未使用keep-alive设置关闭的套接字

目前我有麻烦使用HTTP（S）连接。编译器：XC32 1.44 Health:V2.04（V2.05在这方面没有太大的改变）。我使用NETSIPRESS层打开套接字，直到打开和关闭几个插槽为止。它看起来好像我的插座已经用完了，所以我开始挖一点。有一件事发生了：它看起来像我可以。请输入一个套接字，但对NETAXPROSSOCKIStiON连接的查询从未返回true。在调试之后，我发现我的套接字确实耗尽了。但是NETA PrimeSoCKOPENT不会返回NETApPrimeSimuldIsSub，但有些数字表明一切都很好。原因是：这是不正确的。TrANHANDLE是一个无符号的16位类型，而NETSpRySuxIDIdSoCub被定义为（-1）。这意味着它被签署了。因此，比较是（65535＝＝1）……我还想指出，在一个错误的情况下，TrANHANDLE将被设置为无效的SuthSoT（从TCP。H）。不幸的是，NETA PrimeSimuldIsSub是单独定义的。在我看来，NETSyPrimeValuiSoSub应该定义为无效的套接字。至少两个都应该有一个共同的父定义。但是回到原来的问题（套接字耗尽）：NETAXPROSSOCKOSEND调用TCPIPUTCPY关闭，调用TyTCPDISCONTION。由于某些原因，状态机处于状态TCPIPpTCPYSTATEYLASSTARACK，并与OXTCPHSDENTHOK断开连接，但实际上并没有关闭套接字。NETXPRES-SOCKIT处理程序被释放，但实际的TCP套接字保持打开状态！我将我的代码与TCPIpHelocver示例进行了比较，结果如下：这可能是一个错误：当在请求中设置“连接：保持生存”时，TCP套接字不会被NETApPrimeSockCocket关闭。只有在保持活动设置超时之后，套接字才会关闭。我认为这不是正确的行为。当我显式关闭套接字时，我只是希望套接字立即关闭。当我关闭套接字时，无论如何我不能重新使用连接，因为我失去了所有的socket句柄。作为一种解决办法，我试图降低保持生存超时，但这并没有降低超时，直到套接字被释放。



      以下为原文

    I am currently having trouble to use HTTP(S) connections.
Compiler: xc32 1.44
HARMony: v2.04 (the v2.05 did not change much in that respect)

I use the NET_PRES layer to open the socket and it works until I open and close a few more sockets.
It looked like I was running out of sockets, so I startet to dig a bit.

One thing that came up: it looked like I could open a socket, but the query to NET_PRES_SocketIsConnected never returned true.
After debugging, I discovered that my sockets are indeed depleted. But NET_PRES_SocketOpen would not return NET_PRES_INVALID_SOCKET but some number that idicates that everything went fine.
The reason:

net_pres.c
line 299:
        if (sNetPresSockets[sockIndex].transHandle == NET_PRES_INVALID_SOCKET)

This is not correct. transHandle is a unsigned 16Bit type while NET_PRES_INVALID_SOCKET is defined to (-1). That means it is signed. So the comparison is down to (65535 == -1)...
I also want to point out that, in an error case, transHandle will be set to INVALID_SOCKET (from tcp.h). Unfortunately NET_PRES_INVALID_SOCKET is separately defined. In my opinion NET_PRES_INVALID_SOCKET should be defined to INVALID_SOCKET. At least both should have a common parent define.

But back to the original problem (running out of sockets):
NET_PRES_SocketClose calls TCPIP_TCP_Close which calls _TCPDisconnect. The state machine is in state TCPIP_TCP_STATE_LAST_ACK for some reason and disconnect leaves with _TCP_SEND_OK but is not actually closing the socket. The NET_PRES-Socket-Handler is freed, but the actual TCP Socket stays open!

I compared my code to the tcpip_client example and it came down to the following, which might be a bug:
When "Connection: keep-alive" is set in the request, the tcp-socket is not closed with NET_PRES_SocketClose. Only after the timeout of the keep alive settings, the socket is closed.
I do not think this is a correct behaviour. When I explicitly close the socket, I just expect the socket to be closed immediately. When I close the socket, I cannot re-use the connection anyway, because I lost all the handles to the socket.


As a workaround, I tried to lower the keep-alive timeout with

    TCP_OPTION_KEEP_ALIVE_DATA keepAliveData;
    keepAliveData.keepAliveEnable = true;
    keepAliveData.keepAliveTmo = 500;
    keepAliveData.keepAliveUnackLim = 1;
    if(!NET_PRES_SocketOptionsSet(socket, TCP_OPTION_KEEP_ALIVE, &keepAliveData))
    {
         ....
    }

but that did not lower the timeout until the socket is freed.