Jakub Sitnicki wrote:
> On Sat, May 28, 2022 at 09:54 AM +08, wangyufen wrote:
> > 在 2022/5/28 5:37, Cong Wang 写道:
> >> On Tue, May 24, 2022 at 03:53:11PM +0800, Wang Yufen wrote:
> >>> During TCP sockmap redirect pressure test, the following warning is triggered:
> >>> WARNING: CPU: 3 PID: 2145 at net/core/stream.c:205 sk_stream_kill_queues+0xbc/0xd0
> >>> CPU: 3 PID: 2145 Comm: iperf Kdump: loaded Tainted: G W 5.10.0+ #9
> >>> Call Trace:
> >>> inet_csk_destroy_sock+0x55/0x110
> >>> inet_csk_listen_stop+0xbb/0x380
> >>> tcp_close+0x41b/0x480
> >>> inet_release+0x42/0x80
> >>> __sock_release+0x3d/0xa0
> >>> sock_close+0x11/0x20
> >>> __fput+0x9d/0x240
> >>> task_work_run+0x62/0x90
> >>> exit_to_user_mode_prepare+0x110/0x120
> >>> syscall_exit_to_user_mode+0x27/0x190
> >>> entry_SYSCALL_64_after_hwframe+0x44/0xa9
> >>>
> >>> The reason we observed is that:
> >>> When the listener is closing, a connection may have completed the three-way
> >>> handshake but not accepted, and the client has sent some packets. The child
> >>> sks in accept queue release by inet_child_forget()->inet_csk_destroy_sock(),
> >>> but psocks of child sks have not released.
> >>>
> >> Hm, in this scenario, how does the child socket end up in the sockmap?
> >> Clearly user-space does not have a chance to get an fd yet.
> >>
> >> And, how does your patch work? Since the child sock does not even inheirt
> >> the sock proto after clone (see the comments above tcp_bpf_clone()) at
> >> all?
> >>
> >> Thanks.
> >> .
> > My test cases are as follows:
> >
> > __section("sockops")
> > int bpf_sockmap(struct bpf_sock_ops *skops)
> > {
> > switch (skops->op) {
> > case BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB:
> > case BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB:
> > ...
> > bpf_sock_hash_update(skops, &sock_ops_map, &key, BPF_NOEXIST);
> > break;
> > ...
> > }
>
> Right, when processing the final ACK in tcp_rcv_state_process(), we
> invoke the BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB BPF callback.
>
> This gives a chance to install sockmap sk_prot callbacks.
>
> An accept() without ever calling accept() ;-)
>
> [...]
LGTM as well.
Acked-by: John Fastabend <john.fastabend@xxxxxxxxx>
