[CRIU] [PATCH] unix: allow --external unix[ino]:stub[string]

[Pavel Emelyanov]

On 02/28/2017 04:47 PM, Ruslan Kuprieiev wrote:
> Needles to say it's useful for debug, but it is also
> very useful when you syncronise checkpoint/restore
> whith a dumpee using unix socketpair and want it to
> painlessly receive notification on restore.
> 
> I.e. on dump, if --external unix[ino]:foo is specified,
> criu will treat specified unix socket almost the same
> way it handles criu service socket. It will dump it as
> closed and will put "foo" string into images, to put
> that "foo" string into that socket on restore. This is
> almost identical to what criu does with service socket,
> but instead of putting a Criu_Resp message into it on
> restore, it puts an arbitrary string.

OK, with the explanation why this option is good for dump,
I have one question left on the patch itself, inline :)

> @@ -1165,8 +1179,15 @@ static int open_unixsk_standalone(struct unix_sk_info *ui, int *new_fd)
>  			return -1;
>  		}
>  
> -		if (send_criu_dump_resp(sks[1], true, true) == -1)
> -			return -1;
> +		if (ui->ue->stub) {
> +			pr_warn("Writing stub %s to unix socket %#x\n",
> +				(char *)ui->ue->stub, ui->ue->ino);
> +			if (write(sks[1], ui->ue->stub, strlen(ui->ue->stub)) != strlen(ui->ue->stub))
> +				return -1;
> +		} else {
> +			if (send_criu_dump_resp(sks[1], true, true) == -1)
> +				return -1;
> +		}

This is the only place that you send the stub value at, but it's
not external socket handling, it's the RPC socket that client uses 
to connect to CRIU and via which it receives the "restored" message.

So your goal is actually not to allow for random stuff to be sent via
external unix sockets, but rather to make criu send something
meaningful via the RPC socket for self-dump case. Is it?

>  
>  		close(sks[1]);
>  		sk = sks[0];