Re: [PATCH] bpf: export the net namespace for bpf_sock_ops

[王黎 <wangli09@kuaishou.com>]

I just saw the commit 13d70f5a5ecff367db2fb18ed4ebe433eab8a74c Sun May 24 09:51:15 2020 that already added "struct bpf_sock *, sk” for sk_msg_md.

So Just ignore my question. Thank you all anyway.

> 在 2020年6月8日，上午10:51，王黎 <wangli09@kuaishou.com> 写道：
> 
> Daniel, thanks for your comments.
> 
> bpf_get_netns_cookie_sock really sounds better. 
> 
> But I still have another question that is how can I get the netns cookie when I invoke the function “msg_redirect_hash" because its parameter "struct sk_msg_md” does not have any netns information, and there is no "struct sock * ctx” to be used for the helper function “bpf_get_netns_cookie_sock" too. Add a field for sk_msg_md ? And just like I did for bpf_sock_ops ?
> 
> 
> 
>> 在 2020年6月5日，下午11:22，Daniel Borkmann <daniel@iogearbox.net> 写道：
>> 
>> On 6/5/20 4:53 PM, Jakub Sitnicki wrote:
>>> On Fri, Jun 05, 2020 at 02:40 PM CEST, Wang Li wrote:
>>>> Sometimes we need net namespace as part of the key for BPF_MAP_TYPE_SOCKHASH to
>>>> distinguish the connections with same five-tuples, for example when we do the
>>>> sock_map acceleration for the proxy that uses 127.0.0.1 to 127.0.0.1 connections
>>>> in different containers on same node.
>>>> And we export the netns inum instead of the real pointer of struct net to avoid
>>>> the potential security issue.
>>>> 
>>>> Signed-off-by: Wang Li <wangli09@kuaishou.com>
>>>> Signed-off-by: huangxuesen <huangxuesen@kuaishou.com>
>>>> Signed-off-by: yangxingwu <yangxingwu@kuaishou.com>
>>>> ---
>>>> include/uapi/linux/bpf.h       |  2 ++
>>>> net/core/filter.c              | 17 +++++++++++++++++
>>>> tools/include/uapi/linux/bpf.h |  2 ++
>>>> 3 files changed, 21 insertions(+)
>>>> 
>>>> diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
>>>> index c65b374a5090..0fe7e459f023 100644
>>>> --- a/include/uapi/linux/bpf.h
>>>> +++ b/include/uapi/linux/bpf.h
>>>> @@ -3947,6 +3947,8 @@ struct bpf_sock_ops {
>>>> 				 * there is a full socket. If not, the
>>>> 				 * fields read as zero.
>>>> 				 */
>>>> +	__u32 netns_inum;	/* The net namespace this sock belongs to */
>>>> +
>>> In uapi/linux/bpf.h we have a field `netns_ino` for storing net
>>> namespace inode number in a couple structs (bpf_prog_info,
>>> bpf_map_info). Would be nice to keep the naming constent.
>> 
>> Adding in the middle will break programs. Also, currently we have the
>> merge window open and as such bpf-next is closed. Check status here [0].
>> 
>> Regarding above, we recently added bpf_get_netns_cookie() helper, have
>> you tried to enable this one instead?
>> 
>> [0] http://vger.kernel.org/~davem/net-next.html
>> 
>>>> diff --git a/net/core/filter.c b/net/core/filter.c
>>>> index d01a244b5087..bfe448ace25f 100644
>>>> --- a/net/core/filter.c
>>>> +++ b/net/core/filter.c
>>>> @@ -8450,6 +8450,23 @@ static u32 sock_ops_convert_ctx_access(enum bpf_access_type type,
>>>> 					       is_fullsock));
>>>> 		break;
>>>> 
>>>> +	case offsetof(struct bpf_sock_ops, netns_inum):
>>>> +#ifdef CONFIG_NET_NS
>>>> +		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
>>>> +						struct bpf_sock_ops_kern, sk),
>>>> +				      si->dst_reg, si->src_reg,
>>>> +				      offsetof(struct bpf_sock_ops_kern, sk));
>>>> +		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(
>>>> +						struct sock_common, skc_net),
>>>> +				      si->dst_reg, si->dst_reg,
>>>> +				      offsetof(struct sock_common, skc_net));
>>>> +		*insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg,
>>>> +				      offsetof(struct net, ns.inum));
>>>> +#else
>>>> +		*insn++ = BPF_MOV32_IMM(si->dst_reg, 0);
>>>> +#endif
>>>> +		break;
>>>> +
>>>> 	case offsetof(struct bpf_sock_ops, state):
>>>> 		BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1);
>>>> 
>>>> diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
>>>> index c65b374a5090..0fe7e459f023 100644
>>>> --- a/tools/include/uapi/linux/bpf.h
>>>> +++ b/tools/include/uapi/linux/bpf.h
>>>> @@ -3947,6 +3947,8 @@ struct bpf_sock_ops {
>>>> 				 * there is a full socket. If not, the
>>>> 				 * fields read as zero.
>>>> 				 */
>>>> +	__u32 netns_inum;	/* The net namespace this sock belongs to */
>>>> +
>>>> 	__u32 snd_cwnd;
>>>> 	__u32 srtt_us;		/* Averaged RTT << 3 in usecs */
>>>> 	__u32 bpf_sock_ops_cb_flags; /* flags defined in uapi/linux/tcp.h */
>> 
>