vms-4a8: bootable-image vms.ko carries the kernel-resident ODS-2 codec (unblocks ACP boot)

[baron-3dl]

What

Make the bootable-image vmsfs.ko (built by the in-tree drivers/ovmx/ overlay, not the out-of-tree make -C src/kernel) carry the kernel-resident genuine ODS-2 codec + read-only ODS-2 presentation, so the shipping module can mount+read a genuine ODS-2 volume. Rung of epic vms-208 (Files-11 ODS-2 ACP in the executive); unblocks the ACP boot rung.

ADDITIVE: this makes the shipping module carry the codec. It does not flip the boot default or mount SYS$DISK via the ACP — the bespoke-VMFS path still drives boot until the atomic-flip rung. The codec is present-but-not-yet-driving-boot.

The gap (vms-dcd / #623)

The overlay FLATTENS every module source to a basename. The codec's sources #include "vmsfs/ods2.h" — a subdir path — which the basename-flatten model can't resolve. So the overlay Kbuild deliberately left OVMX_ODS2_KERNEL undefined and dropped the codec objects (to avoid a dangling-symbol modpost break at vmsfs_ods2ro_register). The follow-up was tracked as vms-4a8 in vmsfs_super.c and drivers-ovmx/vmsfs/Kbuild.

Flatten-safe approach

A single additive convention, no source rename:

• overlay-ovmx-drivers.sh: a sources.conf line may now use <glob> -> <subdir>/ to stage matched files under the module dir's <subdir>/ instead of the flat root. Lines without -> flatten to basenames exactly as before — nothing else in the overlay, and no other module, changes.
• vmsfs/sources.conf: stage the codec (src/vmsfs/ods2/*.c,*.h) flat, and stage the one subdir'd public header preserving its path: src/vmsfs/include/vmsfs/ods2.h -> vmsfs/. #include "vmsfs/ods2.h" then resolves via -I$(src) (→ <moddir>/vmsfs/ods2.h). Every other codec include is basename-only.
• vmsfs/Kbuild: add the codec objects (ods2_reader/writer/bdev/path/block_kern + vmsfs_ods2ro) and define -DOVMX_ODS2_KERNEL, so the gated vmsfs_ods2ro_register() call in vmsfs_super.c now resolves — the inverse of vms-dcd: ODS-2 codec runs kernel-resident inside the shared FS engine (epic vms-208) #623's gap.
• Dockerfile.bootable: add binutils and an nm gate asserting the shipped vmsfs.ko carries vmsfs_ods2ro_register + ods2_* (codec-compiled-in proof).

The codec source and the object list are untouched — the out-of-tree Makefile keeps its -I .../vmsfs/include include path, so the fix is invisible to every other build path.

Modpost-clean-WITH-codec proof (local, real kernel headers)

Staged the overlay into a fabricated kernel tree and built the flat layout with the real staged Kbuild + -DOVMX_ODS2_KERNEL against real kernel headers:

• All codec objects compile from the flat dir; vmsfs.ko links.
• modpost clean — no undefined ods2_* / vmsfs_ods2ro_* symbols.
• nm vmsfs.ko shows T vmsfs_ods2ro_register, T ods2_bdev_finish_open, T ods2_wvolume_append_file — the codec is compiled in.

Regression

• Out-of-tree make -C src/kernel (vms.ko) and make -C src/kernel/vmsfs (vmsfs.ko, the QEMU-test module) — both build clean, modpost clean, unchanged (this item edits only distro/).
• Module-home gate tests/integration/test_ovmx_module_home.sh — 29/29 pass (exercises the overlay incl. the new -> line; the ->-free demo module still flattens normally).
• Change touches only distro/ build plumbing (Docker + overlay + Kbuild + sources.conf); nothing in the CMake graph.

⚠ Cross-build heads-up for peers

This edits only the in-tree overlay's include staging; it does not touch the codec sources or the -I .../vmsfs/include include convention the cross-builds rely on. As a precaution please re-run:

• Alpha: tools/cross-alpha/build-vmsko-alpha.sh
• VAX: tools/cross-vax/build-vms-module-vax.sh

Both should be unaffected (their include paths are unchanged), but confirm.

🤖 Generated with Claude Code

[baron-3dl]

Real Dockerfile in-tree build now GREEN (the gate that was red)

The original PR proved staging against a fabricated kernel tree fed the full repo — which masked the real bug. Rebuilt against the actual distro/Dockerfile.bootable kernel-build stage (kernel 6.12.103, --target kernel-build). Three real defects the fabricated proof could not see, each fixed and re-proven:

1. Codec sources never staged in the real build. The kernel-build stage COPYs only src/kernel + src/kernel-core into /tmp/ovmx-src; src/vmsfs was absent, so the overlay's src/vmsfs/ods2/*.c globs matched nothing (silent WARN) and Kbuild died on No rule to make target ods2_reader.o. → COPY src/vmsfs added; overlay now hard-FAILs on any zero-match glob (verified: FAILs loudly without src/vmsfs, passes with it) so this class can't ship silently again.
2. ods2_wvolume_dir_insert frame 3288B > kernel -Werror=frame-larger-than=2048. ~2.6KB of on-stack scratch (hdr+hdr_parsed+newrec+lbns) moved to one ods2_kzalloc heap block with single-exit done: cleanup. Behaviour-identical; an -O2 sweep confirmed it was the only codec function over 2048.
3. vmsfs_ods2ro.c used sb->s_bdev->bd_inode, removed from struct block_device in the 6.x block layer. → bdev_nr_bytes(sb->s_bdev) (same byte count, linux/blkdev.h, 5.16+ so also fine out-of-tree/cross).

Real build result (exit 0):

CC [M]  drivers/ovmx/vmsfs/{ods2_reader,ods2_writer,ods2_bdev,ods2_path,ods2_block_kern,vmsfs_ods2ro}.o
LD [M]  drivers/ovmx/vmsfs/vmsfs.ko
Built OVMX kernel: 6.12.103-ovmx, bzImage 13M
OK: 2 OVMX module(s) built IN-TREE (intree=Y ...) and SIGNED
OK: shipped vmsfs.ko CARRIES the kernel-resident ODS-2 codec (vmsfs_ods2ro_register + ods2_* present)

Regression checks: host ods2 ctest 13/13 (incl. ods2_dirgrow/multiversion/longname/write/master — exercise dir_insert); out-of-tree make -C src/kernel/vmsfs on host 6.8 green. Two pre-existing sub-2048 frame warnings (1584B, 1696B, other functions) are unchanged, under the ship gate, and out of scope here.