Re: [PATCH v5 13/21] evm: add evm_inode_post_init call in gfs2

[Mimi Zohar <zohar@linux.vnet.ibm.com>]

On Mon, 2011-05-16 at 12:25 -0700, Casey Schaufler wrote: 
> On 5/16/2011 11:48 AM, Mimi Zohar wrote:
> > On Mon, 2011-05-16 at 11:23 -0700, Casey Schaufler wrote:

> >> There is a very real possibility that multiple concurrent LSMs will
> >> be supported before too long. Smack already uses multiple attributes
> >> (SMACK64, SMACK64EXEC) on a file. Getting all the attributes in a
> >> single call could result in an interface that requires parsing a
> >> string argument, and we all know how popular those are. Introducing
> >> an interface that we know isn't going to accommodate this upcoming
> >> direction does not seem prudent.
> > I would think that Smack would benefit from Steven's suggestion of
> > returning an array of xattrs. Without his suggestion, I'm not sure how
> > you are, or planning on, initializing multiple xattrs from a single LSM,
> > unless of course you're not using security_inode_init_security().
> 
> The good news is that Smack has one required attribute. The others
> are for special purposes and will usually be absent. It is easy to
> imagine an LSM that always uses multiple attributes on a given file.
> 
> Yes, the array of xattr structures makes sense for any one LSM,
> but there still needs to be the potential for multiple calls for
> the multiple LSM case. I can't see that going away without a radical
> LSM restructuring.
> 
> > Multiple LSMs calling security_inode_init_security() will be an issue
> > for EVM, as EVM assumes there is a single LSM xattr on which to base the
> > initial hmac.
> 
> That is far from the biggest issue with multiple LSMs, but is definitely
> something to worry about.

Ok. After thinking about this a bit more, 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.

Mimi