4. Tutorial: A Real Analyzer
In this chapter we will develop a simple protocol analyzer from scratch. Our analyzer will parse the Trivial File Transfer Protocol (TFTP) in its original incarnation, as described in RFC 1350. TFTP provides a small protocol for copying files from a server to a client system. It is most commonly used these days for providing boot images to devices during initialization. The protocol is sufficiently simple that we can walk through it end to end. See its Wikipedia page for more background.
Contents
4.1. Creating a Spicy Grammar
We start by developing Spicy grammar for TFTP. The protocol is packet-based, and our grammar will parse the content of one TFTP packet at a time. While TFTP is running on top of UDP, we will Spicy parse just the actual UDP application-layer payload, as described in Section 5 of the protocol standard.
4.1.1. Parsing One Packet Type
TFTP is a binary protocol that uses a set of standardized, numerical opcodes to distinguish between different types of packets—a common idiom with such protocols. Each packet contains the opcode inside the first two bytes of the UDP payload, followed by further fields that then differ by type. For example, the following is the format of a TFTP “Read Request” (RRQ) that initiates a download from a server:
2 bytes string 1 byte string 1 byte (from RFC 1350)
------------------------------------------------
| Opcode | Filename | 0 | Mode | 0 |
------------------------------------------------
A Read Request uses an opcode of 1. The filename is a sequence of
ASCII bytes terminated by a null byte. The mode is another
null-terminated byte sequence that usually is either netascii,
octet, or mail, describing the desired encoding for data that
will be received.
Let’s stay with the Read Request for a little bit and write a Spicy parser just for this one packet type. The following is a minimal Spicy unit to parse the three fields:
module TFTP; # [1]
public type ReadRequest = unit { # [2]
opcode: uint16; # [3]
filename: bytes &until=b"\x00"; # [4]
mode: bytes &until=b"\x00"; # [5]
on %done { print self; } # [6]
};
Let’s walk through:
[1]All Spicy source files must start with amoduleline defining a namespace for their content. By convention, the namespace should match what is being parsed, so we call oursTFTP. Naming our moduleTFTPalso implies saving it under the nametftp.spicy, so that other modules can find it throughimport TFTP;. See Modules for more on all of this.
[2]In Spicy, one will typically create aunittype for each of the main data units that a protocol defines. We want to parse a Read Request, so we call our type accordingly. We declare it as public because we want to use this unit as the starting point for parsing data. The following lines then lay out the elements of such a request in the same order as the protocol defines them.
[3]Per the TFTP specification, the first field contains theopcodeas an integer value encoded over two bytes. For multi-byte integer values, it is important to consider the byte order for parsing. TFTP uses network byte order which matches Spicy’s default, so there is nothing else for us to do here. (If we had to specify the order, we would add the &byte-order attribute).
[4]The filename is a null-terminated byte sequence, which we can express directly as such in Spicy: Thefilenamefield will accumulate bytes until a null byte is encountered. Note that even though the specification of a Read Request shows the0as separate element inside the packet, we don’t create a field for it, but rather exploit it as a terminator for the file name (which will not be included into thefilenamestored).
[5]Themodeoperates just the same as thefilename.
[6]Once we are done parsing a Read Request, we print out the result for debugging.
We should now be able to parse a Read Request. To try it, we need the
actual payload of a corresponding packet. With TFTP, the format is
simple enough that we can start by faking data with printf
and pipe that into the Spicy tool spicy-driver:
# printf '\000\001rfc1350.txt\000octet\000' | spicy-driver tftp.spicy
[$opcode=1, $filename=b"rfc1350.txt", $mode=b"octet"]
Here, spicy-driver compiles our ReadRequest unit into an
executable parser and then feeds it with the data it is receiving on
standard input. The output of spicy-driver is the result of our
print statement executing at the end.
What would we do with a more complex protocol where we cannot easily use
printf to create some dummy payload? We would probably have access
to some protocol traffic in pcap traces, however we can’t just feed
those into spicy-driver directly as they will contain all the
other network layers as well that our grammar does not handle (e.g.,
IP and UDP). One way to test with a trace would be proceeding with
Zeek integration at this point, so that we could let Zeek strip off
the base layers and then feed our parser only the TFTP payload.
However, during development it is often easier at first to extract
application-layer protocol data from the traces ourselves, write it
into files, and then feed those files into spicy-driver.
We can leverage Zeek for doing this extraction into files. If we had a
TCP-based protocol, doing so would be trivial because Zeek has that
functionality built in: When you run Zeek on a pcap trace and add
Conn::default_extract=T to the command line, it will write out all
the TCP streams into individual files. As TFTP is UDP-based, however,
we will use a custom script, udp-contents.zeek. When you run Zeek with that script on
trace, you will get one file per UDP packet each containing the
corresponding application-layer UDP payload (make sure to use this
with small traces only …).
Let’s use the UDP script with tftp_rrq.pcap, a tiny TFTP trace containing a single file
download from Wireshark’s pcap archive. tcpdump shows
us that the first packet indeed contains a Read Request:
# tcpdump -ttnr tftp_rrq.pcap
1367411051.972852 IP 192.168.0.253.50618 > 192.168.0.10.69: 20 RRQ "rfc1350.txtoctet" [\|tftp]
1367411052.077243 IP 192.168.0.10.3445 > 192.168.0.253.50618: UDP, length 516
1367411052.081790 IP 192.168.0.253.50618 > 192.168.0.10.3445: UDP, length 4
[...]
Running Zeek on the trace with the udp-contents scripts produces
the expected content files:
# zeek -r tftp_rrq.pcap udp-contents
# ls udp-contents.orig.*
udp-contents.orig.1367411051.972852.dat
udp-contents.orig.1367411052.077243.dat
udp-contents.orig.1367411052.086300.dat
udp-contents.orig.1367411052.088995.dat
udp-contents.orig.1367411052.091675.dat
[...]
Per the timestamps included with the names, the first file is the one containing our Read Request. We can pass that into our Spicy parser:
# cat udp-contents.orig.1367411051.972852.dat | spicy-driver tftp.spicy
[$opcode=1, $filename=b"rfc1350.txt", $mode=b"octet"]
That gives us an easy way to test our TFTP parser.
4.1.2. Generalizing to More Packet Types
So far we can parse a Read Request, but nothing else. In fact, we are
not even examining the opcode yet at all to see if our input
actually is a Read Request. To generalize our grammar to other TFTP
packet types, we will need to parse the opcode on its own first,
and then use the value to decide how to handle subsequent data. Let’s
start over with a minimal version of our TFTP grammar that looks at
just the opcode:
module TFTP;
public type Packet = unit {
opcode: uint16;
on %done { print self; }
};
# cat udp-contents.orig.1367411051.972852.dat | spicy-driver tftp.spicy
[$opcode=1]
Next we create a separate type to parse the fields that are specific to a Read Request:
type ReadRequest = unit {
filename: bytes &until=b"\x00";
mode: bytes &until=b"\x00";
};
We do not declare this type as public because we will use it only
internally inside our grammar; it is not a top-level entry point for
parsing (that’s Packet now).
Now we need to tie the two units together. We can do that by adding
the ReadRequest as a field to the Packet, which will let Spicy
parse it as a sub-unit:
module TFTP;
public type Packet = unit {
opcode: uint16;
rrq: ReadRequest;
on %done { print self; }
};
# cat udp-contents.orig.1367411051.972852.dat | spicy-driver tftp.spicy
[$opcode=1, $rrq=[$filename=b"rfc1350.txt", $mode=b"octet"]]
However, this does not help us much yet: it still resembles our
original version in that it continues to hardcode one specific packet
type. But the direction of using sub-units is promising, we only need
to instruct the parser to leverage the opcode to decide what
particular sub-unit to use. Spicy provides a switch construct for
such dispatching:
module TFTP;
public type Packet = unit {
opcode: uint16;
switch ( self.opcode ) {
1 -> rrq: ReadRequest;
};
on %done { print self; }
};
# cat udp-contents.orig.1367411051.972852.dat | spicy-driver tftp.spicy
[$opcode=1, $rrq=[$filename=b"rfc1350.txt", $mode=b"octet"]]
The self keyword always refers to the unit instance currently
being parsed, and we use that to get to the opcode for switching on.
If it is 1, we descend down into a Read Request.
What happens if it is something other than 1? Let’s try it with
the first server-side packet, which contains a TFTP acknowledgment
(opcode 4):
# cat udp-contents.resp.1367411052.081790.dat | spicy-driver tftp.spicy
[fatal error] terminating with uncaught exception of type spicy::rt::ParseError: parse error: no matching case in switch statement (:7:5-9:7)
Of course it is now easy to add another unit type for handling such acknowledgments:
public type Packet = unit {
opcode: uint16;
switch ( self.opcode ) {
1 -> rrq: ReadRequest;
4 -> ack: Acknowledgement;
};
on %done { print self; }
};
type Acknowledgement = unit {
num: uint16; # block number being acknowledged
};
# cat udp-contents.resp.1367411052.081790.dat | spicy-driver tftp.spicy
[$opcode=4, $rrq=(not set), $ack=[$num=1]]
As expected, the output shows that our TFTP parser now descended into
the ack sub-unit while leaving rrq unset.
TFTP defines three more opcodes for other packet types: 2 is a
Write Request, 3 is file data being sent, and 5 is an error.
We will add these to our grammar as well, so that we get the whole
protocol covered (please refer to the RFC for specifics of each packet
type):
module TFTP;
public type Packet = unit {
opcode: uint16;
switch ( self.opcode ) {
1 -> rrq: ReadRequest;
2 -> wrq: WriteRequest;
3 -> data: Data;
4 -> ack: Acknowledgement;
5 -> error: Error;
};
on %done { print self; }
};
type ReadRequest = unit {
filename: bytes &until=b"\x00";
mode: bytes &until=b"\x00";
};
type WriteRequest = unit {
filename: bytes &until=b"\x00";
mode: bytes &until=b"\x00";
};
type Data = unit {
num: uint16;
data: bytes &eod; # parse until end of data (i.e., packet) is reached
};
type Acknowledgement = unit {
num: uint16;
};
type Error = unit {
code: uint16;
msg: bytes &until=b"\x00";
};
This grammar works well already, but we can improve it a bit more.
4.1.3. Using Enums
The use of integer values inside the switch construct is not
exactly pretty: they are hard to read and maintain. We can improve our
grammar by using an enumerator type with descriptive labels instead.
We first declare an enum type that provides one label for each
possible opcode:
type Opcode = enum { RRQ = 1, WRQ = 2, DATA = 3, ACK = 4, ERROR = 5 };
Now we can change the switch to look like this:
switch ( self.opcode ) {
Opcode::RRQ -> rrq: ReadRequest;
Opcode::WRQ -> wrq: WriteRequest;
Opcode::DATA -> data: Data;
Opcode::ACK -> ack: Acknowledgement;
Opcode::ERROR -> error: Error;
};
Much better, but there is a catch still: this will not compile because
of a type mismatch. The switch cases’ expressions have type
Opcode, but self.opcode remains of type uint16. That is
because Spicy cannot know on its own that the integers we parse into
opcode match the numerical values of the Opcode labels. But
we can convert the former into the latter explicitly by adding a
&convert attribute to the opcode field:
public type Packet = unit {
opcode: uint16 &convert=Opcode($$);
...
};
This does two things:
Each time an
uint16gets parsed for this field, it is not directly stored inopcode, but instead first passed through the expression that&convertspecifies. Spicy then stores the result of that expression, potentially adapting the field’s type accordingly. Inside the&convertexpression, the parsed value is accessible through the special identifier$$.Our
&convertexpression passes the parsed integer into the constructor for theOpcodeenumerator type, which lets Spicy create anOpcodevalue with the label that corresponds to the integer value.
With this transformation, the opcode field now has type Opcode
and hence can be used with our updated switch statement. You can see
the new type for opcode in the output as well:
# cat udp-contents.orig.1367411051.972852.dat | spicy-driver tftp.spicy
[$opcode=Opcode::RRQ, $rrq=[$filename=b"rfc1350.txt", $mode=b"octet"], $wrq=(not set), $data=(not set), $ack=(not set), $error=(not set)]
See On-the-fly Type Conversion with &convert for more on &convert, and
Enum for more on the enum type.
Note
What happens when Opcode($$) receives an integer that does not
correspond to any of the labels? Spicy permits that and will
substitute an implicitly defined Opcode::Undef label. It will
also retain the actual integer value, which can be recovered by
converting the enum value back to an integer.
4.1.4. Using Unit Parameters
Looking at the two types ReadRequest and WriteRequest, we see
that both are using exactly the same fields. That means we do not
really need two separate types here, and could instead define a
single Request unit to cover both cases. Doing so is
straight-forward, except for one issue: when parsing such a
Request, we would now lose the information whether we are seeing
read or a write operation. For a potential Zeek integration later it will be
useful to retain that distinction, so let us leverage a Spicy
capability that allows passing state into a sub-unit: unit
parameters. Here’s the corresponding excerpt after
that refactoring:
public type Packet = unit {
opcode: uint16 &convert=Opcode($$);
switch ( self.opcode ) {
Opcode::RRQ -> rrq: Request(True);
Opcode::WRQ -> wrq: Request(False);
# ...
};
on %done { print self; }
};
type Request = unit(is_read: bool) {
filename: bytes &until=b"\x00";
mode: bytes &until=b"\x00";
on %done { print "We got a %s request." % (is_read ? "read" : "write"); }
};
We see that the switch now passes either True or False
into the Request type, depending on whether it is a Read Request
or Write Request. For demonstration, we added another print
statement, so that we can see how that boolean becomes available
through the is_read unit parameter:
# cat udp-contents.orig.1367411051.972852.dat | spicy-driver tftp.spicy
We got a read request.
[$opcode=Opcode::RRQ, $rrq=[$filename=b"rfc1350.txt", $mode=b"octet"], $wrq=(not set), $data=(not set), $ack=(not set), $error=(not set)]
Admittedly, the unit parameter is almost overkill in this
example, but it proves very useful in more complex grammars where one
needs access to state information, in particular also from
higher-level units. For example, if the Packet type stored
additional state that sub-units needed access to, they could receive
the Packet itself as a parameter.
4.1.5. Complete Grammar
Combining everything discussed so far, this leaves us with the following complete grammar for TFTP, including the packet formats in comments as well:
# Copyright (c) 2021 by the Zeek Project. See LICENSE for details.
#
# Trivial File Transfer Protocol
#
# Specs from https://tools.ietf.org/html/rfc1350
module TFTP;
import spicy;
# Common header for all messages:
#
# 2 bytes
# ---------------
# | TFTP Opcode |
# ---------------
public type Packet = unit { # public top-level entry point for parsing
op: uint16 &convert=Opcode($$);
switch ( self.op ) {
Opcode::RRQ -> rrq: Request(True);
Opcode::WRQ -> wrq: Request(False);
Opcode::DATA -> data: Data;
Opcode::ACK -> ack: Acknowledgement;
Opcode::ERROR -> error: Error;
};
};
# TFTP supports five types of packets [...]:
#
# opcode operation
# 1 Read request (RRQ)
# 2 Write request (WRQ)
# 3 Data (DATA)
# 4 Acknowledgment (ACK)
# 5 Error (ERROR)
type Opcode = enum {
RRQ = 0x01,
WRQ = 0x02,
DATA = 0x03,
ACK = 0x04,
ERROR = 0x05
};
# Figure 5-1: RRQ/WRQ packet
#
# 2 bytes string 1 byte string 1 byte
# ------------------------------------------------
# | Opcode | Filename | 0 | Mode | 0 |
# ------------------------------------------------
type Request = unit(is_read: bool) {
filename: bytes &until=b"\x00";
mode: bytes &until=b"\x00";
};
# Figure 5-2: DATA packet
#
# 2 bytes 2 bytes n bytes
# ----------------------------------
# | Opcode | Block # | Data |
# ----------------------------------
type Data = unit {
num: uint16;
data: bytes &eod;
};
# Figure 5-3: ACK packet
#
# 2 bytes 2 bytes
# ---------------------
# | Opcode | Block # |
# ---------------------
type Acknowledgement = unit {
num: uint16;
};
# Figure 5-4: ERROR packet
#
# 2 bytes 2 bytes string 1 byte
# -----------------------------------------
# | Opcode | ErrorCode | ErrMsg | 0 |
# -----------------------------------------
type Error = unit {
code: uint16;
msg: bytes &until=b"\x00";
};
4.2. Next Steps
This tutorial provides an introduction to the Spicy language and toolchain. Spicy’s capabilities go much further than what we could show here. Some pointers for what to look at next:
Programming in Spicy provides an in-depth discussion of the Spicy language, including in particular all the constructs for parsing data and a reference of language elements. Note that most of Spicy’s types come with operators and methods for operating on values. The Debugging section helps understanding Spicy’s operation if results do not match what you would expect.
Examples summarizes grammars coming with the Spicy distribution.
Zeek’s Spicy tutorial continues the TFTP example by turning the Spicy code developed here into a full Zeek analyzer.
Zeek Integration discusses Spicy’s integration into Zeek.