A halfway competent malloc implementation that uses doubly linked lists:
- One used list that stores all blocks currently in use.
- N free lists, each storing unused blocks of the same size, 2^N bytes large.
When an allocation request comes in, we first try to satisfy it using a block from the free list that stores blocks of a matching size. If no block can be found, we increase the size of the heap to create a new block.
No blocks from the free lists are ever released back to the OS. The heap only grows, never shrinks.
Features:
- Allocations are aligned at 16 byte boundaries.
- Blocks are sized by rounding up the size requested to the next power of 2, in order to make the blocks more reusable.
- A mutex is used to ensure exclusive access to shared data structures.
- All list operations are O(1).
The implementation is functional enough and fast enough to run most programs, both single threaded and multi threaded, both command line programs and gui programs.
Examples: ls, ps, cat, hostname, uname, gcc, javac, firefox, vim, kate:
$ ./runprogram ls -l
The code was developed in a test driven style. Run the tests with:
$ ./unittest
Portability:
- Assumes 64bit system.
- Assumes
size_tis an unsigned long (64bit).
When you start out trying to write a malloc implementation the first milestone is to get it to run on your own toy example programs. Once you've done that it's time to start testing it against real programs.
You should definitely use the LD_PRELOAD trick where you compile your malloc
into a shared library and run existing programs using it. It's almost magical
because you don't have to recompile any of them; you just insert your
malloc/free functions into their symbol search path and they will get called.
$ ulimit -c unlimited # so you can get a core file on a crash
$ LD_PRELOAD=mymalloc.so ls
You'll probably want to print from your malloc to see if it's being called or
not, but it's a mistake to use printf because that itself calls malloc, so
it's not going to work. You'll need to use a lower level function like write.
At this point it's most likely that almost no real programs will run with your
malloc. The first program I was able to run was ls (without arguments), but
not ls -l (segfault). When you get a crash run gdb <program> core to see
the stack and this may give you a clue as to what's wrong.
My first crash was in a function that was calling strcmp. If you're lucky you
might see something like this, where the clue "unaligned" is literally part of
the function name:
(gdb) bt
#0 __strcmp_sse2_unaligned () at ../sysdeps/x86_64/multiarch/strcmp-sse2-unaligned.S:32
One by one you will have to tackle these issues (not necessarily in this order):
- Make sure the address malloc returns is aligned, because some code expects to be able to read a word of memory at a time (8 bytes on a 64bit architecture), starting at an aligned address.
- Make sure the chunk you return is also aligned (ie. padded) at the end, for the same reason.
- Synchronize access to data structures that are used from more than one thread. Without this multi threaded programs will crash.
- Speed, because your malloc is most likely too slow to run interactive programs.
- Efficient memory management. This is the ultimate goal of a malloc implementation.