Java SSL connect, add server cert to keystore programmatically
I am connecting an SSL client to my SSL server. When the client fails to verify a certificate due to the root not existing in the client's key store, I need the option to add that certificate to the local key store in code and continue. There are examples for always accepting all certificates, but I want the user to verify the cert and add it to local key store without leaving the application.
SSLSocketFactory sslsocketfactory = (SSLSocketFactory) SSLSocketFactory.getDefault();
SSLSocket sslsocket = (SSLSocket) sslsocketfactory.createSocket("localhost", 23467);
try{
sslsocket.startHandshake();
} catch (IOException e) {
//here I want to get the peer's certificate, conditionally add to local key store, then rea开发者_运维百科uthenticate successfully
}
There is a whole lot of stuff about custom SocketFactory, TrustManager, SSLContext, etc and I don't really understand how they all fit together or which would be the shortest path to my goal.
You could implement this using a X509TrustManager.
Obtain an SSLContext with
SSLContext ctx = SSLContext.getInstance("TLS");
Then initialize it with your custom X509TrustManager
by using SSLContext#init. The SecureRandom and the KeyManager[] may be null. The latter is only useful if you perform client authentication, if in your scenario only the server needs to authenticate you don't need to set it.
From this SSLContext, get your SSLSocketFactory using SSLContext#getSocketFactory and proceed as planned.
As concerns your X509TrustManager implementation, it could look like this:
TrustManager tm = new X509TrustManager() {
public void checkClientTrusted(X509Certificate[] chain,
String authType)
throws CertificateException {
//do nothing, you're the client
}
public X509Certificate[] getAcceptedIssuers() {
//also only relevant for servers
}
public void checkServerTrusted(X509Certificate[] chain,
String authType)
throws CertificateException {
/* chain[chain.length -1] is the candidate for the
* root certificate.
* Look it up to see whether it's in your list.
* If not, ask the user for permission to add it.
* If not granted, reject.
* Validate the chain using CertPathValidator and
* your list of trusted roots.
*/
}
};
Edit:
Ryan was right, I forgot to explain how to add the new root to the existing ones. Let's assume your current KeyStore of trusted roots was derived from cacerts
(the 'Java default trust store' that comes with your JDK, located under jre/lib/security). I assume you loaded that key store (it's in JKS format) with KeyStore#load(InputStream, char[]).
KeyStore ks = KeyStore.getInstance("JKS");
FileInputStream in = new FileInputStream("<path to cacerts"");
ks.load(in, "changeit".toCharArray);
The default password to cacerts
is "changeit" if you haven't, well, changed it.
Then you may add addtional trusted roots using KeyStore#setEntry. You can omit the ProtectionParameter (i.e. null), the KeyStore.Entry would be a TrustedCertificateEntry that takes the new root as parameter to its constructor.
KeyStore.Entry newEntry = new KeyStore.TrustedCertificateEntry(newRoot);
ks.setEntry("someAlias", newEntry, null);
If you'd like to persist the altered trust store at some point, you may achieve this with KeyStore#store(OutputStream, char[].
In the JSSE API (the part that takes care of SSL/TLS), checking whether a certificate is trusted doesn't necessarily involve a KeyStore
. This will be the case in the vast majority of cases, but assuming there will always be one is incorrect. This is done via a TrustManager
.
This TrustManager
is likely to use the default trust store KeyStore
or the one specified via the javax.net.ssl.trustStore
system property, but that's not necessarily the case, and this file isn't necessarily $JAVA_HOME/jre/lib/security/cacerts
(all this depends on the JRE security settings).
In the general case, you can't get hold of the KeyStore
that's used by the trust manager you're using from the application, more so if the default trust store is used without any system property settings.
Even if you were able to find out what that KeyStore
is, you would still be facing two possible problems (at least):
- You might not be able to write to that file (which is not necessarily a problem if you're not saving the changes permanently).
- This
KeyStore
might not even be file-based (e.g. it could be the Keychain on OSX) and you might not have write access to it either.
What I would suggest is to write a wrapper around the default TrustManager
(more specifically, X509TrustManager
) which performs the checks against the default trust manager and, if this initial check fails, performs a callback to the user-interface to check whether to add it to a "local" trust store.
This can be done as shown in this example (with a brief unit test), if you want to use something like jSSLutils.
Preparing your SSLContext
would be done like this:
KeyStore keyStore = // ... Create and/or load a keystore from a file
// if you want it to persist, null otherwise.
// In ServerCallbackWrappingTrustManager.CheckServerTrustedCallback
CheckServerTrustedCallback callback = new CheckServerTrustedCallback {
public boolean checkServerTrusted(X509Certificate[] chain,
String authType) {
return true; // only if the user wants to accept it.
}
}
// Without arguments, uses the default key managers and trust managers.
PKIXSSLContextFactory sslContextFactory = new PKIXSSLContextFactory();
sslContextFactory
.setTrustManagerWrapper(new ServerCallbackWrappingTrustManager.Wrapper(
callback, keyStore));
SSLContext sslContext = sslContextFactory.buildSSLContext();
SSLSocketFactory sslSocketFactory = sslContext.getSslSocketFactory();
// ...
// Use keyStore.store(...) if you want to save the resulting keystore for later use.
(Of course, you don't need to use this library and its SSLContextFactory
, but implement your own X509TrustManager
, "wrapping" or not the default one, as you prefer.)
Another factor you have to take into account is the user interaction with this callback. By the time the user has made the decision to click on accept or reject (for example), the handshake may have timed out, so you may have to make a second attempt to connect when the user has accepted the certificate.
Another point to take into account in the design of the callback is that the trust manager doesn't know which socket is being used (unlike its X509KeyManager
counterpart), so there should be as little ambiguity as to what user action caused that pop-up (or however you want to implement the callback). If multiple connections are made, you wouldn't want to validate the wrong one.
It seems possible to solve this by using a distinct callback, SSLContext and SSLSocketFactory per SSLSocket that's supposed to make a new connection, some way of tying up the SSLSocket and the callback to the action taken by the user to trigger that connection attempt in the first place.
精彩评论