Boundary Mathematics: Weaponizing PAGE_SHIFT Arithmetic via FUSE — Part 3

MAX_LFS_FILESIZE only gates the superblock. Once the FUSE connection is up, every FUSE_GETATTR reply can mutate i_size to 0xFFFFFFFFFFFFFFFF. The page cache’s (pos + count - 1) » PAGE_SHIFT arithmetic wraps unsigned, inverts loop invariants, and turns vma_merge() into an arbitrary OOB-write primitive on Maple Tree-backed kernels.

April 29, 2026 · JM00NJ

Lying to the Kernel: FUSE Trust Boundary & Size Desync as a VFS Attack Surface — Part 1

An unprivileged FUSE daemon controls the semantic authority of an entire filesystem. By lying about i_size in vfs_getattr replies, it desynchronizes kernel allocation from kernel ingestion — turning finit_module(2), the firmware loader, and kexec_file_load(2) into kmalloc-4k slab overflow primitives.

April 29, 2026 · JM00NJ

The Async Abort Race: drop_caches × SIGKILL × fuse_abort_conn = Double Put — Part 4 & Conclusion

struct fuse_req borrows inode references without bumping i_count. A SIGKILL’d reader, an unrelated drop_caches sysctl, and a delayed daemon abort conspire to dereference freed-and-reoccupied slab memory. The result: a refcount decrement on whatever struct cred lands in the freed slot — the entire kill chain in three syscalls and one signal.

April 29, 2026 · JM00NJ

Two Approaches to EDR Evasion: Kernel-Level BYOVD vs User-Space Injection

BYOVD kills your EDR. User-space injection makes it irrelevant. These two approaches to defeating endpoint detection operate at different privilege levels, target different layers, and require entirely different defensive strategies.

April 29, 2026 · JM00NJ

ICMP-Ghost v3.6.2: Fileless C2 with Dual-Channel Pivoting & DPI Evasion

Ghost-C2 v3.6.2 introduces Dual-Channel Protocol Pivoting via an in-memory VTable architecture — seamlessly switching between Raw ICMP and DNS UDP tunneling at runtime. Combined with PIC injection, VESQER compression, and layered evasion, it defeats Suricata v8.0.3. All in pure x64 Assembly.

April 25, 2026 · JM00NJ