Re: include/linux/cgroup.h:566 suspicious rcu_dereference_check() usage!

[Sergey Senozhatsky <sergey.senozhatsky@gmail.com>]

On (10/08/12 12:49), Paul E. McKenney wrote:
> 
> ------------------------------------------------------------------------
> 
> device_cgroup: Restore rcu_read_lock() protection to devcgroup_inode_mknod()
> 
> Commit ad676077 (device_cgroup: convert device_cgroup internally to
> policy + exceptions) restructured devcgroup_inode_mknod(), removing
> rcu_read_lock() in the process.  However, RCU read-side protection
> is required by the call to task_devcgroup(), so this commit restores
> the rcu_read_lock() and rcu_read_unlock().
> 
> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> 
> diff --git a/security/device_cgroup.c b/security/device_cgroup.c
> index 44dfc41..c686110 100644
> --- a/security/device_cgroup.c
> +++ b/security/device_cgroup.c
> @@ -576,9 +576,12 @@ int __devcgroup_inode_permission(struct inode *inode, int mask)
>  
>  int devcgroup_inode_mknod(int mode, dev_t dev)
>  {
> -	struct dev_cgroup *dev_cgroup = task_devcgroup(current);
> +	struct dev_cgroup *dev_cgroup;
> +	int ret;
>  	short type;
>  
> +	rcu_read_lock();
> +	dev_cgroup = task_devcgroup(current);
>  	if (!S_ISBLK(mode) && !S_ISCHR(mode))
>  		return 0;
>  
> @@ -587,7 +590,9 @@ int devcgroup_inode_mknod(int mode, dev_t dev)
>  	else
>  		type = DEV_CHAR;
>  
> -	return __devcgroup_check_permission(dev_cgroup, type, MAJOR(dev),
> +	ret =  __devcgroup_check_permission(dev_cgroup, type, MAJOR(dev),
>  					    MINOR(dev), ACC_MKNOD);
> +	rcu_read_unlock();
> +	return ret;
>  
>  }
> 


I believe the same should be done for __devcgroup_inode_permission() as well. And we
probably can call task_devcgroup() and rcu_read_lock() after "S_ISBLK(mode) && !S_ISCHR(mode)"
checks (I guess we also need to unlock RCU on `return 0').



----------------------------------------------------------------------------

Commit ad676077
 | Author: Aristeu Rozanski <aris@redhat.com>
 | Date:   Thu Oct 4 17:15:17 2012 -0700
 |   device_cgroup: convert device_cgroup internally to policy + exceptions

moved RCU read-side protection from devcgroup_inode_mknod(), which, however is required
by task_devcgroup(). Patch also add RCU read-side protection to __devcgroup_inode_permission()
function, introduced in commit ad676077.

[    0.946303] include/linux/cgroup.h:566 suspicious rcu_dereference_check() usage!
[    0.946511] 
[    0.946606] 2 locks held by kdevtmpfs/28:
[    0.946684]  #0:  (sb_writers){.+.+.+}, at: [<ffffffff81144bcb>] mnt_want_write+0x24/0x4b
[    0.947083]  #1:  (&sb->s_type->i_mutex_key#3/1){+.+.+.}, at: [<ffffffff81133d04>] kern_path_create+0x83/0x144
[    0.947598] 
[    0.947787] Call Trace:
[    0.947868]  [<ffffffff81089644>] lockdep_rcu_suspicious+0x109/0x112
[    0.947958]  [<ffffffff81258fa0>] devcgroup_inode_mknod+0x9e/0xee
[    0.948043]  [<ffffffff81132ee7>] vfs_mknod+0x8a/0xed
[    0.948129]  [<ffffffff813b98af>] handle_create.isra.2+0x144/0x1b5
[    0.948214]  [<ffffffff813b99bf>] ? devtmpfsd+0x9f/0x138
[    0.948298]  [<ffffffff81295d5c>] ? do_raw_spin_lock+0x67/0xde
[    0.948384]  [<ffffffff81295e92>] ? do_raw_spin_unlock+0x8f/0x98
[    0.948469]  [<ffffffff813b9920>] ? handle_create.isra.2+0x1b5/0x1b5
[    0.948554]  [<ffffffff813b9a04>] devtmpfsd+0xe4/0x138
[    0.948638]  [<ffffffff813b9920>] ? handle_create.isra.2+0x1b5/0x1b5
[    0.948724]  [<ffffffff810582b6>] kthread+0xd5/0xdd
[    0.948814]  [<ffffffff814db664>] kernel_thread_helper+0x4/0x10
[    0.948900]  [<ffffffff814d2973>] ? retint_restore_args+0x13/0x13
[    0.948985]  [<ffffffff810581e1>] ? __init_kthread_worker+0x5a/0x5a
[    0.949069]  [<ffffffff814db660>] ? gs_change+0x13/0x13


devcgroup_inode_mknod() part submitted by Paul E. McKenney.


Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>

---

 security/device_cgroup.c | 21 +++++++++++++++++----
 1 file changed, 17 insertions(+), 4 deletions(-)

diff --git a/security/device_cgroup.c b/security/device_cgroup.c
index 44dfc41..043eb00 100644
--- a/security/device_cgroup.c
+++ b/security/device_cgroup.c
@@ -558,7 +558,8 @@ static int __devcgroup_check_permission(struct dev_cgroup *dev_cgroup,
 
 int __devcgroup_inode_permission(struct inode *inode, int mask)
 {
-	struct dev_cgroup *dev_cgroup = task_devcgroup(current);
+	struct dev_cgroup *dev_cgroup;
+	int ret;
 	short type, access = 0;
 
 	if (S_ISBLK(inode->i_mode))
@@ -570,13 +571,20 @@ int __devcgroup_inode_permission(struct inode *inode, int mask)
 	if (mask & MAY_READ)
 		access |= ACC_READ;
 
-	return __devcgroup_check_permission(dev_cgroup, type, imajor(inode),
+	rcu_read_lock();
+
+	dev_cgroup = task_devcgroup(current);
+	ret = __devcgroup_check_permission(dev_cgroup, type, imajor(inode),
 					    iminor(inode), access);
+
+	rcu_read_unlock();
+	return ret;
 }
 
 int devcgroup_inode_mknod(int mode, dev_t dev)
 {
-	struct dev_cgroup *dev_cgroup = task_devcgroup(current);
+	struct dev_cgroup *dev_cgroup;
+	int ret;
 	short type;
 
 	if (!S_ISBLK(mode) && !S_ISCHR(mode))
@@ -587,7 +595,12 @@ int devcgroup_inode_mknod(int mode, dev_t dev)
 	else
 		type = DEV_CHAR;
 
-	return __devcgroup_check_permission(dev_cgroup, type, MAJOR(dev),
+	rcu_read_lock();
+
+	dev_cgroup = task_devcgroup(current);
+	ret = __devcgroup_check_permission(dev_cgroup, type, MAJOR(dev),
 					    MINOR(dev), ACC_MKNOD);
 
+	rcu_read_unlock();
+	return ret; 
 }