On Wed, Jun 2, 2021 at 9:34 PM Martin KaFai Lau <kafai@xxxxxx> wrote:
>
> On Wed, Jun 02, 2021 at 07:08:13PM +0000, Zvi Effron wrote:
> > + data = xdp->data_meta + metalen;
> > + if (data > xdp->data_end)
> This test and...
>
> > + return -EINVAL;
> > + xdp->data = data;
> > +
> > + if (xdp_md->data_end - xdp_md->data != xdp->data_end - xdp->data)
> this one. It is because the user input "xdp_md->data_end" does not
> match with kattr->test.data_size_in? These tests are disconnected from
> where the actual invalid input is. How about direclty testing
> xdp_md->data_end in bpf_prog_test_run_xdp() instead?
>
The first test is required because it's possible that the user provided value
for `xdp_md->data` could be greater than the user provided value for
`xdp_md->data_end`. For example, a user could pass a struct xdp_md that looked
like this:
```
struct xdp_md {
.data_meta = 0,
.data = 12,
.data_end = 8,
};
```
We're moving the test and making it more clear what it's testing for in our
next patch version.
The second test can be eliminated by moving the test for xdp_md->data_end
against kattr->test.data_size_in as you suggest, which also allows simplifying
the surrounding code. We're updating that in our next patch version.
We're also addressing your other feedback in our next patch version.
