[BUG] Bootstrap scheduling changes permissions outside registered VSpace ranges #32
Labels
No labels
alpha-1
bug
ci
docs
duplicate
enhancement
help wanted
invalid
performance
phase-6
question
refactor
security
wontfix
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
erix/kernel#32
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Summary
The original bootstrap scheduler overlay derived an additional user-access span from numeric entry and stack addresses. It could request permission changes outside the selected VSpace inventory, including the hole between separately registered ranges. The correction removes that overlay and its obsolete threshold; scheduler access now comes only from the VSpace's accounted mappings and registered ranges.
Reproduction
At signed Kernel
ced94e22401495cc98aa30143c224c737adb764e, unchanged production operation logic plus a test-only thread-local observer records validated requests at the host permission-range boundary. Six independent controls produce four failures and two passing controls, with Cargo exit 101, warning-free compilation and successful cleanup. A high numeric entry, a high numeric stack, disjoint registered ranges and withdrawal each demonstrate an extra unregistered interval. An ordinary lower context and missing-VSpace refusal pass. The same six controls all pass after correction.The observer neither dereferences high virtual addresses nor mutates host page tables or substitutes operation results. Original sources, hook patch and failed outputs remain retained. This is an original permission-request reproduction, not an original native or CPL3 exploit.
Expected and actual behavior
Only the exact live VSpace's tracked mappings and explicitly registered ranges may supply user permissions. Numeric execution addresses cannot authorize extra access or permission withdrawal in surrounding pages. Direct root launch already registers its image, handoff and stack separately; the correction preserves those registrations and their distinct policies.
The removed helper additionally covered the numeric interval between entry and stack, including unrelated holes and a trailing page. External start-context admission and Process endpoint scope remain separate review boundaries; the source trace alone does not establish unprivileged exploit reachability.
Environment and identity
Host controls use Linux x86_64, Rust 1.97.1, original exact pinned dependencies and warnings denied. The coordinated Integration catalog selects the signed correction in all three Kernel catalogs. Maintained VM scenarios retain their original grants and 60/120-second deadlines.
Evidence and prior checks
Four strict Kernel configurations pass 813/837 tests with three existing ignored cases. Sixteen native builds and binary Clippy profiles, formatting and host/native private rustdoc pass without warnings. All 172 maintained Integration helpers pass; unchanged Rust and profiler sources retain their strict validation.
Six native executions satisfy seven contracts: lifetime, invocation, mapping and sparse checks sharing one configuration, both allocator checks and ordinary exec-successor across all 73 components. Before first user entry, the added isolated lifetime fixture checks actual read-only and writable registered leaves, an untouched internal gap and trailing guard, withdrawal and exact restoration before backing release. It uses a local scalar record, changes no real TCB context or generation, and adds no user grant. The existing later user-mode witnesses also remain required. Exact signed sources and retained packaged artifacts are verified, with no image warnings or QEMU stderr.
The manual passes all 45 tests, complete 2,433-page generation, all 451,723 word bounds and changed-page visual review, without final warnings or overflow.
Completed registered-bootstrap CI review at 2026-09-23T08:13:54.917815+00:00: Kernel 658/659, Docs 1023/1024 and original Integration 1739/1740 pass. Both Integration runs pass all 489 catalog VMs, canonical development/release images, physical input gates and remaining workflow checks. Six complete Integration logs total 26,974,271 bytes with zero warning candidates; together with eight reviewed Kernel/Docs logs, 14 complete logs total 28,579,749 bytes with zero final warnings. Original failures remain retained, without job retries. This completes corrected validation for Kernel #32; later table/heap CI and full guest rebuilding remain separate. The report is closed after this complete corrected CI review. Earlier original input/image failures remain separately tracked; no unchanged job was rerun.
Tracking
Related owned address-space design, Kernel compatibility, Kernel audit, Kernel review, Integration review and Phase 6 completion. This is separate from the corrected initial-stack materialization defect. Private root population, readiness, CR3 activation, CPU residency, live-root retirement and full guest rebuilding remain open.
erikinkinen referenced this issue2026-09-23 07:46:42 +02:00
erikinkinen referenced this issue2026-09-23 07:46:44 +02:00
erikinkinen referenced this issue from erix/integration2026-09-23 07:46:50 +02:00
erikinkinen referenced this issue from erix/docs2026-09-23 07:46:56 +02:00