Android HTTPS(2)HttpURLConnection.getInputStream异常的原因及解决方案

Common Problems Verifying Server Certificates

InputStream in = urlConnection.getInputStream();

getInputStream(), it throws an exception:

javax.net.ssl.SSLHandshakeException: java.security.cert.CertPathValidatorException: Trust anchor for certification path not found.
        at org.apache.harmony.xnet.provider.jsse.OpenSSLSocketImpl.startHandshake(OpenSSLSocketImpl.java:374)
        at libcore.net.http.HttpConnection.setupSecureSocket(HttpConnection.java:209)
        at libcore.net.http.HttpsURLConnectionImpl$HttpsEngine.makeSslConnection(HttpsURLConnectionImpl.java:478)
        at libcore.net.http.HttpsURLConnectionImpl$HttpsEngine.connect(HttpsURLConnectionImpl.java:433)
        at libcore.net.http.HttpEngine.sendSocketRequest(HttpEngine.java:290)
        at libcore.net.http.HttpEngine.sendRequest(HttpEngine.java:240)
        at libcore.net.http.HttpURLConnectionImpl.getResponse(HttpURLConnectionImpl.java:282)
        at libcore.net.http.HttpURLConnectionImpl.getInputStream(HttpURLConnectionImpl.java:177)
        at libcore.net.http.HttpsURLConnectionImpl.getInputStream(HttpsURLConnectionImpl.java:271)

This can happen for several reasons, including:

1. The CA that issued the server certificate was unknown
2. The server certificate wasn't signed by a CA, but was self signed
3. The server configuration is missing an intermediate CA

The following sections discuss how to address these problems while keeping your connection to the server secure.

Unknown certificate authority

　　In this case, the SSLHandshakeException occurs because you have a CA that isn't trusted by the system. It could be because you have a certificate from a new CA that isn't yet trusted by Android or your app is running on an older version without the CA. More often a CA is unknown because it isn't a public CA, but a private one issued by an organization such as a government, corporation, or education institution for their own use.

　　Fortunately, you can teach HttpsURLConnection to trust a specific set of CAs. The procedure can be a little convoluted, so below is an example that takes a specific CA from an InputStream, uses it to create aKeyStore, which is then used to create and initialize a TrustManager. A TrustManager is what the system uses to validate certificates from the server and—by creating one from a KeyStore with one or more CAs—those will be the only CAs trusted by that TrustManager.

　　Given the new TrustManager, the example initializes a new SSLContext which provides anSSLSocketFactory you can use to override the default SSLSocketFactory from HttpsURLConnection. This way the connection will use your CAs for certificate validation.

　　Here is the example in full using an organizational CA from the University of Washington:

 // Load CAs from an InputStream
 // (could be from a resource or ByteArrayInputStream or ...)
 CertificateFactory cf = CertificateFactory.getInstance("X.509");
 // From https://www.washington.edu/itconnect/security/ca/load-der.crt
 InputStream caInput = new BufferedInputStream(new FileInputStream("load-der.crt"));
 Certificate ca;
 try {
     ca = cf.generateCertificate(caInput);
     System.out.println("ca=" + ((X509Certificate) ca).getSubjectDN());
 } finally {
     caInput.close();
 }

 // Create a KeyStore containing our trusted CAs
 String keyStoreType = KeyStore.getDefaultType();
 KeyStore keyStore = KeyStore.getInstance(keyStoreType);
 keyStore.load(null, null);
 keyStore.setCertificateEntry("ca", ca);

 // Create a TrustManager that trusts the CAs in our KeyStore
 String tmfAlgorithm = TrustManagerFactory.getDefaultAlgorithm();
 TrustManagerFactory tmf = TrustManagerFactory.getInstance(tmfAlgorithm);
 tmf.init(keyStore);

 // Create an SSLContext that uses our TrustManager
 SSLContext context = SSLContext.getInstance("TLS");
 context.init(null, tmf.getTrustManagers(), null);

 // Tell the URLConnection to use a SocketFactory from our SSLContext
 URL url = new URL("https://certs.cac.washington.edu/CAtest/");
 HttpsURLConnection urlConnection =
     (HttpsURLConnection)url.openConnection();
 urlConnection.setSSLSocketFactory(context.getSocketFactory());
 InputStream in = urlConnection.getInputStream();
 copyInputStreamToOutputStream(in, System.out);

　　With a custom TrustManager that knows about your CAs, the system is able to validate that your server certificate come from a trusted issuer.

Caution: Many web sites describe a poor alternative solution which is to install a TrustManager that does nothing. If you do this you might as well not be encrypting your communication, because anyone can attack your users at a public Wi-Fi hotspot by using DNS tricks to send your users' traffic through a proxy of their own that pretends to be your server. The attacker can then record passwords and other personal data. This works because the attacker can generate a certificate and—without a TrustManager that actually validates that the certificate comes from a trusted source—your app could be talking to anyone. So don't do this, not even temporarily. You can always make your app trust the issuer of the server's certificate, so just do it.

Self-signed server certificate

　　The second case of SSLHandshakeException is due to a self-signed certificate, which means the server is behaving as its own CA. This is similar to an unknown certificate authority, so you can use the same approach from the previous section.

You can create your own TrustManager, this time trusting the server certificate directly. This has all of the downsides discussed earlier of tying your app directly to a certificate, but can be done securely. However, you should be careful to make sure your self-signed certificate has a reasonably strong key. As of 2012, a 2048-bit RSA signature with an exponent of 65537 expiring yearly is acceptable. When rotating keys, you should check forrecommendations from an authority (such as NIST) about what is acceptable.

Missing intermediate certificate authority

　　The third case of SSLHandshakeException occurs due to a missing intermediate CA. Most public CAs don't sign server certificates directly. Instead, they use their main CA certificate, referred to as the root CA, to sign intermediate CAs. They do this so the root CA can be stored offline to reduce risk of compromise. However, operating systems like Android typically trust only root CAs directly, which leaves a short gap of trust between the server certificate—signed by the intermediate CA—and the certificate verifier, which knows the root CA. To solve this, the server doesn't send the client only it's certificate during the SSL handshake, but a chain of certificates from the server CA through any intermediates necessary to reach a trusted root CA.

To see what this looks like in practice, here's the mail.google.com certificate chain as viewed by the openssls_client command:

$ openssl s_client -connect mail.google.com:443
---
Certificate chain
 0 s:/C=US/ST=California/L=Mountain View/O=Google Inc/CN=mail.google.com
   i:/C=ZA/O=Thawte Consulting (Pty) Ltd./CN=Thawte SGC CA
 1 s:/C=ZA/O=Thawte Consulting (Pty) Ltd./CN=Thawte SGC CA
   i:/C=US/O=VeriSign, Inc./OU=Class 3 Public Primary Certification Authority
---

　　This shows that the server sends a certificate for mail.google.com issued by the Thawte SGC CA, which is an intermediate CA, and a second certificate for the Thawte SGC CA issued by a Verisign CA, which is the primary CA that's trusted by Android.

However, it is not uncommon to configure a server to not include the necessary intermediate CA. For example, here is a server that can cause an error in Android browsers and exceptions in Android apps:

$ openssl s_client -connect egov.uscis.gov:443
---
Certificate chain
 0 s:/C=US/ST=District Of Columbia/L=Washington/O=U.S. Department of Homeland Security/OU=United States Citizenship and Immigration Services/OU=Terms of use at www.verisign.com/rpa (c)05/CN=egov.uscis.gov
   i:/C=US/O=VeriSign, Inc./OU=VeriSign Trust Network/OU=Terms of use at https://www.verisign.com/rpa (c)10/CN=VeriSign Class 3 International Server CA - G3
---

　　What is interesting to note here is that visiting this server in most desktop browsers does not cause an error like a completely unknown CA or self-signed server certificate would cause. This is because most desktop browsers cache trusted intermediate CAs over time. Once a browser has visited and learned about an intermediate CA from one site, it won't need to have the intermediate CA included in the certificate chain the next time.

Some sites do this intentionally for secondary web servers used to serve resources. For example, they might have their main HTML page served by a server with a full certificate chain, but have servers for resources such as images, CSS, or JavaScript not include the CA, presumably to save bandwidth. Unfortunately, sometimes these servers might be providing a web service you are trying to call from your Android app, which is not as forgiving.

There are two approaches to solve this issue:

• Configure the server to include the intermediate CA in the server chain. Most CAs provide documentation on how to do this for all common web servers. This is the only approach if you need the site to work with default Android browsers at least through Android 4.2.
• Or, treat the intermediate CA like any other unknown CA, and create a TrustManager to trust it directly, as done in the previous two sections.