single.dvi
14
Interlude: Memory API
In this interlude, we discuss the memory allocation interfaces in UNIX
systems. The interfaces provided are quite simple, and hence the chapter
is short and to the point1. The main problem we address is this:
CRUX: HOW TO ALLOCATE AND MANAGE MEMORY
In UNIX/C programs, understanding how to allocate and manage
memory is critical in building robust and reliable software. What inter-
faces are commonly used? What mistakes should be avoided?
14.1 Types of Memory
In running a C program, there are two types of memory that are allo-
cated. The first is called stack memory, and allocations and deallocations
of it are managed implicitly by the compiler for you, the programmer; for
this reason it is sometimes called automatic memory.
Declaring memory on the stack in C is easy. For example, let’s say you
need some space in a function func() for an integer, called x. To declare
such a piece of memory, you just do something like this:
void func() {
int x; // declares an integer on the stack
…
}
The compiler does the rest, making sure to make space on the stack
when you call into func(). When you return from the function, the com-
piler deallocates the memory for you; thus, if you want some information
to live beyond the call invocation, you had better not leave that informa-
tion on the stack.
1Indeed, we hope all chapters are! But this one is shorter and pointier, we think.
1
2 INTERLUDE: MEMORY API
It is this need for long-lived memory that gets us to the second type
of memory, called heap memory, where all allocations and deallocations
are explicitly handled by you, the programmer. A heavy responsibility,
no doubt! And certainly the cause of many bugs. But if you are careful
and pay attention, you will use such interfaces correctly and without too
much trouble. Here is an example of how one might allocate an integer
on the heap:
void func() {
int *x = (int *) malloc(sizeof(int));
…
}
A couple of notes about this small code snippet. First, you might no-
tice that both stack and heap allocation occur on this line: first the com-
piler knows to make room for a pointer to an integer when it sees your
declaration of said pointer (int *x); subsequently, when the program
calls malloc(), it requests space for an integer on the heap; the routine
returns the address of such an integer (upon success, or NULL on failure),
which is then stored on the stack for use by the program.
Because of its explicit nature, and because of its more varied usage,
heap memory presents more challenges to both users and systems. Thus,
it is the focus of the remainder of our discussion.
14.2 The malloc() Call
The malloc() call is quite simple: you pass it a size asking for some
room on the heap, and it either succeeds and gives you back a pointer to
the newly-allocated space, or fails and returns NULL2.
The manual page shows what you need to do to use malloc; type man
malloc at the command line and you will see:
#include
…
void *malloc(size_t size);
From this information, you can see that all you need to do is include
the header file stdlib.h to use malloc. In fact, you don’t really need to
even do this, as the C library, which all C programs link with by default,
has the code for malloc() inside of it; adding the header just lets the
compiler check whether you are calling malloc() correctly (e.g., passing
the right number of arguments to it, of the right type).
The single parameter malloc() takes is of type size t which sim-
ply describes how many bytes you need. However, most programmers
do not type in a number here directly (such as 10); indeed, it would be
considered poor form to do so. Instead, various routines and macros are
2Note that NULL in C isn’t really anything special at all, just a macro for the value zero.
OPERATING
SYSTEMS
[VERSION 1.01]
WWW.OSTEP.ORG
INTERLUDE: MEMORY API 3
TIP: WHEN IN DOUBT, TRY IT OUT
If you aren’t sure how some routine or operator you are using behaves,
there is no substitute for simply trying it out and making sure it behaves
as you expect. While reading the manual pages or other documentation
is useful, how it works in practice is what matters. Write some code and
test it! That is no doubt the best way to make sure your code behaves as
you desire. Indeed, that is what we did to double-check the things we
were saying about sizeof() were actually true!
utilized. For example, to allocate space for a double-precision floating
point value, you simply do this:
double *d = (double *) malloc(sizeof(double));
Wow, that’s lot of double-ing! This invocation of malloc() uses the
sizeof() operator to request the right amount of space; in C, this is
generally thought of as a compile-time operator, meaning that the actual
size is known at compile time and thus a number (in this case, 8, for a
double) is substituted as the argument to malloc(). For this reason,
sizeof() is correctly thought of as an operator and not a function call
(a function call would take place at run time).
You can also pass in the name of a variable (and not just a type) to
sizeof(), but in some cases you may not get the desired results, so be
careful. For example, let’s look at the following code snippet:
int *x = malloc(10 * sizeof(int));
printf(“%d\n”, sizeof(x));
In the first line, we’ve declared space for an array of 10 integers, which
is fine and dandy. However, when we use sizeof() in the next line,
it returns a small value, such as 4 (on 32-bit machines) or 8 (on 64-bit
machines). The reason is that in this case, sizeof() thinks we are sim-
ply asking how big a pointer to an integer is, not how much memory we
have dynamically allocated. However, sometimes sizeof() does work
as you might expect:
int x[10];
printf(“%d\n”, sizeof(x));
In this case, there is enough static information for the compiler to
know that 40 bytes have been allocated.
Another place to be careful is with strings. When declaring space for a
string, use the following idiom: malloc(strlen(s) + 1), which gets
the length of the string using the function strlen(), and adds 1 to it
in order to make room for the end-of-string character. Using sizeof()
may lead to trouble here.
c© 2008–19, ARPACI-DUSSEAU
THREE
EASY
PIECES
4 INTERLUDE: MEMORY API
You might also notice that malloc() returns a pointer to type void.
Doing so is just the way in C to pass back an address and let the pro-
grammer decide what to do with it. The programmer further helps out
by using what is called a cast; in our example above, the programmer
casts the return type of malloc() to a pointer to a double. Casting
doesn’t really accomplish anything, other than tell the compiler and other
programmers who might be reading your code: “yeah, I know what I’m
doing.” By casting the result of malloc(), the programmer is just giving
some reassurance; the cast is not needed for the correctness.
14.3 The free() Call
As it turns out, allocating memory is the easy part of the equation;
knowing when, how, and even if to free memory is the hard part. To free
heap memory that is no longer in use, programmers simply call free():
int *x = malloc(10 * sizeof(int));
…
free(x);
The routine takes one argument, a pointer returned by malloc().
Thus, you might notice, the size of the allocated region is not passed in
by the user, and must be tracked by the memory-allocation library itself.
14.4 Common Errors
There are a number of common errors that arise in the use of malloc()
and free(). Here are some we’ve seen over and over again in teaching
the undergraduate operating systems course. All of these examples com-
pile and run with nary a peep from the compiler; while compiling a C
program is necessary to build a correct C program, it is far from suffi-
cient, as you will learn (often in the hard way).
Correct memory management has been such a problem, in fact, that
many newer languages have support for automatic memory manage-
ment. In such languages, while you call something akin to malloc()
to allocate memory (usually new or something similar to allocate a new
object), you never have to call something to free space; rather, a garbage
collector runs and figures out what memory you no longer have refer-
ences to and frees it for you.
Forgetting To Allocate Memory
Many routines expect memory to be allocated before you call them. For
example, the routine strcpy(dst, src) copies a string from a source
pointer to a destination pointer. However, if you are not careful, you
might do this:
char *src = “hello”;
char *dst; // oops! unallocated
strcpy(dst, src); // segfault and die
OPERATING
SYSTEMS
[VERSION 1.01]
WWW.OSTEP.ORG
INTERLUDE: MEMORY API 5
TIP: IT COMPILED OR IT RAN 6= IT IS CORRECT
Just because a program compiled(!) or even ran once or many times cor-
rectly does not mean the program is correct. Many events may have con-
spired to get you to a point where you believe it works, but then some-
thing changes and it stops. A common student reaction is to say (or yell)
“But it worked before!” and then blame the compiler, operating system,
hardware, or even (dare we say it) the professor. But the problem is usu-
ally right where you think it would be, in your code. Get to work and
debug it before you blame those other components.
When you run this code, it will likely lead to a segmentation fault3,
which is a fancy term for YOU DID SOMETHING WRONG WITH
MEMORY YOU FOOLISH PROGRAMMER AND I AM ANGRY.
In this case, the proper code might instead look like this:
char *src = “hello”;
char *dst = (char *) malloc(strlen(src) + 1);
strcpy(dst, src); // work properly
Alternately, you could use strdup() and make your life even easier.
Read the strdup man page for more information.
Not Allocating Enough Memory
A related error is not allocating enough memory, sometimes called a buffer
overflow. In the example above, a common error is to make almost enough
room for the destination buffer.
char *src = “hello”;
char *dst = (char *) malloc(strlen(src)); // too small!
strcpy(dst, src); // work properly
Oddly enough, depending on how malloc is implemented and many
other details, this program will often run seemingly correctly. In some
cases, when the string copy executes, it writes one byte too far past the
end of the allocated space, but in some cases this is harmless, perhaps
overwriting a variable that isn’t used anymore. In some cases, these over-
flows can be incredibly harmful, and in fact are the source of many secu-
rity vulnerabilities in systems [W06]. In other cases, the malloc library
allocated a little extra space anyhow, and thus your program actually
doesn’t scribble on some other variable’s value and works quite fine. In
even other cases, the program will indeed fault and crash. And thus we
learn another valuable lesson: even though it ran correctly once, doesn’t
mean it’s correct.
3Although it sounds arcane, you will soon learn why such an illegal memory access is
called a segmentation fault; if that isn’t incentive to read on, what is?
c© 2008–19, ARPACI-DUSSEAU
THREE
EASY
PIECES
6 INTERLUDE: MEMORY API
Forgetting to Initialize Allocated Memory
With this error, you call malloc() properly, but forget to fill in some val-
ues into your newly-allocated data type. Don’t do this! If you do forget,
your program will eventually encounter an uninitialized read, where it
reads from the heap some data of unknown value. Who knows what
might be in there? If you’re lucky, some value such that the program still
works (e.g., zero). If you’re not lucky, something random and harmful.
Forgetting To Free Memory
Another common error is known as a memory leak, and it occurs when
you forget to free memory. In long-running applications or systems (such
as the OS itself), this is a huge problem, as slowly leaking memory even-
tually leads one to run out of memory, at which point a restart is required.
Thus, in general, when you are done with a chunk of memory, you should
make sure to free it. Note that using a garbage-collected language doesn’t
help here: if you still have a reference to some chunk of memory, no
garbage collector will ever free it, and thus memory leaks remain a prob-
lem even in more modern languages.
In some cases, it may seem like not calling free() is reasonable. For
example, your program is short-lived, and will soon exit; in this case,
when the process dies, the OS will clean up all of its allocated pages and
thus no memory leak will take place per se. While this certainly “works”
(see the aside on page 7), it is probably a bad habit to develop, so be wary
of choosing such a strategy. In the long run, one of your goals as a pro-
grammer is to develop good habits; one of those habits is understanding
how you are managing memory, and (in languages like C), freeing the
blocks you have allocated. Even if you can get away with not doing so,
it is probably good to get in the habit of freeing each and every byte you
explicitly allocate.
Freeing Memory Before You Are Done With It
Sometimes a program will free memory before it is finished using it; such
a mistake is called a dangling pointer, and it, as you can guess, is also a
bad thing. The subsequent use can crash the program, or overwrite valid
memory (e.g., you called free(), but then called malloc() again to
allocate something else, which then recycles the errantly-freed memory).
Freeing Memory Repeatedly
Programs also sometimes free memory more than once; this is known as
the double free. The result of doing so is undefined. As you can imag-
ine, the memory-allocation library might get confused and do all sorts of
weird things; crashes are a common outcome.
OPERATING
SYSTEMS
[VERSION 1.01]
WWW.OSTEP.ORG
INTERLUDE: MEMORY API 7
ASIDE: WHY NO MEMORY IS LEAKED ONCE YOUR PROCESS EXITS
When you write a short-lived program, you might allocate some space
using malloc(). The program runs and is about to complete: is there
need to call free() a bunch of times just before exiting? While it seems
wrong not to, no memory will be “lost” in any real sense. The reason is
simple: there are really two levels of memory management in the system.
The first level of memory management is performed by the OS, which
hands out memory to processes when they run, and takes it back when
processes exit (or otherwise die). The second level of management
is within each process, for example within the heap when you call
malloc() and free(). Even if you fail to call free() (and thus leak
memory in the heap), the operating system will reclaim all the memory of
the process (including those pages for code, stack, and, as relevant here,
heap) when the program is finished running. No matter what the state
of your heap in your address space, the OS takes back all of those pages
when the process dies, thus ensuring that no memory is lost despite the
fact that you didn’t free it.
Thus, for short-lived programs, leaking memory often does not cause any
operational problems (though it may be considered poor form). When
you write a long-running server (such as a web server or database man-
agement system, which never exit), leaked memory is a much bigger is-
sue, and will eventually lead to a crash when the application runs out of
memory. And of course, leaking memory is an even larger issue inside
one particular program: the operating system itself. Showing us once
again: those who write the kernel code have the toughest job of all…
Calling free() Incorrectly
One last problem we discuss is the call of free() incorrectly. After all,
free() expects you only to pass to it one of the pointers you received
from malloc() earlier. When you pass in some other value, bad things
can (and do) happen. Thus, such invalid frees are dangerous and of
course should also be avoided.
Summary
As you can see, there are lots of ways to abuse memory. Because of fre-
quent errors with memory, a whole ecosphere of tools have developed to
help find such problems in your code. Check out both purify [HJ92] and
valgrind [SN05]; both are excellent at helping you locate the source of
your memory-related problems. Once you become accustomed to using
these powerful tools, you will wonder how you survived without them.
c© 2008–19, ARPACI-DUSSEAU
THREE
EASY
PIECES
8 INTERLUDE: MEMORY API
14.5 Underlying OS Support
You might have noticed that we haven’t been talking about system
calls when discussing malloc() and free(). The reason for this is sim-
ple: they are not system calls, but rather library calls. Thus the malloc li-
brary manages space within your virtual address space, but itself is built
on top of some system calls which call into the OS to ask for more mem-
ory or release some back to the system.
One such system call is called brk, which is used to change the loca-
tion of the program’s break: the location of the end of the heap. It takes
one argument (the address of the new break), and thus either increases or
decreases the size of the heap based on whether the new break is larger
or smaller than the current break. An additional call sbrk is passed an
increment but otherwise serves a similar purpose.
Note that you should never directly call either brk or sbrk. They
are used by the memory-allocation library; if you try to use them, you
will likely make something go (horribly) wrong. Stick to malloc() and
free() instead.
Finally, you can also obtain memory from the operating system via the
mmap() call. By passing in the correct arguments, mmap() can create an
anonymous memory region within your program — a region which is not
associated with any particular file but rather with swap space, something
we’ll discuss in detail later on in virtual memory. This memory can then
also be treated like a heap and managed as such. Read the manual page
of mmap() for more details.
14.6 Other Calls
There are a few other calls that the memory-allocation library sup-
ports. For example, calloc() allocates memory and also zeroes it be-
fore returning; this prevents some errors where you assume that memory
is zeroed and forget to initialize it yourself (see the paragraph on “unini-
tialized reads” above). The routine realloc() can also be useful, when
you’ve allocated space for something (say, an array), and then need to
add something to it: realloc() makes a new larger region of memory,
copies the old region into it, and returns the pointer to the new region.
14.7 Summary
We have introduced some of the APIs dealing with memory allocation.
As always, we have just covered the basics; more details are available
elsewhere. Read the C book [KR88] and Stevens [SR05] (Chapter 7) for
more information. For a cool modern paper on how to detect and correct
many of these problems automatically, see Novark et al. [N+07]; this
paper also contains a nice summary of common problems and some neat
ideas on how to find and fix them.
OPERATING
SYSTEMS
[VERSION 1.01]
WWW.OSTEP.ORG
INTERLUDE: MEMORY API 9
References
[HJ92] “Purify: Fast Detection of Memory Leaks and Access Errors” by R. Hastings, B. Joyce.
USENIX Winter ’92. The paper behind the cool Purify tool, now a commercial product.
[KR88] “The C Programming Language” by Brian Kernighan, Dennis Ritchie. Prentice-Hall
1988. The C book, by the developers of C. Read it once, do some programming, then read it again, and
then keep it near your desk or wherever you program.
[N+07] “Exterminator: Automatically Correcting Memory Errors with High Probability” by
G. Novark, E. D. Berger, B. G. Zorn. PLDI 2007, San Diego, California. A cool paper on finding
and correcting memory errors automatically, and a great overview of many common errors in C and
C++ programs. An extended version of this paper is available CACM (Volume 51, Issue 12, December
2008).
[SN05] “Using Valgrind to Detect Undefined Value Errors with Bit-precision” by J. Seward, N.
Nethercote. USENIX ’05. How to use valgrind to find certain types of errors.
[SR05] “Advanced Programming in the UNIX Environment” by W. Richard Stevens, Stephen
A. Rago. Addison-Wesley, 2005. We’ve said it before, we’ll say it again: read this book many times
and use it as a reference whenever you are in doubt. The authors are always surprised at how each time
they read something in this book, they learn something new, even after many years of C programming.
[W06] “Survey on Buffer Overflow Attacks and Countermeasures” by T. Werthman. Avail-
able: www.nds.rub.de/lehre/seminar/SS06/Werthmann BufferOverflow.pdf. A nice survey of
buffer overflows and some of the security problems they cause. Refers to many of the famous exploits.
c© 2008–19, ARPACI-DUSSEAU
THREE
EASY
PIECES
10 INTERLUDE: MEMORY API
Homework (Code)
In this homework, you will gain some familiarity with memory allo-
cation. First, you’ll write some buggy programs (fun!). Then, you’ll use
some tools to help you find the bugs you inserted. Then, you will realize
how awesome these tools are and use them in the future, thus making
yourself more happy and productive. The tools are the debugger (e.g.,
gdb) and a memory-bug detector called valgrind [SN05].
Questions
1. First, write a simple program called null.c that creates a pointer
to an integer, sets it to NULL, and then tries to dereference it. Com-
pile this into an executable called null. What happens when you
run this program?
2. Next, compile this program with symbol information included (with
the -g flag). Doing so let’s put more information into the exe-
cutable, enabling the debugger to access more useful information
about variable names and the like. Run the program under the de-
bugger by typing gdb null and then, once gdb is running, typing
run. What does gdb show you?
3. Finally, use the valgrind tool on this program. We’ll use the memcheck
tool that is a part of valgrind to analyze what happens. Run
this by typing in the following: valgrind –leak-check=yes
null. What happens when you run this? Can you interpret the
output from the tool?
4. Write a simple program that allocates memory using malloc() but
forgets to free it before exiting. What happens when this program
runs? Can you use gdb to find any problems with it? How about
valgrind (again with the –leak-check=yes flag)?
5. Write a program that creates an array of integers called data of size
100 using malloc; then, set data[100] to zero. What happens
when you run this program? What happens when you run this
program using valgrind? Is the program correct?
6. Create a program that allocates an array of integers (as above), frees
them, and then tries to print the value of one of the elements of
the array. Does the program run? What happens when you use
valgrind on it?
7. Now pass a funny value to free (e.g., a pointer in the middle of the
array you allocated above). What happens? Do you need tools to
find this type of problem?
OPERATING
SYSTEMS
[VERSION 1.01]
WWW.OSTEP.ORG
INTERLUDE: MEMORY API 11
8. Try out some of the other interfaces to memory allocation. For ex-
ample, create a simple vector-like data structure and related rou-
tines that use realloc() to manage the vector. Use an array to
store the vectors elements; when a user adds an entry to the vec-
tor, use realloc() to allocate more space for it. How well does
such a vector perform? How does it compare to a linked list? Use
valgrind to help you find bugs.
9. Spend more time and read about using gdb and valgrind. Know-
ing your tools is critical; spend the time and learn how to become
an expert debugger in the UNIX and C environment.
c© 2008–19, ARPACI-DUSSEAU
THREE
EASY
PIECES