[PATCH v2 5/8] net: af_can: do not leave a dangling sk pointer in can_create()

Ignat Korchagin posted 8 patches 1 month, 2 weeks ago
There is a newer version of this series
[PATCH v2 5/8] net: af_can: do not leave a dangling sk pointer in can_create()
Posted by Ignat Korchagin 1 month, 2 weeks ago
On error can_create() frees the allocated sk object, but sock_init_data()
has already attached it to the provided sock object. This will leave a
dangling sk pointer in the sock object and may cause use-after-free later.

Signed-off-by: Ignat Korchagin <ignat@cloudflare.com>
---
 net/can/af_can.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/net/can/af_can.c b/net/can/af_can.c
index 707576eeeb58..01f3fbb3b67d 100644
--- a/net/can/af_can.c
+++ b/net/can/af_can.c
@@ -171,6 +171,7 @@ static int can_create(struct net *net, struct socket *sock, int protocol,
 		/* release sk on errors */
 		sock_orphan(sk);
 		sock_put(sk);
+		sock->sk = NULL;
 	}
 
  errout:
-- 
2.39.5
Re: [PATCH v2 5/8] net: af_can: do not leave a dangling sk pointer in can_create()
Posted by Vincent MAILHOL 1 month, 2 weeks ago
Hi Ignat,

Thanks for the patch.

On Tue. 8 Oct. 2024 at 06:37, Ignat Korchagin <ignat@cloudflare.com> wrote:
> On error can_create() frees the allocated sk object, but sock_init_data()
> has already attached it to the provided sock object. This will leave a
> dangling sk pointer in the sock object and may cause use-after-free later.

I was about to suggest that this should be backported to stable, but
after reading the cover letter, I now understand that this patch is
more an improvement to avoid false positives on kmemleak & Cie. Maybe
the description could be a bit more nuanced? After patch 1/8 of this
series, it seems that the use-after-free is not possible anymore.

> Signed-off-by: Ignat Korchagin <ignat@cloudflare.com>

See above comment as notwithstanding. This said:

Reviewed-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>


> ---
>  net/can/af_can.c | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/net/can/af_can.c b/net/can/af_can.c
> index 707576eeeb58..01f3fbb3b67d 100644
> --- a/net/can/af_can.c
> +++ b/net/can/af_can.c
> @@ -171,6 +171,7 @@ static int can_create(struct net *net, struct socket *sock, int protocol,
>                 /* release sk on errors */
>                 sock_orphan(sk);
>                 sock_put(sk);
> +               sock->sk = NULL;
>         }
>
>   errout:
> --
> 2.39.5
>
>
Re: [PATCH v2 5/8] net: af_can: do not leave a dangling sk pointer in can_create()
Posted by Ignat Korchagin 1 month, 2 weeks ago
On Tue, Oct 8, 2024 at 3:38 AM Vincent MAILHOL
<mailhol.vincent@wanadoo.fr> wrote:
>
> Hi Ignat,
>
> Thanks for the patch.
>
> On Tue. 8 Oct. 2024 at 06:37, Ignat Korchagin <ignat@cloudflare.com> wrote:
> > On error can_create() frees the allocated sk object, but sock_init_data()
> > has already attached it to the provided sock object. This will leave a
> > dangling sk pointer in the sock object and may cause use-after-free later.
>
> I was about to suggest that this should be backported to stable, but
> after reading the cover letter, I now understand that this patch is
> more an improvement to avoid false positives on kmemleak & Cie. Maybe
> the description could be a bit more nuanced? After patch 1/8 of this
> series, it seems that the use-after-free is not possible anymore.

If we go with Kuniyuki's suggestion in the cover email to replace the
explicit NULL with DEBUG_NET_WARN_ON_ONCE(sock->sk) in __sock_create()
then use-after-free would be possible again except we will easily
catch it. But I will review the description, when I respin the patches
to net-next as requested by Jakub.

> > Signed-off-by: Ignat Korchagin <ignat@cloudflare.com>
>
> See above comment as notwithstanding. This said:
>
> Reviewed-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr>

Thank you.

>
> > ---
> >  net/can/af_can.c | 1 +
> >  1 file changed, 1 insertion(+)
> >
> > diff --git a/net/can/af_can.c b/net/can/af_can.c
> > index 707576eeeb58..01f3fbb3b67d 100644
> > --- a/net/can/af_can.c
> > +++ b/net/can/af_can.c
> > @@ -171,6 +171,7 @@ static int can_create(struct net *net, struct socket *sock, int protocol,
> >                 /* release sk on errors */
> >                 sock_orphan(sk);
> >                 sock_put(sk);
> > +               sock->sk = NULL;
> >         }
> >
> >   errout:
> > --
> > 2.39.5
> >
> >