# DF-3043 verification

Kernel: DragonFly dfbsd 6.5-DEVELOPMENT #0: Thu Jul 2 06:02:54 UTC 2026 root@dfbsd:/usr/obj/usr/src/sys/X86_64_GENERIC x86_64

Not demonstrated end-to-end. The malicious-claims route (oversized GETATTR size + client-side EFBIG readahead leak) required the custom NFSv3 server rig which, after extensive bring-up (record-mark, fattr XDR and fh-length bugs fixed), still could not complete a functional client mount against the stock kernel in available time; the forced-unmount abort site (d) of this same accounting bug was exercised heavily under DF-3052 (126 early aborts observed) but nm_bioqlen itself is not userland-visible and the mount dies with the aborts, so the mount-wide read-ahead-death observable was never measurable. Accounting sites remain code-certain: increment only at nfs_bio.c:957, decrement only in nfs_iod.c:139/212, bypasses at nfs_bio.c:1346-1373/:1463-1478/:1687-1693 and the commit-to-write reissue at :1753-1755.

## Kernel references
- sys/vfs/nfs/nfs_bio.c:957
- sys/vfs/nfs/nfs_bio.c:1346
- sys/vfs/nfs/nfs_bio.c:1753
- sys/vfs/nfs/nfs_iod.c:139

## Fix
See fix.diff.

## Fix status (batch validation)
Applied with the 36 sibling fix.diffs of this batch in findings/poc/_batch41/all_fixes.patch; single in-guest `make -j6 nativekernel KERNCONF=X86_64_GENERIC` completed RC=0 with zero errors under -Werror (full log: findings/poc/_batch41/fix_build8.log). Behavior A/B intentionally skipped for this Info-tier item.