[CRIU] [PATCH v3 2/2] unix: don't drop the path on unix sockets if they don't exist

[Pavel Emelyanov]

On 07/13/2016 05:25 PM, Tycho Andersen wrote:
> Hi Pavel,
> 
> On Wed, Jul 13, 2016 at 04:01:59PM +0300, Pavel Emelyanov wrote:
>> Both applied, thanks :)
>>
>> Would you also check whether this thing git fixed:
>>
>> static int bind_unix_sk(int sk, struct unix_sk_info *ui)
>> {
>>         struct sockaddr_un addr;
>>         int cwd_fd = -1;
>>         int ret = -1;
>>
>>         if ((ui->ue->type == SOCK_STREAM) && (ui->ue->state == TCP_ESTABLISHED)) {
>>                 /*
>>                  * FIXME this can be done, but for doing this properly we
>>                  * need to bind socket to its name, then rename one to
>>                  * some temporary unique one and after all the sockets are
>>                  * restored we should walk those temp names and rename
>>                  * some of them back to real ones.
>>                  */
>>                 ret = 0;
>>                 goto done;
>>         }
>>
>> this is for the case when we have established socket that inherited its name
>> from the listening one. Presumably it has, and, id I'm right, would you add
>> a test for this and drop this comment? :)
> 
> Won't established sockets have the same inode as listen sockets?

No, the new pair of sockets will be created upon connect()/accept() calls.

> I think in this case we have to do something a little more clever, as
> the code won't detect them as "deleted". Anyway, I can figure it out
> and send some patches.

Awesome :)

> Tycho
> .
>