munmap (nr 11)

Linux Signature

int munmap(void *addr, size_t length);

Description

Removes mappings for the specified address range, causing further references to addresses within the range to generate page faults.

Current Implementation

Fully implemented. Validates arguments, splits/removes VMAs, unmaps page table entries, frees physical frames to the free list, and flushes the TLB.

Source: osl/src/syscalls/mem.rs — sys_munmap

Behaviour

1. addr must be page-aligned; length must be > 0. Returns -EINVAL otherwise.
2. length is rounded up to the next page boundary.
3. Overlapping VMAs are split or removed:
   • Entire VMA consumed — removed from vma_map.
   • Front consumed — VMA start/len adjusted forward.
   • Tail consumed — VMA len shortened.
   • Middle consumed — VMA split into two fragments.
4. Each page in the unmapped range is removed from the page table. Physical frames are released via refcount-aware logic: shared frames (from MAP_SHARED mappings) are only freed when their reference count reaches 0 (i.e. all processes have unmapped the frame and the backing shmem_create fd has been closed). Non-shared frames are freed immediately.
5. If no VMAs overlap the range, returns 0 (Linux no-op semantics).

Lock ordering

PROCESS_TABLE is acquired first (to call munmap_vmas), then released before acquiring MEMORY (to unmap and free pages). Same ordering as sys_mmap and sys_brk.

Errors