On Fri, 2011-06-03 at 12:21 +1000, Dave Chinner wrote: > On Thu, Jun 02, 2011 at 08:23:31AM -0400, Mimi Zohar wrote: > > Initialize 'security.evm' for new files. Reduce number of arguments > > by defining 'struct xattr'. > > why does this need a new security callout from every filesystem? > Once the security xattr is initialised, the name, len and value is > not going to change so surely the evm xattr can be initialised at > the same time the lsm xattr is initialised. Steve Whitehouse asked a similar question, suggesting that security_inode_init_security() return a vector of xattrs to minimize the number of xattr writes. Casey pointed out the "stacking" of LSMs will result in multiple calls to security_inode_init_security(), once for each LSM. The conclusion (http://lkml.org/lkml/2011/5/19/125) was: Moving evm_inode_init_security() into security_inode_init_security() only works for the single LSM and EVM case, but not for the multiple LSMs and EVM case, as the 'stacker' would call each LSM's security_inode_iint_security(). Having the 'stacker' return an array of xattrs would make sense and, at the same time, resolve the EVM issue. In evm_inode_post_init_security(), EVM could then walk the list of xattrs. > Then all you need to do in each filesystem is add the evm_xattr > structure to the existing security init call and a: > > #ifdef CONFIG_EVM > /* set evm.xattr */ > #endif > > to avoid adding code that is never executed when EVM is not > configured into the kernel. > > That way you don't create the lsm_xattr at all if the evm_xattr is > not created, and then the file creation should fail in an atomic > manner, right? i.e. you don't leave files with unverified security > attributes around when interesting failure corner cases occur (e.g. > ENOSPC). That would imply EVM must be enabled for all LSMs that define a security xattr. That's definitely a good goal, but probably not a good idea for right now. > And while you are there, it's probably also be a good idea to add > support for all filesystems that support xattrs, not just a random > subset of them... > > Cheers, > > Dave. The EVM xattr is initialized based on the LSM xattr. At this point, as far as I'm aware, the only remaining filesystems that call security_inode_init_security() to initialize the LSM xattr, are ocfs2 and reiserfs. Both of which might have memory leaks. Tiger Yang is addressing the memory leak for ocfs2. thanks, Mimi -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html