Re: [PATCH] bpf: add missing NULL check for __dev_get_by_index
From: Alexei Starovoitov
Date: Sun Mar 09 2025 - 06:00:38 EST
On Sat, Mar 8, 2025 at 10:56 PM Qasim Ijaz <qasdev00@xxxxxxxxx> wrote:
>
> The __dev_get_by_index function can return NULL if it fails to
> find a device with the provided ifindex.
>
> We should handle this case by adding a NULL check
> and cleaning up if it does happened.
>
> Signed-off-by: Qasim Ijaz <qasdev00@xxxxxxxxx>
> Fixes: a38845729ea3 ("bpf: offload: add map offload infrastructure")
> ---
> kernel/bpf/offload.c | 6 ++++++
> 1 file changed, 6 insertions(+)
>
> diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c
> index a10153c3be2d..28a30fa4457a 100644
> --- a/kernel/bpf/offload.c
> +++ b/kernel/bpf/offload.c
> @@ -530,6 +530,12 @@ struct bpf_map *bpf_map_offload_map_alloc(union bpf_attr *attr)
> bpf_map_init_from_attr(&offmap->map, attr);
> rtnl_lock();
> offmap->netdev = __dev_get_by_index(net, attr->map_ifindex);
> + if (!offmap->netdev) {
> + rtnl_unlock();
> + bpf_map_area_free(offmap);
> + return ERR_PTR(-ENODEV);
> + }
1. you're using some old git tree.
there is no such code in bpf_map_offload_map_alloc().
2. The actual code has bpf_dev_offload_check(offmap->netdev)
that does the NULL check already.
pw-bot: cr
> +
> netdev_lock_ops(offmap->netdev);
> down_write(&bpf_devs_lock);
> err = bpf_dev_offload_check(offmap->netdev);
> --
> 2.39.5
>