Create ts-022-dpi-fragmentation.md

Author: joelanders

test-specs/ts-022-dpi-fragmentation.md

@@ -0,0 +1,69 @@
+# Specification version number
+
+2017-12-18-001
+
+# Specification name
+
+[XXX] DPI Detection Test?
+
+# Test preconditions
+
+  * An internet connection.
+  * A URL we suspect is being blocked by a stateless/non-reassembling DPI box.
+
+# Expected impact
+
+  * If a URL is being blocked by DPI (not by IP or DNS blocking) into the TCP
+    stream, we should be able to determine whether that DPI box reassembles
+    streams or if it only looks at one packet at a time.
+
+# Expected inputs
+
+  * A list of URLs to be tested (that we already know are blocked in
+    some fashion).
+
+## Semantics
+
+The test takes as input a list of URLs, one per line. For example:
+
+    http://torproject.org
+    https://ooni.nu
+
+# Test description
+
+For every hostname, we perform two HTTP connections--one with fragmentation
+and one without--and compare them to see if the response differs. If the
+input scheme is http, we fragment on the HTTP Host header; if the scheme
+is https, we fragment on the SNI header in the TLS Client Hello.
+
+# Expected output
+
+## Parent data format
+
+df-001-httpt [XXX?]
+
+## Semantics
+
+[XXX] I think there will be one boolean value for each URL input: whether
+or not fragmenting around the plaintext hostname results in a different
+HTTP response. Also, we will want to include the DNS requests
+and responses, and the full HTTP requests and responses.
+
+## Possible conclusions
+
+Determing whether or not a censorship device reassembles TCP streams can
+narrow down what type of technology is being used. For example, an HTTP
+proxy like Squid has a stream-level view of the connection, while a DPI
+box from Cisco probably does not reassemble lower-level packets into a
+stream.
+
+## Example output sample
+
+```
+{
+}
+```
+
+# Privacy considerations
+
+[XXX]