[CRIU] Re: [RFC] inet unbound sockets restore

[Pavel Emelyanov]

On 05/12/2012 06:35 PM, Cyrill Gorcunov wrote:
> On Sat, May 12, 2012 at 06:30:02PM +0400, Pavel Emelyanov wrote:
>> On 05/12/2012 06:24 PM, Cyrill Gorcunov wrote:
>>> Pavel, take a look please, this patch brings ability to
>>> restore unbound inet sockets, thus test unbound_sockets
>>> become alive.
>>>
>>> (ps. not for merge, I'll send out the whole series later
>>>  once you confirm if this patch looks good for you)
>>>
>>> 	Cyrill
>>
>>> @@ -264,7 +431,7 @@ static int open_inet_sk(struct file_desc *d)
>>>  	 * bind() and listen(), and that's all.
>>>  	 */
>>>  
>>> -	if (inet_bind(sk, ii))
>>> +	if (ii->ie.state != TCP_CLOSE && inet_bind(sk, ii))
>>>  		goto err;
>>>  
>>>  	if (ii->ie.state == TCP_LISTEN) {
>>
>> O_o
> 
> Not really sure I follow, what do yo don't like here?

What is this check here for? I mean -- e.g. unbound UDP socket can be
connected. Here should go something like "if (is_bound(&ii->ie))" where
is_bound() checks for the name being non empty or smth like this.

>>> +static struct inet_sk_desc *gen_uncon_sk(int lfd, const struct fd_parms *p)
>>> +{
>>
>> From here till the relevant } comes some not yet brushed up code, thus I
>> haven't looked at it close enough.
> 
> ok, i'll clean it up more.
> 
>>
>>> +static unsigned int self_sk_inode(int lfd)
>>> +{
>>> +	char path[64], buf[32], *tok;
>>> +
>>> +	snprintf(path, sizeof(path), "/proc/self/fd/%d", lfd);
>>> +	readlink(path, buf, sizeof(buf));
>>> +
>>> +	tok = strtok(buf, "[");
>>> +	tok = strtok(NULL, "]");
>>> +	BUG_ON(!tok);
>>> +
>>> +	return atoi(tok);
>>> +}
>>
>> It's MUCH simpler to call fstat for this.
> 
> Ah, indeed. Thanks!
> 
> 	Cyrill
> _______________________________________________
> CRIU mailing list
> CRIU at openvz.org
> https://openvz.org/mailman/listinfo/criu
> .
>