New SMB2/3 features and SMB_VFS_* and connection_struct...

[simo]

On Tue, 2012-05-29 at 15:56 +0200, Stefan (metze) Metzmacher wrote: 
> Am 29.05.2012 15:21, schrieb simo:
> > On Tue, 2012-05-29 at 14:23 +0200, Stefan (metze) Metzmacher wrote: 
> >> Hi Simo,
> >>
> >>>> We should also impersonate more correctly, so that all operations
> >>>> on a file handle run as the same user, including TCP disconnects.
> >>>> To do this sanely we need to have an tevent_context wrapper,
> >>>> which impersonates before calling any event handler.
> >>>> See
> >>>> https://gitweb.samba.org/?p=metze/samba/wip.git;a=shortlog;h=refs/heads/master3-impersonate
> >>>>
> >>>> Comments, please:-)
> >>>
> >>> I see how you become the desired user, but there is no way to go 'back'.
> >>> This seem to imply you cannot mix tevent_impersonate with any other
> >>> tevent call, as the process will change user and will not change it back
> >>> once done.
> >>> Where is the trick/technical detail I am missing ?
> >>
> >> Currently 'smbd' also changes the user only if needed.
> >> It doesn't change back to root in every event loop.
> >> There're explicit 'change_to_root()' calls if really needed.
> >>
> >> That's why we'd have 3 types of tevent_context pointers:
> >> 1. the raw tevent_context that doesn't do any impersonation
> >> 2. a tevent_context that runs the handlers as root
> >>    tevent_change_to_root()
> >> 3. a tevent_context that runs as the correct user
> >>    tevent_change_to_user(),
> >>    Note: that this also needs to call set_current_service()
> >>
> >> And I think the SMB_VFS modules should only have access to 3.
> > 
> > Hmm if you are using a different tevent context, doesn't it mean we will
> > not be fully asynchronous ?
> > 
> > If we are looping in the 'impersonation' context and a file descriptor
> > becomes ready for reading in the raw context, then we will not server
> > the one in the raw context until we return back from the loop handling
> > the impersonation context ?
> > 
> > The same is true in reverse also ?
> > 
> > Isn't there a risk of starvation here ?
> > What about interactions between code that run in 2 different context ?
> 
> We have 3 tevent_context structures, but two are just wrappers,
> which inject a handler function in order to impersonate before
> calling the real handler.
> 
> tevent_loop_once/_wait will call smb_panic() if called via the wrappers.
> 
> Did you get what I mean?

Yes, makes sense.

I wonder if we shouldn't make this wrapping functionality a standard
feature of tevent. Not the impersonation specific feature, just the
wrapping part.

Simo.


-- 
Simo Sorce
Samba Team GPL Compliance Officer <simo at samba.org>
Principal Software Engineer at Red Hat, Inc. <simo at redhat.com>