Patch for adding SO_REUSEPORT socket option
Michael McMahon
michael.x.mcmahon at oracle.com
Mon Nov 23 17:35:10 UTC 2015
Lucy,
We can run the change through the standard test system here that will check
it out on all platforms.
Alan mentioned to me earlier that we could tweak the wording that I
suggested
to be more consistent with other options. So, I just need to check that
with him.
Thanks
Michael.
On 23/11/15 17:21, Lu, Yingqi wrote:
> Michael/Volker/Alan,
>
> Thank you all very much for your feedback!
>
> Michael, I like your wording for SO_REUSEPORT in java.net.StandardSocketOptions and I will remove the link to the Linux page. It will be updated for the next version of the patch. Regarding to "reference this text from everywhere else", do you mean doing something like {@see StandardSocketOptions#SO_REUSEPORT}?
>
> Michael/Volker, yes, I agree that we should enable this feature on all the supporting platforms. I will try to add the code in the next version as well. However, I do not have all the OSes so that I might not be able to test them. Can any of you please help with the testing?
>
> Alan, I will add the isReusePortSupported method in the Net.c and only add SO_REUSEPORT into the options set only when it is supported.
>
> Thank you,
> Lucy
>
> -----Original Message-----
> From: Michael McMahon [mailto:michael.x.mcmahon at oracle.com]
> Sent: Monday, November 23, 2015 3:42 AM
> To: Volker Simonis <volker.simonis at gmail.com>; Alan Bateman <Alan.Bateman at oracle.com>
> Cc: Kharbas, Kishor <kishor.kharbas at intel.com>; net-dev at openjdk.java.net; Lu, Yingqi <yingqi.lu at intel.com>
> Subject: Re: Patch for adding SO_REUSEPORT socket option
>
> How about the following for the apidoc in j.n.StandardSocketOptions?
>
> We can then reference this text from everywhere else rather than repeating it.
>
> (just noticed Lucy was not included in the last few mails)
>
> - Michael
>
> /**
> * Re-use port.
> *
> * <p> The value of this socket option is a {@code Boolean} that represents
> * whether the option is enabled or disabled. The exact semantics of this
> * socket option are socket type and system dependent.
> *
> * SO_REUSEPORT can be used with both TCP and UDP sockets.
> *
> * <p> In the case of stream-oriented sockets, this socket option allows
> * multiple listening sockets to be bound to both the same address
> * and same port.
> *
> * <p> For datagram-oriented sockets the socket option is used to allow
> * multiple UDP sockets to be bound to the same address and port.
> *
> * @implNote Generally, this option must be set before the
> * socket is bound or connected. Changing the value of this socket option
> * after the socket is bound has no effect. There may be further
> * restrictions imposed by implementations such as the requirement that
> * all sockets bound to the same port, must have set the option.
> */
>
> On 23/11/15 10:54, Michael McMahon wrote:
>> I agree we should enable the option on all platforms.
>> We can add the code to do that and run the tests.
>>
>> On the existing use of SO_REUSEPORT on AIX and Mac it appears that is
>> set to emulate expected behavior on other platforms when SO_REUSEADDR
>> is set for datagram sockets.
>> The expectation is that ports can be reused for datagram sockets and
>> the JCK tests this. So, I guess we have to leave this behavior by
>> default, except if SO_REUSEPORT is explicitly disabled maybe. Though
>> this code hasn't been forward ported to JDK 9 yet.
>>
>> For reference, SO_REUSEPORT on Linux is documented here
>> http://man7.org/linux/man-pages/man7/socket.7.html
>>
>> - Michael
>>
>> On 23/11/15 09:13, Volker Simonis wrote:
>>> Hi Lucy,
>>>
>>> in general I support the addition of SO_REUSEPORT to the set of
>>> standard socket options. However for me the problem is not that this
>>> new option is not supported on all platforms, but instead that it has
>>> such different semantics on different platforms. If you look at the
>>> code, you'll see that we already implicitly set SO_REUSEPORT on Mac
>>> and AIX for datagram sockets for which we set SO_REUSEADDR. So maybe
>>> we have to rethink this, once SO_REUSEPORT becomes available as a
>>> standard socket option.
>>>
>>> I like the new wording you've posted for JavaDoc of SO_REUSEPORT, but
>>> I think the sentence:
>>>
>>> * Although SO_REUSEADDR option already enables similar
>>> * functionality, SO_REUSEPORT prevents port hijacking and
>>> * distributes the involving datagrams evenly across all of the
>>> * receiving threads.
>>>
>>> refers to a Linux-specific implementation detail which shouldn't be
>>> mentioned in the general documentation. You already have the sentence
>>> "The exact semantics of this socket option are socket type and system
>>> dependent" which should let everybody think twice before using this
>>> option. I'm also not sure about the link to the Linux article but I
>>> again think it is inappropriate in a general API documentation
>>> (otherwise we would have to add links for every platform which
>>> supports SO_REUSEPORT).
>>>
>>> As far as I can see (and please correct me if I'm wrong) you actually
>>> only add the new option for Linux platforms. But this socket option
>>> is also supported on Solaris (>= 11), MacOS X, AIX. So could you
>>> please enable it on the other platforms as well.
>>>
>>> Finally I want to mention the good stackoverflow article at
>>> http://stackoverflow.com/questions/14388706/socket-options-so-reusead
>>> dr-and-so-reuseport-how-do-they-differ-do-they-mean-t
>>>
>>> which covers the topic SO_REUSEADDR vs. SO_REUSEPORT quite well. And
>>> I've collected the man-page entries for SO_REUSEADDR and SO_REUSEPORT
>>> for the systems I have (unfortunately, I couldn't find an updated
>>> Linux man-page which mentions SO_REUSEPORT):
>>>
>>> Linux
>>> =====
>>>
>>> SO_REUSEADDR
>>> Indicates that the rules used in validating addresses
>>> supplied in a bind(2) call should allow reuse of local
>>> addresses. For AF_INET sockets this means that a socket
>>> may bind, except when there is an active listening
>>> socket bound to the address. When the listening socket
>>> is bound to INADDR_ANY with a specific port then it is
>>> not possi- ble to bind to this port for any local
>>> address. Argument is an integer boolean flag.
>>>
>>> Linux will only allow port reuse with the SO_REUSEADDR option
>>> when this option was set both in the previous program that
>>> performed a bind(2) to the port and in the program that wants
>>> to reuse the port. This differs from some implementations
>>> (e.g., FreeBSD) where only the later program needs to set the
>>> SO_REUSEADDR option. Typically this difference is invisi- ble,
>>> since, for example, a server program is designed to always set
>>> this option.
>>>
>>> MacOS X
>>> =======
>>> SO_REUSEADDR enables local address reuse
>>> SO_REUSEPORT enables duplicate address and port bindings
>>>
>>> SO_REUSEADDR indicates that the rules used in validating
>>> addresses supplied in a bind(2) call should allow reuse of local
>>> addresses.
>>>
>>> SO_REUSEPORT allows completely duplicate bindings by multiple
>>> processes if they all set SO_REUSEPORT before bind- ing the port.
>>> This option permits multiple instances of a program to each
>>> receive UDP/IP multicast or broadcast datagrams destined for the
>>> bound port.
>>>
>>> Solaris
>>> =======
>>>
>>> SO_REUSEADDR enable/disable local address reuse
>>>
>>>
>>> SO_REUSEPORT enable/disable local port reuse for
>>> PF_INET/PF_INET6 socket
>>>
>>> The SO_REUSEADDR/SO_REUSEPORT options indi- cate that the rules
>>> used in validating addresses and ports supplied in a
>>> bind(3SOCKET) call should allow reuse of local addresses or
>>> ports.
>>>
>>> AIX
>>> ===
>>>
>>> SO_REUSEADDR
>>> Specifies that the rules used in validating
>>> addresses supplied by a bind subroutine should
>>> allow reuse of a local port. A particular IP
>>> address can only be bound once to the same
>>> port. This option enables or disables reuse of
>>> local ports.
>>>
>>> SO_REUSEADDR allows an application to explicitly
>>> deny subsequent bind subroutine to the port/address
>>> of the socket with SO_REUSEADDR set. This allows an
>>> application to block other applications from
>>> binding with the bind subroutine.
>>>
>>> SO_REUSEPORT
>>> Specifies that the rules used in validating
>>> addresses supplied by a bind subroutine should
>>> allow reuse of a local port/address
>>> combination. Each binding of the port/address
>>> combination must specify the SO_REUSEPORT socket
>>> option. This option enables or disables the reuse
>>> of local port/address combinations.
>>>
>>> HPUX
>>> ====
>>>
>>> SO_REUSEADDR
>>> (int; boolean; AF_INET sockets only) If enabled, allows
>>> a local address to be reused in subsequent calls to
>>> bind(). Default: disallowed.
>>>
>>> SO_REUSEPORT
>>> (int; boolean; AF_INET sockets only) If enabled, allows
>>> a local address and port to be reused in subsequent
>>> calls to bind(). Default: disallowed.
>>>
>>> Setting the SO_REUSEADDR option allows the local socket address
>>> to be reused in subsequent calls to bind(). This permits
>>> multiple SOCK_STREAM sockets to be bound to the same local
>>> address, as long as all existing sockets with the desired local
>>> address are in a connected state before bind() is called for a
>>> new socket. For SOCK_DGRAM sockets, SO_REUSEADDR allows
>>> multiple sockets to receive UDP multicast datagrams addressed to
>>> the bound port number. For all SOCK_DGRAM sockets bound to the
>>> same local address, SO_REUSEADDR must be set before calling
>>> bind().
>>>
>>> Setting the SO_REUSEPORT option allows multiple SOCK_DGRAM
>>> sockets to share the same address and port. Each one of those
>>> sockets, including the first one to use that port, must specify
>>> this option before calling bind().
>>>
>>> Regards,
>>> Volker
>>>
>>>
>>> On Mon, Nov 23, 2015 at 9:00 AM, Alan Bateman
>>> <Alan.Bateman at oracle.com> wrote:
>>>> On 23/11/2015 04:12, Lu, Yingqi wrote:
>>>>
>>>> Hi Alan,
>>>>
>>>>
>>>>
>>>> One more question please J I want to make sure I understand
>>>> correctly on your following suggestion. In order to use
>>>> supportedOptions method to test SO_REUSEPORT, I will need to first
>>>> write a native function to check if SO_REUSEPORT is supported. Then,
>>>> in the defaultOptions method, I do a conditional add for
>>>> StandardSocketOptions.SO_REUSEPORT if it is supported on the
>>>> platform? Is this a preferred way to implement? Please let me know!
>>>>
>>>>
>>>> Yes as supportedOptions() shouldn't return SO_REUSEPORT in the set
>>>> when it's not supported. It might be simplest to put that code in
>>>> sun.nio.ch.Net, maybe isReusePortSupported or some such method. In
>>>> the implementation
>>>> (Net.c) then you can return true or false depending on the platform
>>>> and maybe kernel version.
>>>>
>>>> -Alan
More information about the net-dev
mailing list