetova <elena.reshetova@intel.com>
> Cc: Hans Liljestrand <ishkamiel@gmail.com>
> Cc: James Morris <james.l.morris@oracle.com>
> Cc: Kees Cook <keescook@chromium.org>
> Cc: Peter Zijlstra <peterz@infradead.org>

Thanks for catching this!

Reviewed-by: Kees Cook <keescook@chromium.org>

-Kees

> ---
>  security/keys/key.c | 18 +++++++++++-------
>  1 file changed, 11 insertions(+), 7 deletions(-)
>
> diff --git a/security/keys/key.c b/security/keys/key.c
> index 455c04d..150f51d 100644
> --- a/security/keys/key.c
> +++ b/security/keys/key.c
> @@ -632,6 +632,12 @@ void key_put(struct key *key)
>  }
>  EXPORT_SYMBOL(key_put);
>
> +static bool key_get_not_free(struct key *key)
> +{
> +       lockdep_assert_held(&key_serial_lock);
> +       return refcount_inc_not_zero(&key->usage);
> +}
> +
>  /*
>   * Find a key by its serial number.
>   */
> @@ -660,14 +666,12 @@ struct key *key_lookup(key_serial_t id)
>         goto error;
>
>  found:
> -       /* pretend it doesn't exist if it is awaiting deletion */
> -       if (refcount_read(&key->usage) == 0)
> -               goto not_found;
> -
> -       /* this races with key_put(), but that doesn't matter since key_put()
> -        * doesn't actually change the key
> +       /*
> +        * Pretend it doesn't exist if it is awaiting deletion. This races with
> +        * key_put(), but we can peek at the key until we drop key_serial_lock.
>          */
> -       __key_get(key);
> +       if (!key_get_not_free(key))
> +               goto not_found;
>
>  error:
>         spin_unlock(&key_serial_lock);
> --
> 1.9.1
>



-- 
Kees Cook
Pixel Security

Re: [PATCH] KEYS: fix refcount_inc() on zero

[David Howells]

Mark Rutland <mark.rutland@arm.com> wrote:

> This patch uses refcount_inc_not_zero() to perform the peek and
> increment atomically.

I think the helper is unnecessary.  Better to adjust the comment if you really
want to explain it.  Anyone editing the code should be that this is inside a
critical section.

> A helper with lockdep annotation is added to document why this is safe.

This doesn't explain why this is safe.

> +	/*
> +	 * Pretend it doesn't exist if it is awaiting deletion. This races with
> +	 * key_put(), but we can peek at the key until we drop key_serial_lock.
>  	 */

With your change, there is no race with key_put() - so the second sentence is
unnecessary.

I've adjusted your patch - see attached.

David
---
commit f66bf831c45306ebbc28aecd407e238983457251
Author: Mark Rutland <mark.rutland@arm.com>
Date:   Fri May 26 18:37:34 2017 +0100

    KEYS: fix refcount_inc() on zero
    
    If a key's refcount is dropped to zero between key_lookup() peeking at
    the refcount and subsequently attempting to increment it, refcount_inc()
    will see a zero refcount.  Here, refcount_inc() will WARN_ONCE(), and
    will *not* increment the refcount, which will remain zero.
    
    Once key_lookup() drops key_serial_lock, it is possible for the key to
    be freed behind our back.
    
    This patch uses refcount_inc_not_zero() to perform the peek and increment
    atomically.
    
    Fixes: fff292914d3a2f1e ("security, keys: convert key.usage from atomic_t to refcount_t")
    Signed-off-by: Mark Rutland <mark.rutland@arm.com>
    Signed-off-by: David Howells <dhowells@redhat.com>
    Cc: David Windsor <dwindsor@gmail.com>
    Cc: Elena Reshetova <elena.reshetova@intel.com>
    Cc: Hans Liljestrand <ishkamiel@gmail.com>
    Cc: James Morris <james.l.morris@oracle.com>
    Cc: Kees Cook <keescook@chromium.org>
    Cc: Peter Zijlstra <peterz@infradead.org>

diff --git a/security/keys/key.c b/security/keys/key.c
index 455c04d80bbb..d84ee2a87da6 100644
--- a/security/keys/key.c
+++ b/security/keys/key.c
@@ -660,14 +660,11 @@ struct key *key_lookup(key_serial_t id)
 	goto error;
 
 found:
-	/* pretend it doesn't exist if it is awaiting deletion */
-	if (refcount_read(&key->usage) == 0)
-		goto not_found;
-
-	/* this races with key_put(), but that doesn't matter since key_put()
-	 * doesn't actually change the key
+	/* A key is allowed to be looked up only if someone still owns a
+	 * reference to it - otherwise it's awaiting the gc.
 	 */
-	__key_get(key);
+	if (!refcount_inc_not_zero(&key->usage))
+		goto not_found;
 
 error:
 	spin_unlock(&key_serial_lock);

Re: [PATCH] KEYS: fix refcount_inc() on zero

[Mark Rutland]

On Wed, May 31, 2017 at 01:20:44PM +0100, David Howells wrote:
> Mark Rutland <mark.rutland@arm.com> wrote:
> 
> > This patch uses refcount_inc_not_zero() to perform the peek and
> > increment atomically.
> 
> I think the helper is unnecessary.  Better to adjust the comment if you really
> want to explain it.  Anyone editing the code should be that this is inside a
> critical section.
> 
> > A helper with lockdep annotation is added to document why this is safe.
> 
> This doesn't explain why this is safe.
> 
> > +	/*
> > +	 * Pretend it doesn't exist if it is awaiting deletion. This races with
> > +	 * key_put(), but we can peek at the key until we drop key_serial_lock.
> >  	 */
> 
> With your change, there is no race with key_put() - so the second sentence is
> unnecessary.

Fair enough, on all counts.

> I've adjusted your patch - see attached.

That looks fine to me, thanks!

Mark.

> 
> David
> ---
> commit f66bf831c45306ebbc28aecd407e238983457251
> Author: Mark Rutland <mark.rutland@arm.com>
> Date:   Fri May 26 18:37:34 2017 +0100
> 
>     KEYS: fix refcount_inc() on zero
>     
>     If a key's refcount is dropped to zero between key_lookup() peeking at
>     the refcount and subsequently attempting to increment it, refcount_inc()
>     will see a zero refcount.  Here, refcount_inc() will WARN_ONCE(), and
>     will *not* increment the refcount, which will remain zero.
>     
>     Once key_lookup() drops key_serial_lock, it is possible for the key to
>     be freed behind our back.
>     
>     This patch uses refcount_inc_not_zero() to perform the peek and increment
>     atomically.
>     
>     Fixes: fff292914d3a2f1e ("security, keys: convert key.usage from atomic_t to refcount_t")
>     Signed-off-by: Mark Rutland <mark.rutland@arm.com>
>     Signed-off-by: David Howells <dhowells@redhat.com>
>     Cc: David Windsor <dwindsor@gmail.com>
>     Cc: Elena Reshetova <elena.reshetova@intel.com>
>     Cc: Hans Liljestrand <ishkamiel@gmail.com>
>     Cc: James Morris <james.l.morris@oracle.com>
>     Cc: Kees Cook <keescook@chromium.org>
>     Cc: Peter Zijlstra <peterz@infradead.org>
> 
> diff --git a/security/keys/key.c b/security/keys/key.c
> index 455c04d80bbb..d84ee2a87da6 100644
> --- a/security/keys/key.c
> +++ b/security/keys/key.c
> @@ -660,14 +660,11 @@ struct key *key_lookup(key_serial_t id)
>  	goto error;
>  
>  found:
> -	/* pretend it doesn't exist if it is awaiting deletion */
> -	if (refcount_read(&key->usage) == 0)
> -		goto not_found;
> -
> -	/* this races with key_put(), but that doesn't matter since key_put()
> -	 * doesn't actually change the key
> +	/* A key is allowed to be looked up only if someone still owns a
> +	 * reference to it - otherwise it's awaiting the gc.
>  	 */
> -	__key_get(key);
> +	if (!refcount_inc_not_zero(&key->usage))
> +		goto not_found;
>  
>  error:
>  	spin_unlock(&key_serial_lock);