[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Tue, 3 Mar 2020 09:56:32 -0800
From: Martin KaFai Lau <kafai@...com>
To: Lorenz Bauer <lmb@...udflare.com>
CC: <john.fastabend@...il.com>, Daniel Borkmann <daniel@...earbox.net>,
Jakub Sitnicki <jakub@...udflare.com>,
"David S. Miller" <davem@...emloft.net>,
Jakub Kicinski <kuba@...nel.org>,
Alexey Kuznetsov <kuznet@....inr.ac.ru>,
Hideaki YOSHIFUJI <yoshfuji@...ux-ipv6.org>,
Alexei Starovoitov <ast@...nel.org>,
<kernel-team@...udflare.com>, <linux-kernel@...r.kernel.org>,
<netdev@...r.kernel.org>, <bpf@...r.kernel.org>
Subject: Re: [PATCH bpf-next v2 5/9] bpf: sockmap: allow UDP sockets
On Fri, Feb 28, 2020 at 11:53:40AM +0000, Lorenz Bauer wrote:
> Add basic psock hooks for UDP sockets. This allows adding and
> removing sockets, as well as automatic removal on unhash and close.
>
> sock_map_sk_state_allowed is called from the syscall path, and
> ensures that only established or listening sockets are added.
> No such check exists for the BPF path: we rely on sockets being
> in particular states when a BPF sock ops hook is executed.
> For the passive open hook this means that sockets are actually in
> TCP_SYN_RECV state (and unhashed) when they are added to the
> sock map.
>
> UDP sockets are not saddled with this inconsistency, and so the
> checks for both syscall and BPF path should be identical. Rather
> than duplicating the logic into sock_map_sk_state_allowed merge
> it with sock_map_sk_is_suitable.
>
> Signed-off-by: Lorenz Bauer <lmb@...udflare.com>
> ---
> MAINTAINERS | 1 +
> include/linux/udp.h | 4 ++++
> net/core/sock_map.c | 52 ++++++++++++++++++++++++++------------------
> net/ipv4/Makefile | 1 +
> net/ipv4/udp_bpf.c | 53 +++++++++++++++++++++++++++++++++++++++++++++
> 5 files changed, 90 insertions(+), 21 deletions(-)
> create mode 100644 net/ipv4/udp_bpf.c
>
> diff --git a/MAINTAINERS b/MAINTAINERS
> index 2af5fa73155e..495ba52038ad 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -9358,6 +9358,7 @@ F: include/linux/skmsg.h
> F: net/core/skmsg.c
> F: net/core/sock_map.c
> F: net/ipv4/tcp_bpf.c
> +F: net/ipv4/udp_bpf.c
>
> LANTIQ / INTEL Ethernet drivers
> M: Hauke Mehrtens <hauke@...ke-m.de>
> diff --git a/include/linux/udp.h b/include/linux/udp.h
> index aa84597bdc33..2485a35d113c 100644
> --- a/include/linux/udp.h
> +++ b/include/linux/udp.h
> @@ -143,4 +143,8 @@ static inline bool udp_unexpected_gso(struct sock *sk, struct sk_buff *skb)
>
> #define IS_UDPLITE(__sk) (__sk->sk_protocol == IPPROTO_UDPLITE)
>
> +#ifdef CONFIG_BPF_STREAM_PARSER
> +int udp_bpf_init(struct sock *sk);
> +#endif
> +
> #endif /* _LINUX_UDP_H */
> diff --git a/net/core/sock_map.c b/net/core/sock_map.c
> index c84cc9fc7f6b..d742e1538ae9 100644
> --- a/net/core/sock_map.c
> +++ b/net/core/sock_map.c
[ ... ]
> @@ -466,15 +479,20 @@ static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops)
> ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB;
> }
>
> -static bool sock_map_sk_is_suitable(const struct sock *sk)
> +static bool sock_map_sk_is_udp(const struct sock *sk)
> {
> - return sk->sk_type == SOCK_STREAM &&
> - sk->sk_protocol == IPPROTO_TCP;
> + return sk->sk_type == SOCK_DGRAM && sk->sk_protocol == IPPROTO_UDP;
> }
>
> -static bool sock_map_sk_state_allowed(const struct sock *sk)
> +static bool sock_map_sk_is_suitable(const struct sock *sk, bool from_bpf)
"from_bpf" seems unnecessary.
> {
> - return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN);
> + int tcp_flags = TCPF_ESTABLISHED | TCPF_LISTEN;
> +
> + if (from_bpf)
> + tcp_flags |= TCPF_SYN_RECV;
> +
> + return (sock_map_sk_is_udp(sk) && sk_hashed(sk)) ||
> + (sock_map_sk_is_tcp(sk) && (1 << sk->sk_state) & tcp_flags);
> }
>
> static int sock_map_update_elem(struct bpf_map *map, void *key,
Powered by blists - more mailing lists