2015/AllocDebug

From CS61
Jump to: navigation, search

Computer Science 61 and E61
Systems Programming and Machine Organization
Fall 2015

Assignment 1: Debugging memory allocator

  • Assigned Thu 9/3
  • Intermediate checkin due Fri 9/11 at 11:59:59pm (see Timing, below, for more; every student should turn in an ungraded individual checkin)
  • Due Fri 9/18 at 11:59:59pm (Sat 9/19 at 11:59:59pm for extension students)
  • This assignment may be completed in pairs, although we encourage students to try the early parts individually.
  • We recommend using a Linux virtual machine for programming problem sets. See the Infrastructure page to see how to obtain and install a VM. But you can likely do this problem set on a Mac OS X machine natively. (You must have Xcode installed, including the “Command Line Tools.” We also recommend installing Homebrew or MacPorts so you can easily add software as needed.)

Goal

C programmers (that would be all of us in this class) allocate and free memory explicitly. The good news about this is that we can write fast code for modern machines, because we have full control over memory. The bad news is that it's all too easy to write programs that crash due to memory problems. But wait -- there is more good news: As systems programmers, we can build tools to help us debug memory allocation problems. For instance, in this assignment, you will transform a simple memory allocator (e.g., implementation of malloc and friends) into a debugging memory allocator.

Task

1. Transform the malloc library we give you into a debugging malloc library that:

  • tracks memory usage,
  • catches common programming errors (e.g., use after free, double free),
  • detects writing off the end of dynamically allocated memory (e.g., writing 65 bytes into a 64-byte piece of memory), and
  • catches less common, somewhat devious, programming errors.

2. Augment your debugging malloc library with "heavy hitter reporting" that tells a programmer where most of their dynamically allocated memory is allocated.

Sign up for code.seas here, and then obtain the problem set code by cloning the cs61-psets repository: https://code.seas.harvard.edu/cs61/cs61-psets

The handout code for this assignment is in the cs61-psets/pset1 directory. This contains a skeleton memory allocator that simply calls the default malloc and free library routines and many test programs.

Your code for this problem set will go in m61.c. Once you have created a working copy of the problem set code, run make check or make check-all to see how many tests your implementation passes (hint: before you've done anything, you will find that you fail all the tests). make check stops at the first error; make check-all runs all tests.

Context

C memory allocation uses two basic functions, malloc and free.

void* malloc(size_t size)
Allocates size bytes of memory and returns a pointer to the dynamically allocated memory. This memory is not initialized (it can contain anything). Returns NULL if the allocation failed (because size was too big, or memory is exhausted, or for whatever other reason).
void free(void* ptr)
Frees a single block of memory previously allocated by malloc.

The rules of malloc and free are simple: Allocate once, then free once.

  • Dynamically-allocated memory remains active until it explicitly freed with a call to free.
  • A successful call to malloc(sz) returns a pointer ptr to “new” dynamically allocated memory. This means that the sz bytes of data starting at address ptr are guaranteed not to overlap with the program's code, its global variables, its stack variables, or with any other active dynamically allocated memory.
  • The pointer argument in free(ptr) must either equal NULL or a pointer to active dynamically-allocated memory. In particular:
    • It is not OK to call free(ptr) if ptr points to the program's code, or into its global variables, or into the stack.
    • It is not OK to call free(ptr) unless ptr was returned by a previous call to malloc.
    • It is not OK to call free(ptr) if ptr is currently inactive (i.e., free(ptr) was previously called with the same pointer argument, and the ptr memory block was not reused by another malloc()).

These errors are called invalid frees. The third error is also called a double free.

Some notes on boundary cases:

  • malloc(0) may return either NULL or a non-NULL pointer. If ptr = malloc(0) is not NULL, then ptr does not overlap with any other allocation and can be passed to free().
  • free(NULL) is allowed. It does nothing.
  • malloc(sz) returns memory whose alignment works for any object. (We’ll discuss alignment in class; for a preview, see CS:APP2e §3.9.3.) On 32-bit x86 machines, this means that the address value returned by malloc() must be evenly divisible by 4. But in practice, 32-bit malloc implementations return 8-byte aligned memory, meaning that returned addresses are always evenly divisible by 8. You should do this too. (On x86-64/IA-64 machines, the maximum data alignment is 8, but malloc implementations return 16-byte aligned memory.)

Two secondary memory allocation functions are also commonly used in C, calloc and realloc. The calloc function allocates memory and clears it to zero. The realloc function can allocate, free, or resize memory depending on its arguments. These functions work like this:

 void* calloc(size_t nmemb, size_t sz) {
     void* ptr = malloc(sz * nmemb);
     if (ptr != NULL)
         memset(ptr, 0, sz * nmemb);     // clear memory to 0
     return ptr;
 }
 void* realloc(void* ptr, size_t sz) {
     void* new_ptr = NULL;
     if (sz != 0)
         new_ptr = malloc(sz);
     if (ptr != NULL && new_ptr != NULL) {
         size_t old_sz = size of memory block allocated at ptr;
         if (old_sz < sz)
             memcpy(new_ptr, ptr, old_sz);
         else
             memcpy(new_ptr, ptr, sz);
     }
     free(ptr);
     return new_ptr;
 }

(NB: There’s actually a bug in that implementation of calloc! One of our tests would find it.)

You will work on our replacements for these functions, which are called m61_malloc, m61_free, m61_calloc, and m61_realloc. Our versions of these functions simply call the system versions. Note that the m61 functions take extra arguments that the system versions don’t, namely a filename and a line number. Our header file, m61.h, uses macros to supply these arguments automatically. You’ll use filenames and line numbers to track where memory was allocated and to report where errors occur.

Assignment

Your debugging allocator must support the following functionality. Our handout code contains tests for all this functionality (though we may run other tests when grading).

  1. Overall statistics -- how many allocations/frees, how many bytes have been allocated/freed, etc
  2. Secondary allocation functions (calloc and realloc) and integer overflow protection.
  3. Invalid free detection.
  4. Writing past the beginning/end of an allocation.
  5. Reporting memory that has been allocated, but not freed.
  6. Advanced reports and checking.

This functionality is ordered from easier to harder. We encourage each student to work this out on their own. Our solution involves tens of lines of code (not hundreds), which is modest, but doing the work yourself will teach you a lot.

In addition to the debugging allocator, you must design and implement more useful functionality, namely heavy hitter reports. We want you to design your solution, implement it, and test it.

The grading breakdown:

  • 70% tested debugging allocator functionality (the test programs we hand out, plus others). If running make check reports *** All tests succeeded! you've probably got all these points.
  • 10% coding style.
  • 20% design and implementation of heavy hitter reports.

Timing

This assignment serves several purposes for us and for you.

  • It gets you back into C programming, if you are a little rusty.
  • It teaches you about byte wrangling, sizes, and the memory representations of information. This is great preparation for the rest of class.
  • It introduces you to some common memory errors and shows how careful code can defend against them.
  • You’ll learn how malloc works!
  • It shows you some new algorithms.
  • It will be fun!

We are instituting a graduated submission process to ensure that you get started on the assignment early. Only the last submission will be graded, but we may examine earlier submissions as well.

  1. First, every individual student must commit and upload an initial checkin by Friday 9/11. Try to do this work on your own. Your m61 should pass at least tests test001.c through test011.c.
  2. Second, every student or pair of students must commit and upload their final version, including heavy hitters, by Friday 9/18. Only the grade from this version will count.

In both cases, extension students’ deadlines are 24 hours after college students’ deadlines. In a team of a college student and an extension student (which we allow), the extension deadline counts.

It is important to get started early—this problem set is not trivial even though you don’t have to write that much code to make it work.

Debugging allocator (70%)

Statistics

Implement the following function:

void m61_getstatistics(struct m61_statistics* stats)
Fill in the stats structure with overall statistics about memory allocations so far.

The m61_statistics structure is defined like this:

struct m61_statistics {
    unsigned long long nactive;           // number of active allocations [#malloc - #free]
    unsigned long long active_size;       // number of bytes in active allocations
    unsigned long long ntotal;            // number of allocations, total
    unsigned long long total_size;        // number of bytes in allocations, total
    unsigned long long nfail;             // number of failed allocation attempts
    unsigned long long fail_size;         // number of bytes in failed allocation attempts
    char* heap_min;                       // smallest address in any region ever allocated
    char* heap_max;                       // largest address in any region ever allocated
};

The hard statistic to track is active_size. To track this, your free(ptr) implementation must find the number of bytes allocated for ptr.

The easiest, and probably best, way to do this is for your malloc code to allocate more space than the user requested. The first part of that space is used to store metadata about the allocation, including the allocated size. This metadata will not be a struct m61_statistics, it’ll be a structure you define yourself. Your malloc will initialize this metadata, and then return a pointer to the payload, which is the portion of the allocation following the metadata. Your free code then will take the payload pointer as input, and then use address arithmetic to calculate the pointer to the corresponding metadata. This is possible because the metadata has fixed size. From that metadata it can read the size of the allocation. See CS:APP2e Figure 9.35 “Format of a simple heap block” for an example of this type of layout.

If you don’t like this idea, you could create a list or hash table size_for_pointer that mapped pointer values to sizes. Your malloc code would add an entry to this data structure. Your free code would check this table and then remove the entry.

Other aspects of the problem set will require you to add more stuff to the metadata.

The other statistics are quite easy to track, and you should tackle the easy ones first. You can pass tests 001, 002, 003, 004, 005, 007, 008, 009, and 010 without per-allocation metadata.

Run make check to test your work. Test programs test001.c through test009.c test your overall statistics functionality. Open one of these programs and look at its code. You will notice some comments at the end of the file, such as this:

//! malloc count: active          0   total          0   fail          0
//! malloc size:  active          0   total          0   fail          0

These lines define the expected output for the test. The make check command checks your actual output against the expected output and reports any discrepancies. (It does this by calling compare.pl.)

Secondary allocation functions, integer overflow protection

Your debugging malloc library should support the secondary allocation functions calloc and realloc. It also needs to be robust against integer overflow attacks. (See, for example CS:APP2e Aside “Security vulnerability in the XDR library”, in section 2.3, p91.)

Our handout code’s m61_calloc and m61_free functions are close, but they don’t quite work. Fix them, and fix any other integer overflow errors you find.

Test programs test010.c through test015.c check your work.

Invalid free and double-free detection

m61_free(ptr, file, line) should print an error message and then call C’s abort() function when ptr does not point to active dynamically-allocated memory.

Some things to watch:

  • Be careful of calls like free((void*) 0x1), where the ptr argument is not NULL but it also doesn't point to valid memory. Your debugging malloc library should not crash when passed such a pointer. It should print an error message and exit in an orderly way. Test program test016.c checks this. (You may assume that the system malloc() allocates memory in contiguous ranges of addresses.)
  • The test programs define the desired error message format. Here's our error message for test016:
MEMORY BUG: test016.c:9: invalid free of pointer 0x7fff683fab8c, not in heap
  • Different error situations require different error messages. See test programs test016.c through test019.c.
  • Your code should print out the file name and line number of the problematic call to free().

Test programs test016.c through test021.c check your work.

Boundary write error detection

A boundary error is when a program reads or writes memory beyond the actual dimensions of an allocated memory block. An example boundary write error is to write the 11th entry in an array of size 10:

int* array = (int*) malloc(10 * sizeof(int));
...
for (int i = 0; i <= 10 /* WHOOPS */; ++i)
    array[i] = calculate(i);

These kinds of errors are relatively common in practice. (Other errors can happen, such as writing to totally random locations in memory or writing to memory before the beginning of an allocated block, rather than after its end; but after-the-end boundary writes seem most common.)

A debugging memory allocator can't detect boundary read errors, but it can detect many boundary write errors. Your m61_free(ptr, file, line) should print an error message and call abort() if it detects that the memory block associated with ptr suffered a boundary write error.

No debugging malloc could possibly detect all boundary write errors. For example, consider this:

int* array = (int*) malloc(10 * sizeof(int));
int secret = array[10];     // save boundary value
array[10] = 1384139431;     // boundary write error
array[10] = secret;         // restore old value! dmalloc can't tell there was an error!

Or this:

int* array = (int*) malloc(10 * sizeof(int));
array[200000] = 0;          // a boundary write error, but very far from the boundary!

We’re just expecting your code to catch common simple cases. You should definitely catch the case where the user writes one or more zero bytes directly after the allocated block.

Test programs test022.c through test024.c check your work.

Memory leak reporting

A memory leak happens when code allocates a block of memory but forgets to free it. Memory leaks are not as serious as other memory errors, particularly in short-running programs. They don't cause a crash directly. (The operating system always reclaims all of a program's memory when the program exits.) But in long-running programs, such as your browser, memory leaks have serious effect and are important to avoid.

Write a m61_printleakreport() function that, when called, prints a report about every allocated object in the system. This report should list every object that has been malloc()ed but not free()d. Print the report to standard output (not standard error). A report should look like this:

LEAK CHECK: test027.c:23: allocated object 0x9b811e0 with size 19
LEAK CHECK: test027.c:21: allocated object 0x9b81170 with size 17
LEAK CHECK: test027.c:20: allocated object 0x9b81140 with size 16
LEAK CHECK: test027.c:19: allocated object 0x9b81110 with size 15
LEAK CHECK: test027.c:18: allocated object 0x9b810e0 with size 14
LEAK CHECK: test027.c:16: allocated object 0x9b81080 with size 12
LEAK CHECK: test027.c:15: allocated object 0x9b81050 with size 11

A programmer would use this leak checker by calling m61_printleakreport() before exiting the program, after cleaning up all the memory they could using free() calls. Any missing free()s would show up in the leak report.

To implement a leak checker, you'll need to keep track of every active allocated block of memory. It's easiest to do this by adding more information to the block metadata. You will use the file and line arguments to m61_malloc()/m61_realloc()/m61_calloc().

Note: You may assume that the file argument to these functions has static storage duration. This means you don't need to copy the string’s contents into your block metadata—it is safe to use the string pointer.

Test programs test025.c through test027.c check your work.

Advanced reports and checking

Test programs test028.c, test029.c, and test030.c require you to update your reporting and error detection code. You’ll print better information and defend against more diabolically invalid frees. You will undoubtedly need to read the test code and understand what is being tested to defend against it.

Update your invalid free message. After determining that a pointer is invalid, your code should check whether the pointer is inside a different allocated block. This will use the same structures you created for the leak checker. If the invalid pointer is inside another block, print out that block, like so:

MEMORY BUG: test028.c:10: invalid free of pointer 0x833306c, not allocated
 test028.c:9: 0x833306c is 100 bytes inside a 2001 byte region allocated here

And make sure your invalid free detector can handle the diabolical situations in the other tests. What situations? Check the test code to find out!

Style (10%)

Your code should be clean, clear, correct, and consistent. The most important style guideline is consistency. Our handout code follows a certain format (indentation amount, comment usage, variable names). We won't grade you down if you choose a different format. But pick a format. Don't write code that changes style from line to line.

For this problem set, additional coding guidelines will be incorporated into your grade.

  • Document your metadata. Explain what additional data structures you used to track debugging and heavy-hitter information.
  • Your metadata should be compact. The debugging allocator should introduce at most tens of bytes of overhead per allocation, not hundreds or thousands.
  • Your malloc and free functions should be relatively fast. Ideally, you should add only O(1) additional work to a successful (i.e., non-erroneous) malloc or free call, but this is not a strict requirement. (Your heavy-hitter reports implementation might be more expensive than the other parts.)
  • Your implementation should not introduce memory errors of its own! Check your code by running tests under valgrind.

Heavy hitter reports (20%)

Memory allocation is one of the more expensive things a program can do. It is possible to make a program run much faster by optimizing how that program uses malloc and by optimizing malloc itself. (Did you know that both Google and Facebook have malloc specialists? Here’s Google’s tcmalloc, and Facebook liked jemalloc so much that they hired Jason Evans.)

But before optimizing a program, we must measure that program’s performance -- your (the programmer's) intuition is frequently wrong -- programmers tend to assume the slowest code is either the code they found most difficult to write or the last thing they worked on. Thus, before optimizing anything, you want to have data to guide your optimization. In this case, it useful to have a memory allocation profiler -- a tool that tracks and reports potential memory allocation problems.

Your job is to design and implement a particular kind of profiling, heavy hitter reports, for your memory allocator. This has two parts. You will:

  1. Track the heaviest users of malloc() by code location (file and line). A “heavy” location is a location that is responsible for allocating many bytes.
  2. Generate a readable report that summarizes this information.

Rule 1: If a program makes lots of allocations, and a single line of code is responsible for 20% or more of the total bytes allocated by a program, then your heavy-hitter report should mention that line of code (possibly among others).

Rule 2: Your design should handle both large numbers of allocations and large numbers of allocation sites. In particular, you should be able to handle a program that calls malloc() at 10,000 different file-line pairs.

Rule 3: Your report should include some information that helps the user decide which lines are likely to be the heaviest hitters, such as exact or estimated byte counts, or even just by ordering the hitters.

How should you implement this? That’s up to you, but here are some pointers.

  • Sampling is acceptable. It would be OK, for example, to sample 1/100th of all allocations and report information for only the sampled allocations. This can cut down the amount of data you need to store.
    • You could sample exactly every Nth allocation, but random sampling is usually better, since it avoids synchronization effects. (For instance, if the program cycled among 4 different allocation sites, then sampling every 20th allocation would miss 75% of the allocation sites!) For random sampling you'll need a source of randomness. Use random() or drand48().
  • Clever, yet easy, algorithms developed quite recently can help you catch all heavy hitters with O(1) space and simple data structures!

We provide a test program for you to test heavy hitter reports, hhtest.c. This program contains 40 different allocators that allocate regions of different sizes. Its first argument, the skew, varies the relative probabilities that each allocator is run.

You should change hhtest.c to call your heavy hitter printer just before main exits.

Running ./hhtest 0 will call every allocator with equal probability. But allocator #39 (which is at httest.c:47) allocates twice as much data as any other. So when we run our dirt-simple heavy hitter detector against ./hhtest 0, it reports:

HEAVY HITTER: hhtest.c:49: 1643786191 bytes (~50.1%)
HEAVY HITTER: hhtest.c:48: 817311692 bytes (~25.0%)
HEAVY HITTER: hhtest.c:47: 403156951 bytes (~12.4%)

(Your detector doesn’t need to follow this output format.)

If we run ./hhtest 1, however, then the first allocator (hhtest.c:10) is called twice as often as the next allocator, which is called twice as often as the next allocator, and so forth. There is almost no chance that allocator #39 is called at all. The report for ./hhtest 1 is:

HEAVY HITTER: hhtest.c:10: 499043 bytes (~50.0%)
HEAVY HITTER: hhtest.c:11: 249136 bytes (~25.0%)
HEAVY HITTER: hhtest.c:12: 123995 bytes (~12.5%)

At some intermediate skews, though, and there may be no heavy hitters at all. Our code reports nothing when run against ./hhtest 0.4.

Negative skews call the large allocators more frequently. ./hhtest -0.4:

HEAVY HITTER: hhtest.c:49: 15862542908 bytes (~62.1%)
HEAVY HITTER: hhtest.c:48: 6004585020 bytes (~23.5%)

Try ./hhtest --help to get a full description of hhtest’s arguments. You should test with many different arguments; for instance, make sure you try different allocation “phases.” A great software engineer would also create tests of her own; we encourage you to do this!

This idea can be taken quite far. Google, for example, links a heavy-hitter detector with many important servers. It is possible (within Google) to connect to many servers and generate a graph of its current heavy hitter allocation sites, including their calling functions and relationships among functions. Here's a little example (scroll down the page); here's a bigger one. Have fun!!

Extra just-for-fun opportunities

  1. Implement a particularly awesome heavy-hitter detector. For example, make your heavy-hitter detector find frequent allocations as well as heavy allocations. Allocating a single byte a billion times is expensive, even if the byte is immediately freed each time! So augment your heavy hitter detector to report call sites that are responsible for 20% or more of the total number of allocations the program made, even if those call sites aren’t responsible for many total bytes relative to other call sites.
  2. Extend your debugging allocator. For instance, include a full heap checker, which should cross-check heap data structures to see if any wild writes have corrupted internal heap state. Check with us if you're not sure whether your extension is worthwhile.

Other links

Debugging allocators have a long history. dmalloc is one of the early ones; here’s a list of some others. In fact, Mac OS and Linux’s memory allocators already turn on some debugging features by default! (We turn off those features in our GNUmakefile.)

Feel free to look at the code and documentation for other allocators to get ideas, but make sure you cite them if you do.

Turnin

You will turn in your code using Git. In the pset1/ subdirectory, you MUST include a text file called README.txt. This file should list:

  • Other collaborators and citations. (Just list "None" if you have none, but we'd be surprised if you had no citations for the heavy hitter part of the assignment.
  • Comments for the graders, if any.

Your submission will not be graded until you configure the grading server. Log in and enter your code.seas username and repository URL. Click "pset1" to initiate a fetch of your repository from code.seas. Verify that all tests that you expect to pass continue to pass on the grading server.

Notes

This lab was originally created for CS 61.