Fix BSS segment corruption in file-backed mmap (dynamic library loading)

[Copilot]

When fmap handles a file-backed mapping where memsz > filesz (BSS extension), it was allocating physical pages based on min(len, file_size - offset) — the file-data portion only — while sys_mmap mapped len-worth of pages using that physical address. BSS pages fell outside the allocation, mapping to adjacent unallocated physical frames with stale content. BSS globals initialized to zero by the ABI held garbage values, causing null-pointer-ish dereferences during shared library init.

Observed crash: lightdm (dynamic binary) faulted at CR2=0x7ce8 with RAX=0x7cd8 — a GLib/GObject BSS global used as a pointer, non-zero due to this bug.

Changes

• filesystem.rs — fmap for OpenFile::Real: compute aligned_len from the full requested len rather than want. The pre-existing write_bytes(..., 0, aligned_len) already zeros the whole region; only the file-data slice (want bytes) is read from disk — no change to the read path.

// Before: allocates pages only for file-data portion, BSS maps to garbage
let aligned_len = (want + 0xFFF) & !0xFFF;

// After: allocates pages for full mapping (file data + BSS extension)
let aligned_len = (len + 0xFFF) & !0xFFF;

[Copilot]

Pull request overview

Fixes incorrect physical backing for file-backed mmap when the requested mapping length (len) exceeds the file-data portion (filesz, i.e., memsz > filesz BSS extension). This prevents BSS pages from mapping to unrelated/unallocated physical frames containing stale data, which can crash dynamic library initialization.

Changes:

• In FileSystemScheme::fmap for OpenFile::Real, compute aligned_len from the full requested len (not from want).
• Clarify comments around want (file bytes) vs len (file bytes + BSS extension) and the zeroing/copy behavior.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

aligned_len is computed with (len + 0xFFF) & !0xFFF, which can overflow/wrap for very large len values (release builds wrap). Since len ultimately comes from userspace mmap, this can lead to allocating/zeroing the wrong amount of memory. Use a checked/saturating align-up (e.g., len.checked_add(0xFFF) + mask) and return EINVAL on overflow.

Suggested change

	let aligned_len = (len + 0xFFF) & !0xFFF;
	let Some(aligned_len) = len.checked_add(0xFFF).map(|v| v & !0xFFF) else {
	return Err(scheme_error::EINVAL);
	};