On Fri, Jan 27, 2023 at 5:55 PM Richard Guy Briggs <rgb@xxxxxxxxxx> wrote:
> On 2023-01-27 17:35, Paul Moore wrote:
> > On Fri, Jan 27, 2023 at 12:24 PM Richard Guy Briggs <rgb@xxxxxxxxxx> wrote:
> > >
> > > Since FADVISE can truncate files and MADVISE operates on memory, reverse
> > > the audit_skip tags.
> > >
> > > Fixes: 5bd2182d58e9 ("audit,io_uring,io-wq: add some basic audit support to io_uring")
> > > Signed-off-by: Richard Guy Briggs <rgb@xxxxxxxxxx>
> > > ---
> > > io_uring/opdef.c | 2 +-
> > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/io_uring/opdef.c b/io_uring/opdef.c
> > > index 3aa0d65c50e3..a2bf53b4a38a 100644
> > > --- a/io_uring/opdef.c
> > > +++ b/io_uring/opdef.c
> > > @@ -306,12 +306,12 @@ const struct io_op_def io_op_defs[] = {
> > > },
> > > [IORING_OP_FADVISE] = {
> > > .needs_file = 1,
> > > - .audit_skip = 1,
> > > .name = "FADVISE",
> > > .prep = io_fadvise_prep,
> > > .issue = io_fadvise,
> > > },
> >
> > I've never used posix_fadvise() or the associated fadvise64*()
> > syscalls, but from quickly reading the manpages and the
> > generic_fadvise() function in the kernel I'm missing where the fadvise
> > family of functions could be used to truncate a file, can you show me
> > where this happens? The closest I can see is the manipulation of the
> > page cache, but that shouldn't actually modify the file ... right?
>
> I don't know. I was going on the advice of Steve Grubb. I'm looking
> for feedback, validation, correction, here.
Keep in mind it's your name on the patch, not Steve's, and I would
hope that you should be able to stand up and vouch for your own patch.
Something to keep in mind for the future.
As it stands, I think the audit_skip line should stay for
IORING_OP_FADVISE, if you feel otherwise please provide more
explanation as to why auditing is necessary for this operation.
> > > [IORING_OP_MADVISE] = {
> > > + .audit_skip = 1,
> > > .name = "MADVISE",
> > > .prep = io_madvise_prep,
> > > .issue = io_madvise,
> >
> > I *think* this should be okay, what testing/verification have you done
> > on this? One of the things I like to check is to see if any LSMs
> > might perform an access check and/or generate an audit record on an
> > operation, if there is a case where that could happen we should setup
> > audit properly. I did a very quick check of do_madvise() and nothing
> > jumped out at me, but I would be interested in knowing what testing or
> > verification you did here.
>
> No testing other than build/boot/audit-testsuite. You had a test you
> had developed that went through several iterations?
There is an io_uring test in the audit-testsuite that verifies basic
audit record generation, it is not exhaustive.
Patch 2/2 is a no-go from a security perspective (we want to see those
records), and I think skipping on IORING_OP_FADVISE is the correct
thing to do. It may be that skipping on IORING_OP_MADVISE is the
correct thing, but given that it doesn't appear a lot of in-depth
investigation has gone into these patches I would really like to see
some more reasoning on this before we change the current behavior.
--
paul-moore.com
