In January 2001, in bug 66243, Sonja Mirtitsh wrote: > According to [Wan-Teh], PR_Shutdown should be called befoe PR_Close on a > socket. This is not necessary for Unix, only for Windows. And Wan-Teh added > On some Windows versions (I don't remember whether it's > the Windows 9x or NT family), if you don't call shutdown() > before you close a socket, the peer may get the WSAECONNRESET > or WSAECONNABORTED error. And I asked > So, this is a platform dependent behavior issue. Shouldn't NSPR fix > this by calling shutdown in PR_Close on the affected platform(s)? This symtom (getting an unexpected ECONNRESET due to close without prior shutdown) was also described in https://bugzilla.mozilla.org/show_bug.cgi?id=64695#c26 This behavior is superficially the subject of bug 68869, which concerned the fact that a negative test case was resulting in a TCP socket error (ECONNRESET), not an SSL alert message, when the server ran on Windows NT/2K. That problem occured in a case where libSSL _was_ shutting down the socket for both send and recv (equivalent to close in the TCP protocol) after sending an alert record, and the fix (or workaround) was to only shutdown the socket for send at that point.
You need to log in before you can comment on or make changes to this bug.