RFR: 8369950: TLS connection to IPv6 address fails with BCJSSE due to IllegalArgumentException

[Jaikiran Pai]

On Mon, 1 Dec 2025 14:12:04 GMT, Sergey Chernyshev <schernyshev at openjdk.org> wrote:

> Hi all,
> 
> Let me propose a fix and a test case for JDK-8369950.
> 
> The failure reproduces with BCJSSE provider and all implementations of SSLSocket other than SSLSocketImpl.
> 
> In the test case an anonymous wrapper is used, over the standard SSLSocketImpl, to simulate an external JSSE provider. The test case shows the same behavior as in BCJSSE (failure due to non-LDH ASCII characters in the SNI host name).
> 
> The fix avoids constructing SNIHostName when the URL host name is an IPv4 or IPv6 literal address. Other than that, all other FQDN host names that have invalid characters (non-LDH ASCII characters) still produce that exception.
> 
> SNIHostName, as defined in
> https://github.com/openjdk/jdk/blob/873f8a696fa45c7d94a164be20cf3c797ce7f2a6/src/java.base/share/classes/javax/net/ssl/SNIHostName.java#L44-L66
> 
> has the fully qualified DNS hostname of the server. As follows from the section 3, "Server Name Indication", RFC 6066, `Literal IPv4 and IPv6 addresses are not permitted in "HostName"`.
> 
> The fix mirrors the behavior of SSLSocketImpl, that avoids constructing the SNIHostName from literal addresses. Please see
> 
> https://github.com/openjdk/jdk/blob/873f8a696fa45c7d94a164be20cf3c797ce7f2a6/src/java.base/share/classes/sun/security/ssl/Utilities.java#L110-L116
> 
> Testing:
> - standard jtreg tests goups showed no regressions
> - the new test passes with the fix and fails otherwise
> - passes also with BCJSSE in FIPS and standard mode 
> 
> <details><summary> BCJSSE standard</summary>
> 
> 
> STDOUT:
> STDERR:
> Dez. 01, 2025 2:44:02 PM org.bouncycastle.jsse.provider.PropertyUtils getBooleanSecurityProperty
> INFORMATION: Found boolean security property [keystore.type.compat]: true
> Dez. 01, 2025 2:44:02 PM org.bouncycastle.jsse.provider.PropertyUtils getStringSecurityProperty
> INFORMATION: Found string security property [jdk.tls.disabledAlgorithms]: SSLv3, TLSv1, TLSv1.1, DTLSv1.0, RC4, DES, MD5withRSA, DH keySize < 1024, EC keySize < 224, 3DES_EDE_CBC, anon, NULL, ECDH, TLS_RSA_*, rsa_pkcs1_sha1 usage HandshakeSignature, ecdsa_sha1 usage HandshakeSignature, dsa_sha1 usage HandshakeSignature
> Dez. 01, 2025 2:44:02 PM org.bouncycastle.jsse.provider.DisabledAlgorithmConstraints create
> WARNUNG: Ignoring unsupported entry in 'jdk.tls.disabledAlgorithms': rsa_pkcs1_sha1 usage HandshakeSignature
> Dez. 01, 2025 2:44:02 PM org.bouncycastle.jsse.provider.DisabledAlgorithmConstraints create
> WARNUNG: Ignoring unsupported entry in 'jdk.tls.disabledAlgorithms': ecdsa_sha1 usage HandshakeSignature
> Dez. 01, 2025 2...

src/java.base/share/classes/sun/net/www/protocol/https/HttpsClient.java line 478:

> 476:                             !IPAddressUtil.isIPv4LiteralAddress(host) &&
> 477:                             !(host.charAt(0) == '[' && host.charAt(host.length() - 1) == ']' &&
> 478:                                 IPAddressUtil.isIPv6LiteralAddress(host.substring(1, host.length() - 1))

The `host` value here comes from `URL.getHost()` which specifies that it returns an IPv6 address enclosed in `[]`brackets. So what you have here looks fine to me.

One additional thing I would suggest is to make this `protected String host` field of this class `final`. It currently gets assigned in the constructor of the `HttpClient` and `HttpsClient` and making this `final` would give an extra assurance that its value will always be coming from `URL.getHost()` call. 

These 2 `sun.net.www.http.HttpClient` and `HttpsClient` classes are internal to the JDK, so changing this protected field to final shouldn't cause any issues for application code.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/28577#discussion_r2580356484