作者:Janes
Jenkins 未授权远程代码执行漏洞, 允许攻击者将序列化的Java SignedObject对象传输给Jenkins CLI处理,反序列化ObjectInputStream作为Command对象,这将绕过基于黑名单的保护机制, 导致代码执行。
SSD的报告披露了完整的漏洞细节,作为才学JAVA的我来说,看完这份报告,依旧不清楚具体的执行流程,因此有了下文,梳理漏洞触发的具体执行流程。
触发jenkins反序列化导致代码执行的漏洞发生在使用HTTP协议实现双向通信通道的代码中,Jenkins利用此通道来接收命令。大致流程如下图:
基于HTTP建立双向Channel的入口函数位于jenkins-2.46.1/core/src/main/java/hudson/cli/CLIAction.java
文件中
@Extension @Symbol("cli")
@Restricted(NoExternalUse.class)
public class CLIAction implements UnprotectedRootAction, StaplerProxy {
private transient final Map<UUID,FullDuplexHttpChannel> duplexChannels = new HashMap<UUID, FullDuplexHttpChannel>();
......
@Override
public Object getTarget() {
StaplerRequest req = Stapler.getCurrentRequest();
if (req.getRestOfPath().length()==0 && "POST".equals(req.getMethod())) {
// CLI connection request
throw new CliEndpointResponse();
} else {
return this;
}
}
private class CliEndpointResponse extends HttpResponseException {
@Override
public void generateResponse(StaplerRequest req, StaplerResponse rsp, Object node) throws IOException, ServletException {
try {
// do not require any permission to establish a CLI connection
// the actual authentication for the connecting Channel is done by CLICommand
UUID uuid = UUID.fromString(req.getHeader("Session"));
rsp.setHeader("Hudson-Duplex",""); // set the header so that the client would know
FullDuplexHttpChannel server;
if(req.getHeader("Side").equals("download")) {
duplexChannels.put(uuid,server=new FullDuplexHttpChannel(uuid, !Jenkins.getActiveInstance().hasPermission(Jenkins.ADMINISTER)) {
@Override
protected void main(Channel channel) throws IOException, InterruptedException {
// capture the identity given by the transport, since this can be useful for SecurityRealm.createCliAuthenticator()
channel.setProperty(CLICommand.TRANSPORT_AUTHENTICATION, Jenkins.getAuthentication());
channel.setProperty(CliEntryPoint.class.getName(),new CliManagerImpl(channel));
}
});
try {
server.download(req,rsp);
} finally {
duplexChannels.remove(uuid);
}
} else {
duplexChannels.get(uuid).upload(req,rsp);
}
} catch (InterruptedException e) {
throw new IOException(e);
}
}
}
}
从上述代码可知,建立一对双向通道(download/upload), 需要发送两次POST请求,根据请求头Session字段的值uuid识别不同的双向通道,Side字段的值识别download或upload通道,请求发送的顺序是先发送download请求再发送upload请求,跟进download
函数(/Users/js/IdeaProjects/vulnhub/jenkins-2.46.1/core/src/main/java/hudson/model/FullDuplexHttpChannel.java
), 当服务器收到download请求时会阻塞请求,等待upload请求,收到upload请求后,新建Channel对象处理upload请求和返回响应,代码如下:
public synchronized void download(StaplerRequest req, StaplerResponse rsp) throws InterruptedException, IOException {
......
{// wait until we have the other channel
long end = System.currentTimeMillis() + CONNECTION_TIMEOUT;
while (upload == null && System.currentTimeMillis()<end)
wait(1000);
if (upload==null)
throw new IOException("HTTP full-duplex channel timeout: "+uuid);
}
try {
channel = new Channel("HTTP full-duplex channel " + uuid,
Computer.threadPoolForRemoting, Mode.BINARY, upload, out, null, restricted);
......
} finally {
// publish that we are done
completed=true;
notify();
}
}
public synchronized void upload(StaplerRequest req, StaplerResponse rsp) throws InterruptedException, IOException {
rsp.setStatus(HttpServletResponse.SC_OK);
InputStream in = req.getInputStream();
if(DIY_CHUNKING) in = new ChunkedInputStream(in);
// publish the upload channel
upload = in;
notify();
// wait until we are done
while (!completed)
wait();
}
以上就是建立双向通道的基本过程。
upload请求作为输入流实例化Channel对象(~/.m2/repository/org/jenkins-ci/main/remoting/3.7/remoting-3.7-sources.jar!/hudson/remoting/Channel.java
), Channel类的构造链比较繁琐如下图,
最终调用的构造方法为Channel(ChannelBuilder settings, CommandTransport transport)
, 该构造方法的transport参数,由ChannelBuilder类的negotiate()方法获得。
protected CommandTransport negotiate(final InputStream is, final OutputStream os) throws IOException {
......
{// read the input until we hit preamble
Mode[] modes={Mode.BINARY,Mode.TEXT};
byte[][] preambles = new byte[][]{Mode.BINARY.preamble, Mode.TEXT.preamble, Capability.PREAMBLE};
int[] ptr=new int[3];
Capability cap = new Capability(0); // remote capacity that we obtained. If we don't hear from remote, assume no capability
while(true) {
int ch = is.read();
......
for(int i=0;i<preambles.length;i++) {
byte[] preamble = preambles[i];
if(preamble[ptr[i]]==ch) {
if(++ptr[i]==preamble.length) {
switch (i) {
case 0:
case 1:
......
return makeTransport(is, os, mode, cap);
case 2:
cap = Capability.read(is);
negotiate()会检查输入(upload请求)的前导码, 所有发往Jenkins CLI的命令中都包含某种格式的前导码(preamble),前导码格式通常为:<===[JENKINS REMOTING CAPACITY]===>rO0ABXNyABpodWRzb24ucmVtb3RpbmcuQ2FwYWJpbGl0eQAAAAAAAAABAgABSgAEbWFza3hwAAAAAAAAAH4=
, 该前导码包含一个经过base64编码的序列化对象。“Capability”类型的序列化对象的功能是告诉服务器客户端具备哪些具体功能(比如HTTP分块编码功能)。
最后调用makeTransport()方法返回CommandTransport
对象, 根据cap是否支持Chunking
返回不同的对象ChunkedCommandTransport
或ClassicCommandTransport
。
protected CommandTransport makeTransport(InputStream is, OutputStream os, Mode mode, Capability cap) throws IOException {
FlightRecorderInputStream fis = new FlightRecorderInputStream(is);
if (cap.supportsChunking())
return new ChunkedCommandTransport(cap, mode.wrap(fis), mode.wrap(os), os);
else {
ObjectOutputStream oos = new ObjectOutputStream(mode.wrap(os));
oos.flush(); // make sure that stream preamble is sent to the other end. avoids dead-lock
return new ClassicCommandTransport(
new ObjectInputStreamEx(mode.wrap(fis),getBaseLoader(),getClassFilter()),
oos,fis,os,cap);
}
}
利用SSD的PoC脚本发送的upload请求返回的是ClassicCommandTransport
对象,其继承关系如下图所示。
Channel构造函数Channel(ChannelBuilder settings, CommandTransport transport)
中, transport.setup()调用SynchronousCommandTransport类的setup()方法来启动一个ReaderThread线程。
public void setup(Channel channel, CommandReceiver receiver) {
this.channel = channel;
new ReaderThread(receiver).start();
}
通过上面的ReaderThread.start()方法启动一个线程,ReaderThread为SynchronousCommandTransport类的内部类,在run()方法中,调用ClassicCommandTransport
类的read()方法读取输入,read()方法实际是调用Command类的readFrom()方法读取,通过反序列化输入返回一个Command对象。
private final class ReaderThread extends Thread {
......
public ReaderThread(CommandReceiver receiver) {
super("Channel reader thread: "+channel.getName());
this.receiver = receiver;
}
@Override
public void run() {
final String name =channel.getName();
try {
while(!channel.isInClosed()) {
Command cmd = null;
try {
cmd = read();
public final Command read() throws IOException, ClassNotFoundException {
try {
Command cmd = Command.readFrom(channel, ois);
在反序列化输入返回一个Command对象时就执行了cmd命令,而不是通过正常的回调handle()方法执行cmd命令,反序列化导致的执行代码触发的相关异常如下:
类型转换异常ClassCastException
: org.apache.commons.collections.map.ReferenceMap cannot be cast to hudson.remoting.Command
.
虽说反序列化时就执行了cmd代码,这里也顺带了解下正常的执行cmd的过程。SynchronousCommandTransport类的run()方法中,获得返回的Command对象(cmd),然后调用receiver.handle(cmd);
处理命令,其实质是回调Channel类构造方法里面的handle方法,而传入handle方法的cmd参数就是反序列化得到的Command对象。
transport.setup(this, new CommandReceiver() {
public void handle(Command cmd) {
......
try {
cmd.execute(Channel.this);
上面过程主要讲述的是漏洞触发的流程,而该漏洞的核心是反序列化Java SignedObject对象会绕过黑名单保护机制,从而导致的代码执行漏洞。
ClassFilter类定义的默认的黑名单如下:
private static final String[] DEFAULT_PATTERNS = {
"^bsh[.].*",
"^com[.]google[.]inject[.].*",
"^com[.]mchange[.]v2[.]c3p0[.].*",
"^com[.]sun[.]jndi[.].*",
"^com[.]sun[.]corba[.].*",
"^com[.]sun[.]javafx[.].*",
"^com[.]sun[.]org[.]apache[.]regex[.]internal[.].*",
"^java[.]awt[.].*",
"^java[.]rmi[.].*",
"^javax[.]management[.].*",
"^javax[.]naming[.].*",
"^javax[.]script[.].*",
"^javax[.]swing[.].*",
"^org[.]apache[.]commons[.]beanutils[.].*",
"^org[.]apache[.]commons[.]collections[.]functors[.].*",
"^org[.]apache[.]myfaces[.].*",
"^org[.]apache[.]wicket[.].*",
".*org[.]apache[.]xalan.*",
"^org[.]codehaus[.]groovy[.]runtime[.].*",
"^org[.]hibernate[.].*",
"^org[.]python[.].*",
"^org[.]springframework[.](?!(\\p{Alnum}+[.])*\\p{Alnum}*Exception$).*",
"^sun[.]rmi[.].*",
"^javax[.]imageio[.].*",
"^java[.]util[.]ServiceLoader$",
"^java[.]net[.]URLClassLoader$"
};
黑名单机制绕过可以通过分析补丁得到印证。