[v2] LSM: Replace secctx/len pairs with lsm_context

[PATCH v2 5/6] LSM: secctx provider check on release

Posted by Casey Schaufler 1 month, 1 week ago

Verify that the LSM releasing the secctx is the LSM that
allocated it. This was not necessary when only one LSM could
create a secctx, but once there can be more than one it is.

Signed-off-by: Casey Schaufler <casey@schaufler-ca.com>
---
 security/apparmor/secid.c | 10 ++--------
 security/selinux/hooks.c  | 10 ++--------
 2 files changed, 4 insertions(+), 16 deletions(-)

diff --git a/security/apparmor/secid.c b/security/apparmor/secid.c
index 5d92fc3ab8b4..974f802cbe5a 100644
--- a/security/apparmor/secid.c
+++ b/security/apparmor/secid.c
@@ -122,14 +122,8 @@ int apparmor_secctx_to_secid(const char *secdata, u32 seclen, u32 *secid)
 
 void apparmor_release_secctx(struct lsm_context *cp)
 {
-	/*
-	 * stacking scaffolding:
-	 * When it is possible for more than one LSM to provide a
-	 * release hook, do this check:
-	 * if (cp->id == LSM_ID_APPARMOR || cp->id == LSM_ID_UNDEF)
-	 */
-
-	kfree(cp->context);
+	if (cp->id == LSM_ID_APPARMOR)
+		kfree(cp->context);
 }
 
 /**
diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
index 79776a5e651d..b9286c2c5efe 100644
--- a/security/selinux/hooks.c
+++ b/security/selinux/hooks.c
@@ -6640,14 +6640,8 @@ static int selinux_secctx_to_secid(const char *secdata, u32 seclen, u32 *secid)
 
 static void selinux_release_secctx(struct lsm_context *cp)
 {
-	/*
-	 * stacking scaffolding:
-	 * When it is possible for more than one LSM to provide a
-	 * release hook, do this check:
-	 * if (cp->id == LSM_ID_SELINUX || cp->id == LSM_ID_UNDEF)
-	 */
-
-	kfree(cp->context);
+	if (cp->id == LSM_ID_SELINUX)
+		kfree(cp->context);
 }
 
 static void selinux_inode_invalidate_secctx(struct inode *inode)
-- 
2.46.0

Re: [PATCH v2 5/6] LSM: secctx provider check on release

Posted by Paul Moore 1 month ago

On Oct 14, 2024 Casey Schaufler <casey@schaufler-ca.com> wrote:
> 
> Verify that the LSM releasing the secctx is the LSM that
> allocated it. This was not necessary when only one LSM could
> create a secctx, but once there can be more than one it is.
> 
> Signed-off-by: Casey Schaufler <casey@schaufler-ca.com>
> ---
>  security/apparmor/secid.c | 10 ++--------
>  security/selinux/hooks.c  | 10 ++--------
>  2 files changed, 4 insertions(+), 16 deletions(-)
> 
> diff --git a/security/apparmor/secid.c b/security/apparmor/secid.c
> index 5d92fc3ab8b4..974f802cbe5a 100644
> --- a/security/apparmor/secid.c
> +++ b/security/apparmor/secid.c
> @@ -122,14 +122,8 @@ int apparmor_secctx_to_secid(const char *secdata, u32 seclen, u32 *secid)
>  
>  void apparmor_release_secctx(struct lsm_context *cp)
>  {
> -	/*
> -	 * stacking scaffolding:
> -	 * When it is possible for more than one LSM to provide a
> -	 * release hook, do this check:
> -	 * if (cp->id == LSM_ID_APPARMOR || cp->id == LSM_ID_UNDEF)
> -	 */
> -
> -	kfree(cp->context);
> +	if (cp->id == LSM_ID_APPARMOR)
> +		kfree(cp->context);

Should we set cp->context to NULL too?  One could argue that it's an
unecessary assignment, given the cp->id checks, and they wouldn't be
wrong, but considering the potential for a BPF LSM to do things with
a lsm_context, I wonder if resetting the pointer to NULL is the
smart thing to do.

This obviously applies to the SELinux code (below) too.

>  }
>  
>  /**
> diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
> index 79776a5e651d..b9286c2c5efe 100644
> --- a/security/selinux/hooks.c
> +++ b/security/selinux/hooks.c
> @@ -6640,14 +6640,8 @@ static int selinux_secctx_to_secid(const char *secdata, u32 seclen, u32 *secid)
>  
>  static void selinux_release_secctx(struct lsm_context *cp)
>  {
> -	/*
> -	 * stacking scaffolding:
> -	 * When it is possible for more than one LSM to provide a
> -	 * release hook, do this check:
> -	 * if (cp->id == LSM_ID_SELINUX || cp->id == LSM_ID_UNDEF)
> -	 */
> -
> -	kfree(cp->context);
> +	if (cp->id == LSM_ID_SELINUX)
> +		kfree(cp->context);
>  }
>  
>  static void selinux_inode_invalidate_secctx(struct inode *inode)

--
paul-moore.com

Re: [PATCH v2 5/6] LSM: secctx provider check on release

Posted by Casey Schaufler 1 month ago

On 10/21/2024 4:39 PM, Paul Moore wrote:
> On Oct 14, 2024 Casey Schaufler <casey@schaufler-ca.com> wrote:
>> Verify that the LSM releasing the secctx is the LSM that
>> allocated it. This was not necessary when only one LSM could
>> create a secctx, but once there can be more than one it is.
>>
>> Signed-off-by: Casey Schaufler <casey@schaufler-ca.com>
>> ---
>>  security/apparmor/secid.c | 10 ++--------
>>  security/selinux/hooks.c  | 10 ++--------
>>  2 files changed, 4 insertions(+), 16 deletions(-)
>>
>> diff --git a/security/apparmor/secid.c b/security/apparmor/secid.c
>> index 5d92fc3ab8b4..974f802cbe5a 100644
>> --- a/security/apparmor/secid.c
>> +++ b/security/apparmor/secid.c
>> @@ -122,14 +122,8 @@ int apparmor_secctx_to_secid(const char *secdata, u32 seclen, u32 *secid)
>>  
>>  void apparmor_release_secctx(struct lsm_context *cp)
>>  {
>> -	/*
>> -	 * stacking scaffolding:
>> -	 * When it is possible for more than one LSM to provide a
>> -	 * release hook, do this check:
>> -	 * if (cp->id == LSM_ID_APPARMOR || cp->id == LSM_ID_UNDEF)
>> -	 */
>> -
>> -	kfree(cp->context);
>> +	if (cp->id == LSM_ID_APPARMOR)
>> +		kfree(cp->context);
> Should we set cp->context to NULL too?  One could argue that it's an
> unecessary assignment, given the cp->id checks, and they wouldn't be
> wrong, but considering the potential for a BPF LSM to do things with
> a lsm_context, I wonder if resetting the pointer to NULL is the
> smart thing to do.

Wouldn't hurt. I'll go ahead and add that. If a BPF LSM does anything
with a lsm_context we're likely to hear about the many issues quite
quickly.

>
> This obviously applies to the SELinux code (below) too.
>
>>  }
>>  
>>  /**
>> diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c
>> index 79776a5e651d..b9286c2c5efe 100644
>> --- a/security/selinux/hooks.c
>> +++ b/security/selinux/hooks.c
>> @@ -6640,14 +6640,8 @@ static int selinux_secctx_to_secid(const char *secdata, u32 seclen, u32 *secid)
>>  
>>  static void selinux_release_secctx(struct lsm_context *cp)
>>  {
>> -	/*
>> -	 * stacking scaffolding:
>> -	 * When it is possible for more than one LSM to provide a
>> -	 * release hook, do this check:
>> -	 * if (cp->id == LSM_ID_SELINUX || cp->id == LSM_ID_UNDEF)
>> -	 */
>> -
>> -	kfree(cp->context);
>> +	if (cp->id == LSM_ID_SELINUX)
>> +		kfree(cp->context);
>>  }
>>  
>>  static void selinux_inode_invalidate_secctx(struct inode *inode)
> --
> paul-moore.com
>

Re: [PATCH v2 5/6] LSM: secctx provider check on release

Posted by Paul Moore 1 month ago

On Mon, Oct 21, 2024 at 8:06 PM Casey Schaufler <casey@schaufler-ca.com> wrote:
> On 10/21/2024 4:39 PM, Paul Moore wrote:
> > On Oct 14, 2024 Casey Schaufler <casey@schaufler-ca.com> wrote:
> >> Verify that the LSM releasing the secctx is the LSM that
> >> allocated it. This was not necessary when only one LSM could
> >> create a secctx, but once there can be more than one it is.
> >>
> >> Signed-off-by: Casey Schaufler <casey@schaufler-ca.com>
> >> ---
> >>  security/apparmor/secid.c | 10 ++--------
> >>  security/selinux/hooks.c  | 10 ++--------
> >>  2 files changed, 4 insertions(+), 16 deletions(-)
> >>
> >> diff --git a/security/apparmor/secid.c b/security/apparmor/secid.c
> >> index 5d92fc3ab8b4..974f802cbe5a 100644
> >> --- a/security/apparmor/secid.c
> >> +++ b/security/apparmor/secid.c
> >> @@ -122,14 +122,8 @@ int apparmor_secctx_to_secid(const char *secdata, u32 seclen, u32 *secid)
> >>
> >>  void apparmor_release_secctx(struct lsm_context *cp)
> >>  {
> >> -    /*
> >> -     * stacking scaffolding:
> >> -     * When it is possible for more than one LSM to provide a
> >> -     * release hook, do this check:
> >> -     * if (cp->id == LSM_ID_APPARMOR || cp->id == LSM_ID_UNDEF)
> >> -     */
> >> -
> >> -    kfree(cp->context);
> >> +    if (cp->id == LSM_ID_APPARMOR)
> >> +            kfree(cp->context);
> > Should we set cp->context to NULL too?  One could argue that it's an
> > unecessary assignment, given the cp->id checks, and they wouldn't be
> > wrong, but considering the potential for a BPF LSM to do things with
> > a lsm_context, I wonder if resetting the pointer to NULL is the
> > smart thing to do.
>
> Wouldn't hurt. I'll go ahead and add that. If a BPF LSM does anything
> with a lsm_context we're likely to hear about the many issues quite
> quickly.

Yes, I suspect you're right about that, at least we can protect
against a UAF in this one case :)

-- 
paul-moore.com