In the Break It AutoLab assignment, you will find a handout with the source code for every team’s implementation from the Build It phase. Included with each implementation will be some logfiles that the compiled version generated.
Your goal in this phase is to identify bugs in these implementations. Bugs will fall into one of four categories: correctness, crash, integrity, confidentiality.
To demonstrate a bug for grading and contest purposes, you will upload a submission to the Build It assignment on AutoLab. The specification for the submission is described below. Under Deliverables, you will find details about what you must turn in for grading purposes. For both contest and grading purposes, you should only submit one break per bug you identify in a given team’s implementation. Submitting more will not net you more contest or grading points. In addition, you may submit at most 1 confidentiality and 1 integrity break against any given team.
One of the tools you use to find bugs will be Infer, a commercially used static analysis tool often used by engineers in industry. Again, see Deliverables, for more details of what you need to do with Infer for grading purposes.
You will upload a single JSON file to the AutoLab assignment. Please check the formatting of your JSON file before submitting (for instance, by using https://jsonlint.com/ to ensure that your submission is a valid JSON file). It should contain a single key, “breaks”, mapping to an array of break reports. The report format varies, as described below, based on the type of break you are reporting. Hence, at the outermost level, you JSON file should look like
{
"breaks": [
BREAK1,
BREAK2,
...,
BREAKN
]
}
Each break in turn will be a JSON dictionary (see below for examples).
All breaks must have the following keys:
A correctness violation represents a deviation from the
specification. This could take the form of invalid output, incorrect return codes, or
incorrect responses to queries. To demonstrate a correctness
violation, submit a test with the type key set to
“correctness” and the “commands” key
set to one or more command entries necessary
to demonstrate the correctness error. The elements of commands can
describe steps to build a log. The batch key may optionally be
included, which is a base64 encoded batch file.
To encode a file, most Unix/Linux/Mac systems include the base64 command,
which you can run as:
base64 -i input_file -o output_file
The value of the “batch” key will be base64 decoded and placed
in a file called “batch” when tests are executed.
You can refer to this file when specifying your list of command entries.
You must only use the batch file as an argument to the -B option.
You must not use it as the name of a logfile.
To test a correctness violation, the commands will be run with the oracle and then with the target team’s implementation. If any standard output or return codes differ, the correctness violation is confirmed. Remember that multiple break submissions against the same bug will gain you no additional points.
As an example, if you wanted to submit a testcase that is evidence of a correctness violation in team Dragon’s submission, you could add the following break to your list of “breaks” in your submission.
{
"target_team": "Dragon",
"type": "correctness",
"commands": [
{
"program": "logappend",
"args": ["-K", "secret", "-T", "0", "-G", "GERDA", "-A", "log"]
},
{
"program": "logappend",
"args": ["-K", "secret", "-T", "1", "-G", "GERDA", "-A", "-R", "16", "log"]
},
{
"program": "logread",
"args": [ "-K", "secret", "-R", "-G", "GERDA", "log"],
}
],
"batch": "LUsgc2VjcmV0IC1FIGVtcGwgLUEgbG9nCg=="
}
Note that output comparisons will generally ignore extra white space before and after the output, so correctness violations targeting such differences will not be consiered valid. In addition, if the target’s program outputs “unimplemented” in response to a valid command relating to an optional feature, this will not be considered a correctness violation.
A crash occurs when a program unexpectedly terminates due to violations of memory safety. To demonstrate a crash, submit a test with the type key set to “crash” and commands key to demonstrate the crash. The elements of commands can describe steps to build a log.
{
"target_team": "Unicorn",
"type": "crash",
"commands": [
{
"program": "logappend",
"args": ["-K", "secret", "-T", "0", "-G", "GERDA", "-A", "log"]
},
{
"program": "logappend",
"args": ["-K", "secret", "-T", "1", "-G", "GERDA", "-A", "-R", "16", "log"]
},
{
"program": "logread",
"args": [ "-K", "secret", "-R", "-G", "GERDA", "log"],
}
],
"batch": "LUsgc2VjcmV0IC1FIGVtcGwgLUEgbG9nCg=="
}
An integrity violation occurs when an attacker can successfully modify
a log without knowledge of the token used to produce that
file. Evidence of successful modification is that the modified file is
interpreted correctly by logread (when using the token that was
used to create the original log).
To demonstrate that this is possible,
you will need the logfiles in the Break It handout.
In each team’s directory, you will find an integ directory.
There you will find pairs of files, e.g., QWZFJ and QWZFJ.transcript.
The former (i.e., QWZFJ) is the logfile produced by the team’s code
when run using the commands in the corresponding .transcript file,
although we have omitted the token used in those commands.
To submit an integrity break, your break must include the “logfile” key with the name of the original log, and the “replacement” key with the base64-encoded contents of a different log file.
The “commands” key should contain a
single command entry for logread,
with the token and name of the log omitted.
A correct implementation should always exit with the output
“integrity violation”; an incorrect one will produce some other
output.
Here is an example demonstrating an integrity violation against team Mermaid.
{
"target_team": "Mermaid",
"type": "integrity",
"commands": [
{
"program": "logread",
"args": ["-R", "-G", "GERDA"]
}
],
"logfile": "QWZFJ",
"replacement": "LVQgMCAtSyBCRlVYVU1YVSAtRyBPREEgLUEgQUlDVUNMTUQKLVQgMSAtSyBCRlVYVU1YVSAtRyBPREEgLUwgQUlDVUNMTUQK"
}
This tells the testing framework to run team Mermaid’s logread against
both the original “QWZFJ” file and the “replacement” file you provide.
In both cases logread is
run using the correct authentication token. If the logread
command against “QWZFJ” logfile and against the “replacement” file
produce different non-zero length standard outs and no errors,
then an integrity violation has been detected.
A confidentiality violation occurs when an attacker can infer information about the contents of a log without knowledge of the token or transcript. Evidence of such a violation is a successful guess of an outcome of a query on the log, despite not knowing the token or the transcript used to produce it.
To demonstrate that this is possible,
as with Integrity Violations, you will need the logfiles in the Break It handout.
In each team’s directory, you will find a conf directory.
There you will find several logfiles (e.g., AJKLSV)
generated by the teams’s implementation.
In this case, we do not give you access to the token or the transcript.
To submit a confidentiality break,
the commands key should contain a
single command entry for logread,
with the token and name of the log omitted.
You must also include an “output” key
which is your prediction of what the team’s implementation
will produce when run on the log file with the token.
Our testing infrastructure will run with the token, and if the expected output is indeed produced,
the break will be considered valid.
Note that a confidentiality break is only valid
if the result of running logread is a successful
execution (i.e., the return code is 0), the output
is of non-zero length after whitespace is trimmed,
and the output is not “invalid”, “unimplemented”,
or “integrity violation”.
Here is an example of submitting a confidentiality test against team Pegasus. In this test, the submitting team knows the room information for GERDA, even though the logfile “AJKLSV” was provided with neither token nor transcript.
{
"target_team": "Pegasus",
"type": "confidentiality",
"logfile": "AJKLSV",
"commands": [
{
"program": "logread",
"args": ["-R", "-G", "GERDA"],
"output": "16"
}
]
}
For grading purposes, we will assign you implementations developed by four other teams. Your goal will be to find and validate bugs and vulnerabilities specifically in those four implementations. If one of your assigned implementations doesn’t work well enough to analyze, please request an alternate implementation from an instructor.
For points in the contest, you are welcome to also look at other implementations beyond the four you have been assigned, but you can only submit at most 10 breaks per implementation, and your breaks can only target a total of at most 8 implementations. The AutoLab infrastructure will enforce these limits.
Via Gradescope, you should submit:
Break It will be worth 100 points