[PATCH] lsm: handle the NULL buffer case in lsm_fill_user_ctx()

[Serge E. Hallyn]

On Fri, Mar 15, 2024 at 12:42:27PM -0400, Paul Moore wrote:
> On Fri, Mar 15, 2024 at 12:28 PM Serge E. Hallyn <serge at hallyn.com> wrote:
> > On Fri, Mar 15, 2024 at 12:19:05PM -0400, Paul Moore wrote:
> > > On Fri, Mar 15, 2024 at 12:13 PM Serge E. Hallyn <serge at hallyn.com> wrote:
> > > > On Fri, Mar 15, 2024 at 09:08:47AM -0700, Casey Schaufler wrote:
> > > > > On 3/15/2024 8:02 AM, Serge E. Hallyn wrote:
> > > > > > On Wed, Mar 13, 2024 at 10:22:03PM -0400, Paul Moore wrote:
> > > > > >> Passing a NULL buffer into the lsm_get_self_attr() syscall is a valid
> > > > > >> way to quickly determine the minimum size of the buffer needed to for
> > > > > >> the syscall to return all of the LSM attributes to the caller.
> > > > > >> Unfortunately we/I broke that behavior in commit d7cf3412a9f6
> > > > > >> ("lsm: consolidate buffer size handling into lsm_fill_user_ctx()")
> > > > > >> such that it returned an error to the caller; this patch restores the
> > > > > >> original desired behavior of using the NULL buffer as a quick way to
> > > > > >> correctly size the attribute buffer.
> > > > > >>
> > > > > >> Cc: stable at vger.kernel.org
> > > > > >> Fixes: d7cf3412a9f6 ("lsm: consolidate buffer size handling into lsm_fill_user_ctx()")
> > > > > >> Signed-off-by: Paul Moore <paul at paul-moore.com>
> > > > > >> ---
> > > > > >>  security/security.c | 8 +++++++-
> > > > > >>  1 file changed, 7 insertions(+), 1 deletion(-)
> > > > > >>
> > > > > >> diff --git a/security/security.c b/security/security.c
> > > > > >> index 5b2e0a15377d..7e118858b545 100644
> > > > > >> --- a/security/security.c
> > > > > >> +++ b/security/security.c
> > > > > >> @@ -780,7 +780,9 @@ static int lsm_superblock_alloc(struct super_block *sb)
> > > > > >>   * @id: LSM id
> > > > > >>   * @flags: LSM defined flags
> > > > > >>   *
> > > > > >> - * Fill all of the fields in a userspace lsm_ctx structure.
> > > > > >> + * Fill all of the fields in a userspace lsm_ctx structure.  If @uctx is NULL
> > > > > >> + * simply calculate the required size to output via @utc_len and return
> > > > > >> + * success.
> > > > > >>   *
> > > > > >>   * Returns 0 on success, -E2BIG if userspace buffer is not large enough,
> > > > > >>   * -EFAULT on a copyout error, -ENOMEM if memory can't be allocated.
> > > > > >> @@ -799,6 +801,10 @@ int lsm_fill_user_ctx(struct lsm_ctx __user *uctx, u32 *uctx_len,
> > > > > >>            goto out;
> > > > > >>    }
> > > > > >>
> > > > > >> +  /* no buffer - return success/0 and set @uctx_len to the req size */
> > > > > >> +  if (!uctx)
> > > > > >> +          goto out;
> > > > > > If the user just passes in *uctx_len=0, then they will get -E2BIG
> > > > > > but still will get the length in *uctx_len.
> > > > >
> > > > > Yes.
> > > > >
> > > > > > To use it this new way, they have to first set *uctx_len to a
> > > > > > value larger than nctx_len could possibly be, else they'll...
> > > > > > still get -E2BIG.
> > > > >
> > > > > Not sure I understand the problem. A return of 0 or E2BIG gets the
> > > > > caller the size.
> > > >
> > > > The problem is that there are two ways of doing the same thing, with
> > > > different behavior.  People are bound to get it wrong at some point,
> > > > and it's more corner cases to try and maintain (once we start).
> > >
> > > I have a different perspective on this, you can supply either a NULL
> > > buffer and/or a buffer that is too small, including a size of zero,
> > > and you'll get back an -E2BIG and a minimum buffer size.  What's the
> > > old wisdom, be conservative in what you send and liberal in what you
> > > accept?
> >
> > But if you pass a NULL uctx, then surely you should send *uctx_len=0.
> > And that is already handled.
> 
> Why should we assume that userspace is always going to behave a
> certain way?  Userspace is going to do crazy stuff, that's a given, I
> just want to make sure that we protect ourselves against the really
> crazy stuff, and if we can do something useful with the moderately
> crazy stuff I think we should.
> 
> > What you are adding is that the user can pass NULL uctx, but a large
> > uctx_len value.
> >
> > Perhaps should change my objection, and say that I would prefer the
> > comment fix to suggest passing in uctx_len=0 and uctx=NULL, and expect
> > a -E2BIG.  The NULL check can stay (though I still think it should
> > return -E2BIG).
> >
> > Because with the current comment update, the user may pass in
> > uctx=NULL, but the actual rv will change between 0 and -E2BIG
> > depending on the uctx_len they sent in.  Which is whack, since
> > the incoming value means nothing.
> 
> I think that's a desirable behavior, if you pass in a NULL buffer
> we'll provide you with the required size and return -E2BIG if the size
> you gave us was too small, and zero/success if the size you provided
> was adequate.
> 
> Maybe I'm being stupid and this really is "whack", but you've got to
> help me understand what harm can come from the behavior above.

Returning success if uctx==NULL and uctx_len is big enough sends the
message that this -sending a non-zero length and uctx=NULL is a
recommended use case.  It should not be a recommended use.

If the user wants to find out the size of the buffer, they can already
do so more reliably in the pre-existing way, passing in a *length=0 (and
a NULL or random uctx).  I say more reliable, because they can predict
the function return value in this case: -E2BIG.

So the only way uctx==NULL and uctx_len!=0 should happen is by accident,
and in that case sending back 'success' is misleading.  

> ...  What's the
> old wisdom, be conservative in what you send and liberal in what you
> accept?

If you've received garbage, you should let the sender know, rather than
return 'success' in the hopes that it wasn't important.

But anyway I'll stop here - it doesn't break anything directly, I just
think it's a bad API with the potential for future harder to spot bugs.

-serge