Re: [PATCH net-next] net: core: add SOL_SOCKET filter for bpf getsockopt hook

[Stanislav Fomichev]

On Wed, May 10, 2023 at 2:41 PM Aleksandr Mikhalitsyn
<aleksandr.mikhalitsyn@xxxxxxxxxxxxx> wrote:
>
> On Wed, May 10, 2023 at 11:31 PM Stanislav Fomichev <sdf@xxxxxxxxxx> wrote:
> >
> > On Wed, May 10, 2023 at 8:23 AM Alexander Mikhalitsyn
> > <aleksandr.mikhalitsyn@xxxxxxxxxxxxx> wrote:
> > >
> > > We have per struct proto ->bpf_bypass_getsockopt callback
> > > to filter out bpf socket cgroup getsockopt hook from being called.
> > >
> > > It seems worthwhile to add analogical helper for SOL_SOCKET
> > > level socket options. First user will be SO_PEERPIDFD.
> > >
> > > This patch was born as a result of discussion around a new SCM_PIDFD interface:
> > > https://lore.kernel.org/all/20230413133355.350571-3-aleksandr.mikhalitsyn@xxxxxxxxxxxxx/
> > >
> > > Cc: Alexei Starovoitov <ast@xxxxxxxxxx>
> > > Cc: Daniel Borkmann <daniel@xxxxxxxxxxxxx>
> > > Cc: Christian Brauner <brauner@xxxxxxxxxx>
> > > Cc: Stanislav Fomichev <sdf@xxxxxxxxxx>
> > > Cc: "David S. Miller" <davem@xxxxxxxxxxxxx>
> > > Cc: Eric Dumazet <edumazet@xxxxxxxxxx>
> > > Cc: Jakub Kicinski <kuba@xxxxxxxxxx>
> > > Cc: Paolo Abeni <pabeni@xxxxxxxxxx>
> > > Cc: Kuniyuki Iwashima <kuniyu@xxxxxxxxxx>
> > > Cc: linux-kernel@xxxxxxxxxxxxxxx
> > > Cc: netdev@xxxxxxxxxxxxxxx
> > > Cc: bpf@xxxxxxxxxxxxxxx
> > > Signed-off-by: Alexander Mikhalitsyn <aleksandr.mikhalitsyn@xxxxxxxxxxxxx>
> > > ---
> > >  include/linux/bpf-cgroup.h | 8 +++++---
> > >  include/net/sock.h         | 1 +
> > >  net/core/sock.c            | 5 +++++
> > >  3 files changed, 11 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/include/linux/bpf-cgroup.h b/include/linux/bpf-cgroup.h
> > > index 57e9e109257e..97d8a49b35bf 100644
> > > --- a/include/linux/bpf-cgroup.h
> > > +++ b/include/linux/bpf-cgroup.h
> > > @@ -387,10 +387,12 @@ static inline bool cgroup_bpf_sock_enabled(struct sock *sk,
> > >         int __ret = retval;                                                    \
> > >         if (cgroup_bpf_enabled(CGROUP_GETSOCKOPT) &&                           \
> > >             cgroup_bpf_sock_enabled(sock, CGROUP_GETSOCKOPT))                  \
> > > -               if (!(sock)->sk_prot->bpf_bypass_getsockopt ||                 \
> > > -                   !INDIRECT_CALL_INET_1((sock)->sk_prot->bpf_bypass_getsockopt, \
> > > +               if (((level != SOL_SOCKET) ||                                  \
> > > +                    !sock_bpf_bypass_getsockopt(level, optname)) &&           \
> > > +                   (!(sock)->sk_prot->bpf_bypass_getsockopt ||                \
> >
> > Any reason we are not putting this into bpf_bypass_getsockopt for
> > af_unix struct proto? SO_PEERPIDFD seems relevant only for af_unix?
>
> Yes, that should work perfectly well. The reason why I'm going this
> way is that we are
> declaring all SOL_SOCKET-level options in the net/core/sock.c which is
> not specific to any address family.
> It seems reasonable to have a way to filter out getsockopt for these
> options too.
>
> But I'm not insisting on that way.

Yeah, let's move it into af_unix struct proto for now. That should
avoid adding extra conditionals for a few places that care about
performance (tcp zerocopy fastpath).
If we'd ever need to filter out generic SOL_SOCKET level options that
apply for all sockets, we might put (and copy-paste) them in the
respective {tcp,udp,unix,etc}_bpf_bypass_getsockopt.

> Kind regards,
> Alex
>
> >
> > > +                    !INDIRECT_CALL_INET_1((sock)->sk_prot->bpf_bypass_getsockopt, \
> > >                                         tcp_bpf_bypass_getsockopt,             \
> > > -                                       level, optname))                       \
> > > +                                       level, optname)))                      \
> > >                         __ret = __cgroup_bpf_run_filter_getsockopt(            \
> > >                                 sock, level, optname, optval, optlen,          \
> > >                                 max_optlen, retval);                           \
> > > diff --git a/include/net/sock.h b/include/net/sock.h
> > > index 8b7ed7167243..530d6d22f42d 100644
> > > --- a/include/net/sock.h
> > > +++ b/include/net/sock.h
> > > @@ -1847,6 +1847,7 @@ int sk_getsockopt(struct sock *sk, int level, int optname,
> > >                   sockptr_t optval, sockptr_t optlen);
> > >  int sock_getsockopt(struct socket *sock, int level, int op,
> > >                     char __user *optval, int __user *optlen);
> > > +bool sock_bpf_bypass_getsockopt(int level, int optname);
> > >  int sock_gettstamp(struct socket *sock, void __user *userstamp,
> > >                    bool timeval, bool time32);
> > >  struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
> > > diff --git a/net/core/sock.c b/net/core/sock.c
> > > index 5440e67bcfe3..194a423eb6e5 100644
> > > --- a/net/core/sock.c
> > > +++ b/net/core/sock.c
> > > @@ -1963,6 +1963,11 @@ int sock_getsockopt(struct socket *sock, int level, int optname,
> > >                              USER_SOCKPTR(optlen));
> > >  }
> > >
> > > +bool sock_bpf_bypass_getsockopt(int level, int optname)
> > > +{
> > > +       return false;
> > > +}
> > > +
> > >  /*
> > >   * Initialize an sk_lock.
> > >   *
> > > --
> > > 2.34.1
> > >