This project is a malloc implementation written using the 42Paris' subject as reference. It is Fancy because even if I already had a working one, I decided to remake it entirely. At least it looks good to me.
Most basic function are reimplemented from scratch instead of relying on the standard library. Dependencies are limited to a few syscalls such as write or mmap.
My files also follows the current 42 Norms, which is starting to become debilitating and will probably become unecessary in the near future.
Homemade implementation of standard functions:
- malloc, free, calloc and realloc
Memory display of my structure:
- show_alloc_mem / show_alloc_mem_ex
Several personal environnment variables:
- FtMallocZone: using every power of 2 between 16 and 4096 for allocation instead of the TINY, SMALL things that the subject ask us to do;
- FtMallocLog: log the state of the malloc datastructure in a .txt file; Access permissions are not set when created by the program: created file might not have R/W rights
- FtMallocScribble: fill the freed zone with 0xAA; Causes crash (sometimes ?) when used with ./run.sh
- FtMallocPreScribble: fill a newly allocated zone with 0xAA;
- FtMallocGuard: protect malloc metadata with PROT_NONE pages.
$ export FtMallocZone=1To launch a program with this library:
LD_PRELOAD=[Library_path]/libft_malloc.so [program]
I wanted my model to handle allocation size between 1 and 2048 bytes with as little metadata as possible, using hypothesis that are largely shared among computer (but not portable according to the C Standard).
To do so, my model will allocate a zone of memory for every power of 2 between 16 and 2048*, which guarantee that at least 50% of a given memory block is actually used. My zone's metadata fit in 256 bytes, and I can fit 15 of them in a single page (the 256 bytes left are the metadata's ... metadata). I don't have alignment problem, and since my metadata is isolated from the memory allocation, I can protect it against accidental overwrite. I use between 1.025 and 2.025 bytes of metadata by address, largely because I need to keep track of the size of the allocation for show_alloc_mem.
This model is far from perfect (especially with large allocation), but I am content with what I could achieve in a 2 weeks time window. Check the image for the general idea.
* 4096 actually, because it fit in my datastructure without further development
If I want to keep my metadata segregated from my actual allocation while keeping memory alignment, I don't think fragmentation is a good solution until an allocation of more than half a page (2049 bytes on system I care about). This is beyond the scope of what I need/want to do.
Library does not seems to work with vim, bash, zsh, stopping with a segfault. The thing is, I even tried filling my block with 0 bytes before returning them, so I'm pretty sure that the block I return is valid...
Using GDB, there seems to be an issue with __GI_getenv, Using valgrind, there seems to be an issue with stpcpy.