Re: Inodes hooks example

Previous message: Crispin Cowan: "Re: Low-cost hooks, multiple modules, per-task data"
In reply to: Greg KH: "Inodes hooks example"
Next in thread: Serge E. Hallyn: "Re: Inodes hooks example"
Next in thread: Andrew Morgan: "Re: Inodes hooks example"
Reply: Serge E. Hallyn: "Re: Inodes hooks example"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

hallynat_private

Quoting Greg KH (gregat_private):
> I'm looking for feedback on this:
> 	- do these hooks look sufficient for what everyone wants to do
> 	  with regards to inode operations?
> 	- have I placed the hooks in good places? (I'm not happy about
> 	  with where the setattr() hook is, I really don't want any of
> 	  these to be called within any kernel lock.  This one violates
> 	  that.)
> 	- have I done anything real dumb with these hooks?
> 	- anything else people want to comment on.

Hey guys,

I've been sitting on the sidelines watching so far.  However, I do
have some particular needs for hooks for DTE.  I need to assign types
to files based on absolute pathname.  For the moment i will ignore the
fact that with binding and multiple mounts I require some other
changes in order to even get a single, system-wide pathname from a
dentry/vfsmnt pair, as I'm not certain it's something for us to worry
about in this context (though it probably is).

However, I believe that any module which will label an inode based
upon pathnames will wish to add the labels at the first open, or first
create.  I'm working on 2.4.0-test11 (sorry), and I'm attaching the
diff for my namei.c.  In each place where dte_set_types is called, I
would love to see a
security_ops->inode_ops->set_pathbased_label(dentry,vfsmnt) hook.

My kernel panics if a type isn't assigned by the time I get to
permission(), so I'm reasonably certain I've caught the important
spots.

-serge