gen_udp:open/2 does not set {reuseaddr, true} on Windows 10

[ziopio]

Describe the bug
On Windows 10 64-bit, when i try to open an UDP port with option {reuseaddr, true} the option is not applied.
The resulting socket owns the port exclusively. This makes it impossible to open multiple sockets on the same port.
The bug DOES NOT occur on an OTP linux installation executing through WSL.

To Reproduce (OTP 23.3 Eshell V11.2, Windows 10 64-bit )
1> {_,S} = gen_udp:open(0,[{reuseaddr, true}]).
{ok,#Port<0.4>}
2> inet:getopts(S,[reuseaddr]).
{ok,[{reuseaddr,false}]}

Expected behavior (OTP 22 Eshell V10.6.4, Ubuntu 20.04 on WSL )
1> {_,S} = gen_udp:open(0,[{reuseaddr, true}]).
{ok,#Port<0.4>}
2> inet:getopts(S,[reuseaddr]).
{ok,[{reuseaddr,true}]}

Affected versions
Erlang/OTP 23.3 with Eshell V 11.2, 64-bit binary installation on Windows 10

Additional context
I my case i am implementing the RTPS protocol by ObjectManagmentGroup in Erlang. The p2p discovery requires multiple participants to listen to the same port for UDP multicast packets. This bug prevents me to start more than one node on my system.

[bmk]

After some searching on the net, I found that this (REUSEADDR) seems to have issues on windows.
For instance Apache HTTPD only use it on non-windows.

I also found the following comment:

"I understand REUSE is popular under *nix because it prevents re-binding process restart errors
for processes stuck in TIME_WAIT.
Supposedly Windows does not have this issue of preventing socket binding for socket addr/port combinations in TIME_WAIT.
Apparently there is little value to using REUSE under Windows except in cases where you are using multicast."

[juhlig]

The same applies to gen_tcp etc. From inet_drv.c, with my favorite explanatory comment of all times 😁:

otp/erts/emulator/drivers/common/inet_drv.c

Lines 6635 to 6637 in d646fa3

	case INET_OPT_REUSEADDR:
	#ifdef __WIN32__
	continue; /* Bjorn says */

[ziopio]

So... it's more like an intended feature then a bug😅

[bmk]

I think it should be "feature" in this case...

[Maria-12648430]

"Bjorn says", priceless 😄

[juhlig]

I think it should be "feature" in this case...

I don't remember all the details, but IIRC reuseaddr means something very different (and dangerous, of course) on Windows...

[juhlig]

https://docs.microsoft.com/en-us/windows/win32/winsock/using-so-reuseaddr-and-so-exclusiveaddruse

[bjorng]

On Tue, May 25, 2021 at 7:26 PM Maria Scott ***@***.***> wrote: "Bjorn says", priceless 😄 I think that comment was written by Tony Rogvall (https://github.com/tonyrog).

The backstory was that at the time the Erlang distribution had its own driver and there was a separate port program for applications that wanted to use TCP and UDP. Tony was working on a new driver called the inet_driver that could serve both purposes. At that time, I was working with the Windows 95/NT build of Erlang. In the socket code for Windows, we had removed the SO_REUSEADDR option, because we had noticed that it would cause unsafe reuse of listen ports. When Tony showed me code for the inet_driver, I saw that it used SO_REUSEADDR on all platforms and I told him that he would have to take it out for Windows. He did.

[bmk]

Added a PR, which solves this by introducing a new option; 'win_reuseaddr' which
functions the "same way" as 'reuseaddr' but reversed. That is, it is ignored on all
non-Windows.

[bmk]

This option, 'reuseaddr', will be handled the same way on Windows as on other OSes. As of 25.0!

[rickard-green]

@ziopio As of OTP 26 the reuseaddr option will revert back to how it behaved prior to OTP 25 on Windows. The reuseport option will however be added in OTP 26, so you will be able to do what you want by passing the [{reuseaddr,true},{reuseport,true}] options on Windows in order to do what you want. For more information see the #6522 pull request.