Upon inspection all inodes appear to have atleast 2 references

Program bug exposed this:
[vi:16] poll(BFBFEF27, 80D11B8, 64, 0) =
yieldig a call to
heapmm_alloc(0x40688c00)

The following posnk original user projects are still in the kernel repo:

• All GUI applications
• Old shell
• Test cases

The following third party projects are still in the kernel repo

• libvterm
• klange's nyancat

Serial output doesn't reveal much:

Debugger console up on ttyS0
Bootloader VBE info @C0010558 : {mode: 324,  lfb: E0000000, w:1024, h: 768, bpp: 32}
info: registered handler for IRQ1!
info: registered handler for IRQ12!
Enumerating PCI...
0:0.0 - Intel Corporation PCI & Memory NO DRIVER FOUND!!!
0:1.0 - Intel Corporation PIIX3 PCI-to-ISA Bridge (Triton II) NO DRIVER FOUND!!!
ata_pci: initializing controller 0:1.1 (0, 0, 0, 0, D000)->IRQ0
info: registered handler for IRQ14!

A jump to address 0x00000000 occurs when sys_execve fails on ARMv7.

Debug log:

[init:4] execve(BFBBEF67, BFBBEFB7, 4020000, 8066440) = 
error loading elf /busybox
sending signal to pid 4 : 12
[0x0] page fault in 4 @ 0xBFFFFFEC (P:0, W:0, U:0) 
Userland exception: Unresolved page fault(10), sending SIGSEGV!
Exception occurred at 0xBFFFFFEC
Register dump: 
sending signal to pid 4 : 11
sighandle: 4
handling signal: 6144
killedby: 6144
4 (Errno: 0)

No filesystem synchronization has been implemented.
On shutdown any modified inodes are lost and the superblock is not updated either

ext2_link yields a directory "file" that has a length greater than the number of bytes stored for it, resulting in a sparse file yielding all-zero dirents

ext2 and probably other fses require variable size dirent structure, which is supported but unusable due to the layout of the fields, suggested fix:
move name field to the end of the struct

Something is corrupting memory around block cache structures

When the inode store operation blocks the inode may be evicted again causing two processes to race in the evictor function thereby causing one to continue trying to remove an nonexistent inode object.

The current build system does not support a seperate output directory, this results in the pollution of the kernel source tree with binaries and object files.

Readdir does not correctly skip deleted dirents for ext2

sbrk() throws ENOMEM on a newly spawned process even though the system has enough memory available.

Debug log:
[init:1] fork(0, 0, 0, 0) = 3 (Errno: 0)
[init:3] sbrk(20, 0, 0, 0) = FFFFFFFF (Errno: 12)
[init:3] sbrk(20, 0, 0, 0) = FFFFFFFF (Errno: 12)
[init:1] sbrk(418, 0, 0, 0) = 8066000 (Errno: 0)

At the moment the configuration header does not automatically provide the correct definitions for the selected architecture

execve allows for non-executable files to be run.

The current path resolution system maintains a list of all open directories and their trees back to the root, this is not necessary and should be replaced by a more memory efficient method of lookup

Background jobs sometimes hang the terminal in busybox ash

To try;
nyan > /dev/tty2
ls
nyan > /dev/tty3
ls