Re: Adding a security parameter to VFS functions

[Casey Schaufler <casey@schaufler-ca.com>]

--- David Howells <dhowells@redhat.com> wrote:

> 
> Hi Linus, Al,
> 
> Would you object greatly to functions like vfs_mkdir() gaining a security
> parameter?

Could you describe how this compares to the proposal that the
AppArmor developers suggested recently? I expect that we can 
reduce the amount of discussion required, and maybe avoid some
confusion if you could do that.

Thank you.

> What I'm thinking of is this:
> 
> 	int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode,
> 		      struct security *security)
> 
> Where the security context is the state of the context at the time the call
> was issued:
> 
> 	struct security {
> 		uid_t			fsuid;
> 		git_t			fsgid;
> 		struct group_info	*group_info;
> 		void			*security;
> 		struct key		*session_keyring;
> 		struct key		*process_keyring;
> 		struct key		*thread_keyring;
> 
> And perhaps:
> 
> 		struct audit_context	*audit_context;
> 		seccomp_t		seccomp;
> 	};
> 
> This would, for the most part, be a temporary affair, being set up by such as
> sys_mkdir()/sys_mkdirat() from data held in task_struct.
> 
> This information would then be passed into the filesystem and LSM layers so
> that files, directories, etc. can be created, opened, deleted, or otherwise
> mangled based on these security items, rather than the one in whichever
> task_struct is current.
> 
> 
> The reason for doing this would be to support an act-as interface, so that
> services such as nfsd and cachefiles could act with different security
> details
> to the ones attached to the task.  This would have a couple of potential
> benefits:
> 
>  (1) nfsd threads don't have to keep changing their security contexts.
> 
>  (2) cachefiles can act on behalf of a process without changing its security
>      context.
> 
> 
> Note that I/O operations such as read, write and ioctl would *not* be passed
> this data as the file struct should contain the relevant security
> information.
> Similarly, page I/O operations would also not need alteration as the VMA
> covering the region points to a file struct, which holds the appropriate
> security.
> 
> David
> 
> 
> 


Casey Schaufler
casey@schaufler-ca.com