Re: [Qemu-devel] [PATCH] qapi: InitSocketAddress: add keepalive option

[Vladimir Sementsov-Ogievskiy]

07.06.2019 20:22, Markus Armbruster wrote:
> Vladimir Sementsov-Ogievskiy <address@hidden> writes:
> 
>> 06.06.2019 14:17, Daniel P. Berrangé wrote:
>>> On Thu, Jun 06, 2019 at 01:15:33PM +0300, Vladimir Sementsov-Ogievskiy 
>>> wrote:
>>>> Signed-off-by: Vladimir Sementsov-Ogievskiy <address@hidden>
>>>> ---
>>>>
>>>> Hi all!
>>>>
>>>> This is a continuation of "[PATCH v2 0/2] nbd: enable keepalive", but
>>>> it's a try from another side, so almost nothing common with v2.
> 
> Please explain intended use of your new option in your commit message.

Will do. Actual reason is keepalive for nbd-client.

> 
>>>>    qapi/sockets.json   |  5 ++++-
>>>>    util/qemu-sockets.c | 13 +++++++++++++
>>>>    2 files changed, 17 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/qapi/sockets.json b/qapi/sockets.json
>>>> index fc81d8d5e8..aefa024051 100644
>>>> --- a/qapi/sockets.json
>>>> +++ b/qapi/sockets.json
>>>> @@ -53,6 +53,8 @@
>>>>    #
>>>>    # @ipv6: whether to accept IPv6 addresses, default try both IPv4 and 
>>>> IPv6
>>>>    #
>>>> +# @keepalive: enable keepalive when connecting to this socket (Since 4.1)
>>>> +#
>>>>    # Since: 1.3
>>>>    ##
>>>>    { 'struct': 'InetSocketAddress',
>>>> @@ -61,7 +63,8 @@
>>>>        '*numeric':  'bool',
>>>>        '*to': 'uint16',
>>>>        '*ipv4': 'bool',
>>>> -    '*ipv6': 'bool' } }
>>>> +    '*ipv6': 'bool',
>>>> +    '*keepalive': 'bool' } }
> 
> I know the C identifier is SO_KEEPALIVE, but let's stick to proper
> English words in QMP: keep-alive.

Ok

> 
>>>>    
>>>>    ##
>>>>    # @UnixSocketAddress:
>>>> diff --git a/util/qemu-sockets.c b/util/qemu-sockets.c
>>>> index 8850a280a8..d2cd2a9d4f 100644
>>>> --- a/util/qemu-sockets.c
>>>> +++ b/util/qemu-sockets.c
>>>> @@ -457,6 +457,19 @@ int inet_connect_saddr(InetSocketAddress *saddr, 
>>>> Error **errp)
>>>>        }
>>>>    
>>>>        freeaddrinfo(res);
>>>> +
>>>> +    if (saddr->keepalive) {
>>>
>>> IIUC, best practice is to use 'saddr->has_keepalive && saddr->keepalive'
>>
>> As I remember, now all optional fields are zeroed. But I'm not against 
>> stricter condition.
> 
> As far as I'm concerned, relying on zero-initialization of optional
> members is fine.
> 
>>
>>>
>>>> +        int val = 1;
>>>> +        int ret = qemu_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
>>>> +                                  &val, sizeof(val));
>>>> +
>>>> +        if (ret < 0) {
>>>> +            error_setg_errno(errp, errno, "Unable to set KEEPALIVE");
>>>> +            close(sock);
>>>> +            return -1;
>>>> +        }
>>>> +    }
>>>> +
>>>>        return sock;
>>>>    }
> 
> Possibly ignorant question: why obey the keep-alive option for active
> connections (made with inet_connect_saddr()), but not passive ones (made
> with inet_listen_saddr() + accept())?

Hmm. It's a bit trickier, as seems I can't do it on socket level, as parameter 
keep-alive I
get for listen part, but keep-alive should be enabled on socket from accept. So 
this should
be implemented on qio_channel level.. I'd prefer not implement it now. We now 
only interested
in keep-alive for client, and seems generally keep-alive is more usable for 
client part.

> 
> Do you need to update inet_parse()?

will do, thank you for reviewing!



-- 
Best regards,
Vladimir