Lisp data security/validation - security

This is really just a conceptual question for me at this point.
In Lisp, programs are data and data are programs. The REPL does exactly that - reads and then evaluates.
So how does one go about getting input from the user in a secure way? Obviously it's possible - I mean viaweb - now Yahoo!Stores is pretty secure, so how is it done?

The REPL stands for Read Eval Print Loop.
(loop (print (eval (read))))
Above is only conceptual, the real REPL code is much more complicated (with error handling, debugging, ...).
You can read all kinds of data in Lisp without evaluating it. Evaluation is a separate step - independent from reading data.
There are all kinds of IO functions in Lisp. The most complex of the provided functions is usually READ, which reads s-expressions. There is an option in Common Lisp which allows evaluation during READ, but that can and should be turned off when reading data.
So, data in Lisp is not necessarily a program and even if data is a program, then Lisp can read the program as data - without evaluation. A REPL should only be used by a developer and should not be exposed to arbitrary users. For getting data from users one uses the normal IO functions, including functions like READ, which can read S-expressions, but does not evaluate them.
Here are a few things one should NOT do:
use READ to read arbitrary data. READ for examples allows one to read really large data - there is no limit.
evaluate during READ ('read eval'). This should be turned off.
read symbols from I/O and call their symbol functions
read cyclical data structures with READ, when your functions expect plain lists. Walking down a cyclical list can keep your program busy for a while.
not handle syntax errors during reading from data.

You do it the way everyone else does it. You read a string of data from the stream, you parse it for your commands and parameters, you validate the commands and parameters, and you interpret the commands and parameters.
There's no magic here.
Simply put, what you DON'T do, is you don't expose your Lisp listener to an unvalidated, unsecure data source.
As was mentioned, the REPL is read - eval - print. #The Rook focused on eval (with reason), but do not discount READ. READ is a VERY powerful command in Common Lisp. The reader can evaluate code on its own, before you even GET to "eval".
Do NOT expose READ to anything you don't trust.
With enough work, you could make a custom package, limit scope of functions avaliable to that package, etc. etc. But, I think that's more work than simply writing a simple command parser myself and not worrying about some side effect that I missed.

Create your own readtable and fill with necessary hooks: SET-MACRO-CHARACTER, SET-DISPATCH-MACRO-CHARACTER et al.
Bind READTABLE to your own readtable.
Bind READ-EVAL to nil to prevent #. (may not be necessary if step 1 is done right)
READ
Probably something else.
Also there is a trick in interning symbols in temporary package while reading.
If data in not LL(1)-ish, simply write usual parser.

This is a killer question and I thought this same thing when I was reading about Lisp. Although I haven't done anything meaningful in LISP so my answer is very limited.
What I can tell you is that eval() is nasty. There is a saying that I like "If eval is the answer then you are asking the wrong question." --Unknown.
If the attacker can control data that is then evaluated then you have a very serious remote code execution vulnerability. This can be mitigated, and I'll show you an example with PHP, because that is what I know:
$id=addslashes($_GET['id']);
eval('$test="$id";');
If you weren't doing an add slashes then an attacker could get remote code execution by doing this:
http://localhost?evil_eval.php?id="; phpinfo();/*
But the add slashes will turn the " into a \", thus keeping the attacker from "breaking out" of the "data" and being able to execute code. Which is very similar to sql injection.

I found that question quit controversial. The eval wont eval your input unless you explicitly ask for it.
I mean your input will not be treat it as a LISP code but instead as a string.
Is not because that your language have powerfull concept like the eval that it is not "safe".
I think the confusion come from SQL where your actually treat an input as a [part of] SQL.
(query (concatenate 'string "SELECT * FROM foo WHERE id = " input-id))
Here input-id is being evaluate by the SQL engine.
This is because you have no nice way to write SQL, or whatever, but the point is that your input become part of what is being evaluate.
So eval don't bring you insecurity unless your are using it eyes closed.
EDIT Forgot to tell that this apply to any language.

Related

Does Lisp's treatment of code as data make it more vulnerable to security exploits than a language that doesn't treat code as data? [closed]

Closed. This question is opinion-based. It is not currently accepting answers.
Want to improve this question? Update the question so it can be answered with facts and citations by editing this post.
Closed 1 year ago.
Improve this question
I know that this might be a stupid question but I was curious.
Since Lisp treats code and data the same, does this mean that it's easier to write a payload and pass it as "innocent" data that can be used to exploit programs? In comparison to languages that don't do so?
For e.g. In python you can do something like this.
malicious_str = "print('this is a malicious string')"
user_in = eval(malicious_str)
>>> this is a malicious string
P.S I have just started learning Lisp.
No, I don't think it does. In fact because of what is normally meant by 'code is data' in Lisp, it is potentially less vulnerable than some other languages.
[Note: this answer is really about Common Lisp: see the end for a note about that.]
There are two senses in which 'code can be data' in a language.
Turning objects into executable code: eval & friends
This is the first sense. What this means is that you can, say, take a string or some other object (not all types of object, obviously) and say 'turn this into something I can execute, and do that'.
Any language that can do this has either
to be extremely careful about doing this on unconstrained data;
or to be able to be certain that a given program does not actually do this.
Plenty of languages have equivalents of eval and its relations, so plenty of languages have this problem. You give an example of Python for instance, which is a good one, and there are probably other examples in Python (I've written programs even in Python 2 which supported dynamic loading of modules at runtime, which executes potentially arbitrary code, and I think this stuff is much better integrated in Python 3).
This is also not just a property of a language: it's a property of a system. C can't do this, right? Well, yes it can if you're on any kind of reasonable *nixy platform. Not only can you use an exec-family function, but you can probably dynamically load a shared library and execute code in it.
So one solution to this problem is to, somehow, be able to be certain that a given program doesn't do this. One thing that helps is if there are a finite, known number of ways of doing it. In Common Lisp I think those are probably
eval of course;
unconstrained read (because of *read-eval*);
load;
compile;
compile-file;
and probably some others that I have forgotten.
Well, can you detect calls to those, statically, in a program? Not really: consider this:
(funcall (symbol-function (find-symbol s)) ...)
and now you're in trouble unless you have very good control over what s is: it might be "EVAL" for instance.
So that's frightening, but I don't think it's more frightening than what Python can do, for instance (almost certainly you can poke around in the namespace to find eval or something?). And something like that in a program ought to be a really big hint that bad things might happen.
I think there are probably two approaches to this, neither of which CL adopts but which implementations could (and perhaps even programs written in CL could).
One would be to be able to run programs in such a way that the finite set of bad functions above simply are disallowed: they'd signal errors if you tried to call them. An implementation could clearly do that (see below).
The other would be to have something like Perl's 'tainting' where data which came from a user needs to be explicitly looked-at by the program somehow before it's used. That doesn't guarantee safety of course, but it does make it harder to make silly mistakes: if s above came from user input and was thus tainted you'd have to explicitly say 'it's OK to use it' and, well, then it's up to you.
So this is a problem, but I don't think it's worse than the problems that very many other languages (and language-families) have.
An example of an implementation that can address the first approach is LispWorks: if you're building an application with LW, you typically create the binary with a function called deliver, which has options which allow you to remove the definitions of functions from the resulting binary whether or not the delivery process would otherwise leave them there. So, for instance
(deliver 'foo "x" 5
:functions-to-remove '(eval load compile compile-file read))
would result in an executable x which, whatever else it did, couldn't call those functions, because they're not present, at all.
Other implementations probably have similar features: I just don't know them as well.
But there's another sense in which 'code is data' in Lisp.
Program source code is available as structured data
This is the sense that people probably really mean when they say 'code is data' in Lisp, even if they don't know that. It's worth looking at your Python example again:
>>> eval("exit('exploded')")
exploded
$
So what eval eats is a string: a completely unstructured vector of characters. If you want to know whether that string contains something nasty, well, you've got a lot of work ahead of you (disclaimer: see below).
Compare this with CL:
> (let ((trying-to-be-bad "(end-the-world :now t)"))
(eval trying-to-be-bad))
"(end-the-world :now t)"
OK, so that clearly didn't end the world. And it didn't end the world because eval evaluates a bit of Lisp source code, and the value of a string, as source code, is the string.
If I want to do something nasty I have to hand it an actual interesting structure:
> (let ((actually-bad '(eval (progn
(format *query-io* "? ")
(finish-output *query-io*)
(read *query-io*)))))
(eval actually-bad))
? (defun foo () (foo))
foo
Now that's potentially quite nasty in at least several ways. But wait: in order to do this nasty thing, I had to hand eval a chunk of source code represented as an s-expression. And the structure of that s-expression is completely open to inspection by me. I can write a program which inspects this s-expression in any arbitrary way I like, and decides whether or not it is acceptable to me. That's just hugely easier than 'given this string, interpret it as a piece of source text for the language and tell me if it is dangerous':
the process of turning the sequence of characters into an s-expression has happened already;
the structure of s-expressions is both simple and standard.
So in this sense of 'code is data', Lisp is potentially much safer than other languages which have versions of eval which eat strings, like Python, say, because code is structured, standard, simple data. Lisp has an answer to the terrible 'language in a string' problem.
I am fairly sure that Python does in fact have some approach to making the parse tree available in a standard way which can be inspected. But eval still happily eats strings.
As I said above, this answer is about Common Lisp. But there are many other Lisps of course, which will have varying versions of this problem. Racket for instance probably can really fairly tightly constrain things, using sandboxed execution and modules, although I haven't explored this.
Any language can be exploited if you are not careful.
A well-known attack against Lisp is via the #. reader macro:
(read-from-string "#.(start-the-war)")
will start the war if *read-eval* is non-nil - this is why one should always bind it when reading from an un-trusted stream.
However, this is not directly related to "code is data" doctrine...

Use of unsafePerformIO appropriate?

Is using unsafePerformIO to allow read-only IO calls to non-changing files in pure code appropriate or is it going to cause a lot of problems?
The main reason is because I'd like to store them in containers and for example, make them an instance of Ord, but I can't seem to imagine how to do that without wrapping IO calls in unsafePerformIO.
On safety
Using unsafePerformIO in the way you describe should not cause any problems.
The thumb rule is: if you are using unsafePerformIO to define a function which could be defined without it in Haskell, then you are using it safely.
In your case, you essentially use it to achieve the same effect of defining some fixed values in your code. That is, you could just include your read-only non-changing files in your source code, at the cost of keeping the whole lot of data in memory. So your use is safe.
For example, if you invented a primality test which somehow exploits a fixed 100MB data table, then it would be alright to use unsafePerformIO to access an immutable file containing it. This would trade code purity for performance (memory footprint), without compromising safety.
On appropriateness
Since unsafePerformIO is indeed unsafe (the burden of proving the program safe is on you), it should be regarded as a last resort, and definitely not as the default way for reading a file's contents.
It's hard to understand whether your case really justifies using unsafePerformIO. You should describe what you are trying to achieve in more detail.
I'd guess that, if your program is going to read the files and store their whole contents in memory, then you would get no performance advantage from unsafePerformIO, and you should use pure code instead.

A language in which everything compiles

I'm trying to do some research for a new project, and I need to create objects dynamically from random data.
For this to work, I need a language / compiler that doesn't have problems with weird uncompilable code lying around.
Basically, I need the random code to compile (or be interpreted) as much as possible - Meaning that the uncompilable parts will be ignored, and only the compilable parts will create the objects (which could be ran).
Object Oriented-ness is not a must, but is a very strong advantage.
I thought of ASM, but it's very messy, and I'd probably need a more readable code
Thanks!
It sounds like you're doing something very much like genetic programming; even if you aren't, GP has to solve some of the same problems—using randomness to generate valid programs. The approach to this that is typically used is to work with a syntax tree: rather than storing x + y * 3 - 2, you store something like the following:
Then, instead of randomly changing the syntax, one can randomly change nodes in the tree instead. And if x should randomly change to, say, +, you can statically know that this means you need to insert two children (or not, depending on how you define +).
A good choice for a language to work with for this would be any Lisp dialect. In a Lisp, the above program would be written (- (+ x (* y 3)) 2), which is just a linearization of the syntax tree using parentheses to show depth. And in fact, Lisps expose this feature: you can just as easily work with the object '(- (+ x (* y 3)) 2) (note the leading quote). This is a three-element list, whose first element is -, second element is another list, and third element is 2. And, though you might or might not want it for your particular application, there's an eval function, such that (eval '(- (+ x (* y 3)) 2)) will take in the given list, treat it as a Lisp syntax tree/program, and evaluate it. This is what makes Lisps so attractive for doing this sort of work; Lisp syntax is basically a reification of the syntax-tree, and if you operate at the syntax-tree level, you can work on code as though it was a value. Lisp won't help you read /dev/random as a program directly, but with a little interpretation layered on top, you should be able to get what you want.
I should also mention, though I don't know anything about it (not that I know much about ordinary genetic programming) the existence of linear genetic programming. This is sort of like the assembly model that you mentioned—a linear stream of very, very simple instructions. The advantage here would seem to be that if you are working with /dev/random or something like it, the amount of interpretation needed is very small; the disadvantage would be, as you mentioned, the low-level nature of the code.
I'm not sure if this is what you're looking for, but any programming language can be made to function this way. For any programming language P, define the language Palways as follows:
If p is a valid program in P, then p is a valid program in Palways whose meaning is the same as its meaning in P.
If p is not a valid program in P, then p is a valid program in Palways whose meaning is the same as a program that immediately terminates.
For example, I could make the language C++always so that this program:
#include <iostream>
using namespace std;
int main() {
cout << "Hello, world!" << endl;
}
would compile as "Hello, world!", while this program:
Hahaha! This isn't legal C++ code!
Would be a legal program that just does absolutely nothing.
To solve your original problem, just take any OOP language like Java, Smalltalk, etc. and construct the appropriate Javaalways, Smalltalkalways, etc. language from it. Again, I'm not sure if this is at all what you're looking for, but it could be done very easily.
Alternatively, consider finding a grammar for any OOP language and then using that grammar to produce random syntactically valid programs. You could then filter those programs down by using the Palways programming language for that language to eliminate syntactically but not semantically valid programs.
Divide the ASCII byte values into 9 classes (division modulo 9 would help). Then assign then to Brainfuck codewords (see http://en.wikipedia.org/wiki/Brainfuck). Then interpret as Brainfuck.
There you go, any sequence of ASCII characters is a program. Not that it's going to do anything sensible... This approach has a much better chance, compared to templatetypedef's answer, to get a nontrivial program from a random byte sequence.
Text Editors
You could try feeding random character strings to an editor like Emacs or VI. Many (most?) characters will perform an editing action but some will do nothing (other than beep, perhaps). You would have to ensure that the random code mutator never generates the character sequence that exits the editor. However, this experience would be much like programming a Turing machine -- the code is not too readable.
Mathematica
In Mathematica, undefined symbols and other expressions evaluate to themselves, without error. So, that language might be a viable choice if you can arrange for the random code mutator to always generate well-formed expressions. This would be readily achievable since the basic Mathematica syntax is trivial, making it easy to operate on syntactic units rather than at the character level. It would be even easier if the mutator were written in Mathematica itself since expression-munging is Mathematica's forte. You could define a mini-language of valid operations within a Mathematica package that does not import the system-defined symbols. This would allow you to generate well-formed expressions to your heart's content without fear of generating a dangerous expression, like DeleteFile[FileNames["*.*", "/", Infinity]].
I believe Common Lisp should suit your needs. I always have some code in my SLIME/Emacs session that wouldn't compile. You can always tweak things, redefine functions in run-time. It is actually very good for prototyping.
A few years ago it took me quite a while to learn. But nowadays we have quicklisp and everything is so much easier.
Here I describe my development environment:
Install lisp on my linux machine
PS: I want to give an example, where Common Lisp was useful for me:
Up to maybe 2004 I used to write small programs in C (the keep it simple Unix way).
The last 3 years I had to get lots of different hardware running. Motorized stages, scientific cameras, IO cards.
The cameras turned out to be quite annoying. Usually you have to cool them down to -50 degree celsius or so and (in some SDKs) they don't like it when you close them. But this
is exactly how my C development cycle worked: write (30s), compile (1s), run (0.1s), repeat.
Eventually I decided to just use Common Lisp. Often it is straight forward to define the foreign function interfaces to talk to the SDKs and I can do this without ever leaving the running Lisp image. I start the editor in the morning define the open-device function, to talk to the device and after 3 hours I have enough of the functions implemented to set gain, temperature, region of interest and obtain the video.
Then I can often put the SDK manual away and just use the camera.
I used the same interactive programming approach when I have to parse some webpage or some weird XML.

Keeping State in a Purely Functional Language

I am trying to figure out how to do the following, assume that your are working on a controller for a DC motor you want to keep it spinning at a certain speed set by the user,
(def set-point (ref {:sp 90}))
(while true
(let [curr (read-speed)]
(controller #set-point curr)))
Now that set-point can change any time via a web a application, I can't think of a way to do this without using ref, so my question is how functional languages deal with this sort of thing? (even though the example is in clojure I am interested in the general idea.)
This will not answer your question but I want to show how these things are done in Clojure. It might help someone reading this later so they don't think they have to read up on monads, reactive programming or other "complicated" subjects to use Clojure.
Clojure is not a purely functional language and in this case it might be a good idea to leave the pure functions aside for a moment and model the inherent state of the system with identities.
In Clojure, you would probably use one of the reference types. There are several to choose from and knowing which one to use might be difficult. The good news is they all support the unified update model so changing the reference type later should be pretty straight forward.
I've chosen an atom but depending on your requirements it might be more appropriate to use a ref or an agent.
The motor is an identity in your program. It is a "label" for some thing that has different values at different times and these values are related to each other (i.e., the speed of the motor). I have put a :validator on the atom to ensure that the speed never drops below zero.
(def motor (atom {:speed 0} :validator (comp not neg? :speed)))
(defn add-speed [n]
(swap! motor update-in [:speed] + n))
(defn set-speed [n]
(swap! motor update-in [:speed] (constantly n)))
> (add-speed 10)
> (add-speed -8)
> (add-speed -4) ;; This will not change the state of motor
;; since the speed would drop below zero and
;; the validator does not allow that!
> (:speed #motor)
2
> (set-speed 12)
> (:speed #motor)
12
If you want to change the semantics of the motor identity you have at least two other reference types to choose from.
If you want to change the speed of the motor asynchronously you would use an agent. Then you need to change swap! with send. This would be useful if, for example, the clients adjusting the motor speed are different from the clients using the motor speed, so that it's fine for the speed to be changed "eventually".
Another option is to use a ref which would be appropriate if the motor need to coordinate with other identities in your system. If you choose this reference type you change swap! with alter. In addition, all state changes are run in a transaction with dosync to ensure that all identities in the transaction are updated atomically.
Monads are not needed to model identities and state in Clojure!
For this answer, I'm going to interpret "a purely functional language" as meaning "an ML-style language that excludes side effects" which I will interpret in turn as meaning "Haskell" which I'll interpret as meaning "GHC". None of these are strictly true, but given that you're contrasting this with a Lisp derivative and that GHC is rather prominent, I'm guessing this will still get at the heart of your question.
As always, the answer in Haskell is a bit of sleight-of-hand where access to mutable data (or anything with side effects) is structured in such a way that the type system guarantees that it will "look" pure from the inside, while producing a final program that has side effects where expected. The usual business with monads is a large part of this, but the details don't really matter and mostly distract from the issue. In practice, it just means you have to be explicit about where side effects can occur and in what order, and you're not allowed to "cheat".
Mutability primitives are generally provided by the language runtime, and accessed through functions that produce values in some monad also provided by the runtime (often IO, sometimes more specialized ones). First, let's take a look at the Clojure example you provided: it uses ref, which is described in the documentation here:
While Vars ensure safe use of mutable storage locations via thread isolation, transactional references (Refs) ensure safe shared use of mutable storage locations via a software transactional memory (STM) system. Refs are bound to a single storage location for their lifetime, and only allow mutation of that location to occur within a transaction.
Amusingly, that whole paragraph translates pretty directly to GHC Haskell. I'm guessing that "Vars" are equivalent to Haskell's MVar, while "Refs" are almost certainly equivalent to TVar as found in the stm package.
So to translate the example to Haskell, we'll need a function that creates the TVar:
setPoint :: STM (TVar Int)
setPoint = newTVar 90
...and we can use it in code like this:
updateLoop :: IO ()
updateLoop = do tvSetPoint <- atomically setPoint
sequence_ . repeat $ update tvSetPoint
where update tv = do curSpeed <- readSpeed
curSet <- atomically $ readTVar tv
controller curSet curSpeed
In actual use my code would be far more terse than that, but I've left things more verbose here in hopes of being less cryptic.
I suppose one could object that this code isn't pure and is using mutable state, but... so what? At some point a program is going to run and we'd like it to do input and output. The important thing is that we retain all the benefits of code being pure, even when using it to write code with mutable state. For instance, I've implemented an infinite loop of side effects using the repeat function; but repeat is still pure and behaves reliably and nothing I can do with it will change that.
A technique to tackle problems that apparently scream for mutability (like GUI or web applications) in a functional way is Functional Reactive Programming.
The pattern you need for this is called Monads. If you really want to get into functional programming you should try to understand what monads are used for and what they can do. As a starting point I would suggest this link.
As a short informal explanation for monads:
Monads can be seen as data + context that is passed around in your program. This is the "space suit" often used in explanations. You pass data and context around together and insert any operation into this Monad. There is usually no way to get the data back once it is inserted into the context, you just can go the other way round inserting operations, so that they handle data combined with context. This way it almost seems as if you get the data out, but if you look closely you never do.
Depending on your application the context can be almost anything. A datastructure that combines multiple entities, exceptions, optionals, or the real world (i/o-monads). In the paper linked above the context will be execution states of an algorithm, so this is quite similar to the things you have in mind.
In Erlang you could use a process to hold the value. Something like this:
holdVar(SomeVar) ->
receive %% wait for message
{From, get} -> %% if you receive a get
From ! {value, SomeVar}, %% respond with SomeVar
holdVar(SomeVar); %% recursively call holdVar
%% to start listening again
{From, {set, SomeNewVar}} -> %% if you receive a set
From ! {ok}, %% respond with ok
holdVar(SomeNewVar); %% recursively call holdVar with
%% the SomeNewVar that you received
%% in the message
end.

What does it mean for something to "compose well"?

Many a times, I've come across statements of the form
X does/doesn't compose well.
I can remember few instances that I've read recently :
Macros don't compose well (context: clojure)
Locks don't compose well (context: clojure)
Imperative programming doesn't compose well... etc.
I want to understand the implications of composability in terms of designing/reading/writing code ? Examples would be nice.
"Composing" functions basically just means sticking two or more functions together to make a big function that combines their functionality in a useful way. Essentially, you define a sequence of functions and pipe the results of each one into the next, finally giving the result of the whole process. Clojure provides the comp function to do this for you, you could do it by hand too.
Functions that you can chain with other functions in creative ways are more useful in general than functions that you can only call in certain conditions. For example, if we didn't have the last function and only had the traditional Lisp list functions, we could easily define last as (def last (comp first reverse)). Look at that — we didn't even need to defn or mention any arguments, because we're just piping the result of one function into another. This would not work if, for example, reverse took the imperative route of modifying the sequence in-place. Macros are problematic as well because you can't pass them to functions like comp or apply.
Composition in programming means assembling bigger pieces out of smaller ones.
Composition of unary functions creates a more complicated unary function by chaining simpler ones.
Composition of control flow constructs places control flow constructs inside other control flow constructs.
Composition of data structures combines multiple simpler data structures into a more complicated one.
Ideally, a composed unit works like a basic unit and you as a programmer do not need to be aware of the difference. If things fall short of the ideal, if something doesn't compose well, your composed program may not have the (intended) combined behavior of its individual pieces.
Suppose I have some simple C code.
void run_with_resource(void) {
Resource *r = create_resource();
do_some_work(r);
destroy_resource(r);
}
C facilitates compositional reasoning about control flow at the level of functions. I don't have to care about what actually happens inside do_some_work(); I know just by looking at this small function that every time a resource is created on line 2 with create_resource(), it will eventually be destroyed on line 4 by destroy_resource().
Well, not quite. What if create_resource() acquires a lock and destroy_resource() frees it? Then I have to worry about whether do_some_work acquires the same lock, which would prevent the function from finishing. What if do_some_work() calls longjmp(), and skips the end of my function entirely? Until I know what goes on in do_some_work(), I won't be able to predict the control flow of my function. We no longer have compositionality: we can no longer decompose the program into parts, reason about the parts independently, and carry our conclusions back to the whole program. This makes designing and debugging much harder and it's why people care whether something composes well.
"Bang for the Buck" - composing well implies a high ratio of expressiveness per rule-of-composition. Each macro introduces its own rules of composition. Each custom data structure does the same. Functions, especially those using general data structures have far fewer rules.
Assignment and other side effects, especially wrt concurrency have even more rules.
Think about when you write functions or methods. You create a group of functionality to do a specific task. When working in an Object Oriented language you cluster your behavior around the actions you think a distinct entity in the system will perform. Functional programs break away from this by encouraging authors to group functionality according to an abstraction. For example, the Clojure Ring library comprises a group of abstractions that cover routing in web applications.
Ring is composable where functions that describe paths in the system (routes) can be grouped into higher order functions (middlewhere). In fact, Clojure is so dynamic that it is possible (and you are encouraged) to come up with patterns of routes that can be dynamically created at runtime. This is the essence of composablilty, instead of coming up with patterns that solve a certain problem you focus on patterns that generate solutions to a certain class of problem. Builders and code generators are just two of the common patterns used in functional programming. Function programming is the art of patterns that generate other patterns (and so on and so on).
The idea is to solve a problem at its most basic level then come up with patterns or groups of the lowest level functions that solve the problem. Once you start to see patterns in the lowest level you've discovered composition. As folks discover second order patterns in groups of functions they may start to see a third level. And so on...
Composition (in the context you describe at a functional level) is typically the ability to feed one function into another cleanly and without intermediate processing. Such an example of composition is in std::cout in C++:
cout << each << item << links << on;
That is a simple example of composition which doesn't really "look" like composition.
Another example with a form more visibly compositional:
foo(bar(baz()));
Wikipedia Link
Composition is useful for readability and compactness, however chaining large collections of interlocking functions which can potentially return error codes or junk data can be hazardous (this is why it is best to minimize error code or null return values.)
Provided your functions use exceptions, or alternatively return null objects you can minimize the requirement for branching (if) on errors and maximize the compositional potential of your code at no extra risk.
Object composition (vs inheritance) is a separate issue (and not what you are asking, but it shares the name). It is one of containment to derive object hierarchy as opposed to direct inheritance.
Within the context of clojure, this comment addresses certain aspects of composability. In general, it seems to emerge when units of the system do one thing well, do not require other units to understand its internals, eschew side-effects, and accept and return the system's pervasive data structures. All of the above can be seen in M2tM's C++ example.
composability, applied to functions, means that the functions are smaller and well-defined, thus easy to integrate into other functions (i have seen this idea in the book "the joy of clojure")
the concept can apply to other things that are supposed be composed into something else.
the purpose of composability is reuse. for example, a function well-build (composable) is easier to reuse
macros aren't that well-composable because you can't pass them as parameters
lock are crap because you can't really give them names (define them well) or reuse them. you just do them inplace
imperative languages aren't that composable because (some of them, at least) don't have closures. if you want functionality passed as parameter, you're screwed. you have to build an object and pass that; disclaimer here: this last idea i'm not entirely convinced is true, therefore research more before taking it for granted
another idea on imperative languages is that they don't compose well because they imply state (from wikipedia knowledgebase :) "Imperative programming - describes computation in terms of statements that change a program state").
state does not compose well because although you have given a specific "something" in input, that "something" generates an output according to it's state. different internal state, different behaviour. and thus you can say good-bye to what you where expecting to happen.
with state, you depend to much on knowing what the current state of an object is... if you want to predict it's behavior. more stuff to keep in the back of your mind, less composable (remember well-defined ? or "small and simple", as in "easy to use" ?)
ps: thinking of learning clojure, huh ? investigating... ? good for you ! :P

Resources