nx_vfsblock_lib.nx

nx_vfsblock_lib.nx -- the SOVEREIGN block-backed VFS + write-ahead JOURNAL library (NishiOS rung 4: closes the FILESYSTEM census gap "block-backed VFS mount + journaling"). ONE on-disk format (NSFS v2) designed once so the mount gate (nx_vfs_blockfs) and the journal gate (nx_journal) share it -- no format churn, rule-15 DRY. ON-DISK LAYOUT (NSFS v2, 256B blocks, 1024 blocks = 256KB image; ADDITIVE-ONLY data, rule 13): block 0 SUPERBLOCK: [0]magic'NSFS' [8]version=2 [16]nblocks [24]next_free [32]j0 [40]nj [48]ino0 [56]ni [64]data0 blocks 1..32 JOURNAL: block 1 = header [0]jmagic'NSJL' [8]seq [16]nrec [24]committed [32]sum [40+8j]target_j (j<27); blocks 2..28 = record payloads (the FULL new content of the target block = physical redo WAL, jbd2-class) blocks 33..96 INODE TABLE (64 inodes, one per block): [0]used [8]type(1=dir,0=file) [16]parent [24]size [32]start(data block#) [40]nblk [48]datasum [64..127]name(63B+NUL). Root = inode 0 (dir, parent -1). blocks 97.. DATA (bump-allocated contiguous, never reused = history-preserving) HIERARCHY = the nx_vfs model persisted: children are inodes whose parent field points at the dir inode; path resolution walks components from root against the ON-DISK inode table. MOUNT = validate superblock (magic+version) -> REPLAY the journal (crash recovery) -> validate root. WAL protocol: stage payloads+targets -> COMMIT (single flag write after a checksum over the staged records) -> APPLY to home blocks -> CHECKPOINT (clear committed). Crash after commit -> replay re-applies (idempotent physical copies). Crash before commit -> txn discarded, home blocks untouched BY CONSTRUCTION (records live only in the journal region). Corrupt committed txn -> checksum REFUSES the replay and discards (the jbd2 rule: an unprovable txn is never applied). Txn cap: 27 record blocks -> journaled creates up to ~25 data blocks (~6.4KB); bigger files go through the direct path at image-build time. NEVER-BRICK: operates on RAM + an image file, writes 0 firmware. license_tier: ORIGINAL