[PATCH] security: Delay freeing inode->i_security till the end of RCU grace period

[Mimi Zohar]

On Tue, 2011-12-06 at 10:09 -0600, MAILER-DAEMON wrote:
> On Tue, 2011-12-06 at 10:14 -0500, Christoph Hellwig wrote:
> > On Mon, Dec 05, 2011 at 12:42:21PM -0600, Chandra Seetharaman wrote:
> > > while running test case 234 from xfstests test suite, I was getting an
> > > occational memory fault in inode_has_perm() with the following stack
> > 
> > Interesting.  Given that have no good way to free other data with the
> > normal inode callback it looks like we indeed need to do this
> > separately.
> > 
> > What about IMA or similar monsters?  Posix ACLs already are covered at
> > least.
> > 
> 
> Hi Christoph,
> 
> The problem is pretty much located in the function
> fs/inode.c:destroy_inode(), which calls __destroy_inode(), which does
> the freeing, and then does the call_rcu() on the inode.
> 
> I looked at all the functions in __destroy_inode() and found only
> security_inode_free() to be problematic. Others would handle the
> situation gracefully.
> 
> Sorry for the lack of knowledge. what is IMA ?
> 
> Chandra

security_inode_free() calls security/integrity/iint.c:
integrity_inode_free(), which frees the 'iint'.  For more information on
IMA, refer to linux-ima.sf.net.

thanks,

Mimi