NPE when nonstandard HTTP method is used in request and broken connection

[kotcrab]

Version

4.2.5

Context

Sending a request with nonstandard HTTP method to a Vert.x app with Dropwizard metrics enabled always results in an NPE. It also seems to leave the connection in a broken state.

Do you have a reproducer?

Yes, see gist. Let me know if you need the entire project though this is just a basic project from the app generator with vertx-web and vertx-dropwizard-metrics.

Steps to reproduce

1. Run app, you should see Server started
2. Run command curl -X FAKE http://localhost:8080 you will see response and NPE.

SEVERE: Unhandled exception in router
java.lang.NullPointerException
	at io.vertx.ext.dropwizard.impl.HttpMetricsImpl.end(HttpMetricsImpl.java:102)
	at io.vertx.ext.dropwizard.impl.HttpServerMetricsImpl.responseEnd(HttpServerMetricsImpl.java:60)
        ...

3. This seems to leave the connection in a broken state. If you try to send any subsequent request using the same connection it will hang indefinitely.

curl -X FAKE --max-time 10 http://localhost:8080 http://localhost:8080
Hello World from Vert.x-Web!
curl: (28) Operation timed out after 10001 milliseconds with 0 bytes received

Extra

Tested on Linux and macOS. JVM 11.

[pendula95]

This is because list of supported methods is loaded here

vertx-dropwizard-metrics/src/main/java/io/vertx/ext/dropwizard/impl/HttpMetricsImpl.java

Line 54 in 4ea6d3f

methodRequests.put(method, throughputTimer(method.toString().toLowerCase() + "-requests"));

As this is not initiated NPE is thrown. This is a critical bug as a server can be rendered useless by attacker in few steps.

Maybe init the stat by method lazily so that this NPE can not fire.

[pendula95]

I was able to replicated but this (with your gist) will not affect the server, it won't get blocked.

[kotcrab]

It doesn't seem critical, the broken connection is your connection i.e. only the next sender request will hang. Though I'm not sure what happens to this connection later, maybe it can lead to resource exhaustion if you send a lot of such requests.

[mochaman]

I get this same stack trace in 4.5.1 whenever my site is scanned by CensysInspect. The cause appears different. Always happens after the following log entries:
"GET / HTTP/1.1" 200 11143 "-" "-"
"GET / HTTP/1.1" 200 11146 "-" "Mozilla/5.0 (compatible; CensysInspect/1.1; +https://about.censys.io/)"

[tsegismont]

Closed by #114