tree 752ac99d9e31c72007ec26fae0cbc89e2ed798c1
parent a56ab5c684e88341e30122a0e6012d7142d10f62
author Pete Bentley <44170157+prbprbprb@users.noreply.github.com> 1674844351 +0000
committer GitHub <noreply@github.com> 1674844351 +0000
gpgsig -----BEGIN PGP SIGNATURE-----
 
 wsBcBAABCAAQBQJj1Bi/CRBK7hj4Ov3rIwAAXUoIAKJEvs2kbbwGS63xWMGkkk4K
 6v6rJwqMfZUVk1fM1o4mnsyzaKqr1QIUwztWVzHMKNZTjYgZpEy44+VudpQ2ywqH
 DmmfpY7YJotIh48SX118ba0SLEz+4KS8l4eNdSeSD/BF3cfO0/roi+tVPsB0q9Eq
 sTElLz1oYirD/1lTzu4WCVlilu2hyy55n5iJCjmNEcR9UJDzv7smpqhl0i0kjhS6
 nNipCXM68J2wkYWhg+QYahIGVRLiKTe4O06dAXON9jUOq+Bk2zGvan4Z+dbxXoqE
 QyzdLDaKgIgWBkFa4NmSatWrrqxECgyUobGR4WBPqcOkWRQ9uNiSLQ6pGMPpFeQ=
 =K5zy
 -----END PGP SIGNATURE-----
 

Rework Ssl*Stream creation. (#1109)

* Rework Ssl*Stream creation.

PR #1106 introduced a subtle bug by changing Ssl*Stream
creation to use the public get*Stream() methods.

The contract for those methods is to throw if the socket has been
closed, which mean two threads could race such that
startHandshake() called getInputStream() after the socket has
been closed by another (e.g. timeout) thread.  The upshot was
that getInputStream() would throw, causing close() to be called
before an output stream is created, causing an NPE when trying to
send a TLS close message.

This change moves the actual creation into a method which doesn't
chack the socket state and so is suitable for calling from
startHandshake().

Also added some additional tests around shutdownInput() and
shutdownOutput() as it became apparent these were missing.