Linux 6.12.113 livepatch releases

This page covers livepatches built from the final system configuration's Linux 6.12.113 package. Each configuration supplies the kernel source, development tree, config and toolchain used by its livepatch producer.

Eligibility

  • Configured kernel: the final configuration's vpsAdminOS Linux 6.12.113 package and its matching producer inputs
  • Eligible running kernels: healthy native 6.12.113 boots with compatible patched code and ABI, covered by the applicable qualification
  • Native-fixed recovery coverage: not yet established; Linux 6.12.117 source contains the complete fix, but source inclusion does not qualify it as an operator recovery target
  • Supported mixed versions: native 6.12.113 and release 6.12.113.1 within one boot epoch, after qualification and rollout approval
  • Maximum stack depth: one, qualified for the captured realization below
  • Boot loading: services.live-patches.loadOnBoot, default true
  • Live deployment loading: services.live-patches.loadOnSwitch, default false

Do not activate this preventive livepatch on a boot that has already hit the target failure, has an unexplained Oops, or is otherwise unreliable. Follow the recovery section instead.

Release set

State Predecessor Fixes Qualification Operator status
Native 6.12.113 none none source applicability and tested-build compatibility reviewed eligible when healthy and covered by compatibility qualification
6.12.113.1 native nfs-sysfs-client-lifetime capture and required safety profiles accepted for the linked realization requires matching qualified artifacts and rollout approval

Release 6.12.113.1

Identity and provenance

  • Release: 6.12.113.1
  • Payload revision: 1
  • Predecessor: native 6.12.113
  • Mode: additive, replace=0
  • Source commit: vpsfreecz/linux@02beaa9040c47f489991bac14c0bdc00b7a737c9
  • Provenance status: downstream only
  • CVEs: none assigned
  • Prerequisites: none

The patch prevents NFS server sysfs callbacks from outliving the RPC client state they access. It does not change the sysfs ABI or NFS wire protocol.

Backport selection and provenance

  • Selected source: vpsfreecz/linux@02beaa9040c47f489991bac14c0bdc00b7a737c9
  • Source release and status: downstream-only v2 change; no upstream or stable commit is assigned
  • Prerequisites: none
  • Source-review conclusion: the isolated change drains server sysfs before client release, retains the NFSv2/v3 client reference through common teardown, and ignores unavailable main RPC clients without adding adjacent NLM changes
  • Target-applicability conclusion: Linux 6.12.113 has the affected client lifetime, while the reviewed Linux 6.12.117 source contains the complete selected correction
  • Expected emitted semantic closure: vmlinux:nfs_free_server@0, vmlinux:nfs_init_server.isra.0@1, and vmlinux:shutdown_store@1
  • Accepted selection evidence: reviewed downstream history, source analysis and exact native-kernel validation of the affected and corrected source paths

Fixes and emitted closure

Fix ID Prevention or correction Patch IDs Affected object/functions Oracle IDs
nfs-sysfs-client-lifetime drain server sysfs callbacks before releasing NFS client state nfs-drain-server-sysfs-before-client-release vmlinux:nfs_free_server@0, vmlinux:nfs_init_server.isra.0@1, vmlinux:shutdown_store@1 nfs-client-operation-safety

The source changes nfs_init_server() and nfs_free_server() in fs/nfs/client.o, plus the static shutdown_client() helper in fs/nfs/sysfs.o. The captured realization emitted the replacement closure shown above: it cloned nfs_init_server(), inlined shutdown_client() into shutdown_store(), and retained nfs_free_server() under its source name. Verify that each newly configured build emits the expected closure.

Livepatch-safety qualification

The realization captured from vpsAdminOS cf062322 passed all required qualification profiles. The capture binds the module, manifest, inspection and emitted closure. AMD and Intel/KVM profiles passed lifecycle, busy operation, failure/recovery and NFS operation, with normal forward and reverse convergence, native return, cleanup and clean health checks. Both aging rows completed the 30-minute dwell with 30 one-minute checkpoints and progress in all six workload dimensions.

A separate compatibility test booted the older Linux 6.12.113 build from vpsAdminOS 252b80df with Nixpkgs 8c259fc0, then switched to the cf062322 configuration with automatic loading disabled. The switch left livepatch state unchanged. Explicit production-manager loading, NFS operation and integrity, reversal, native return and cleanup passed with the same captured module. Numbered release links used the older boot's module tree.

These results apply to the captured realization and tested kernel/configuration pair. Another build of 6.12.113.1 needs the applicable artifact and compatibility qualification; matching version strings alone do not transfer these results. Rollout approval remains separate.

Held-frame proof is not required: none of the expected replacements has a reviewed natural sleepable hold point, so this release does not add a synthetic hold.

Generic rows cover lifecycle, busy transition, aged transition and failure/recovery on AMD and Intel/KVM without importing or running the NFS adapter. Dedicated rows run the ordinary NFS operation and integrity adapter. Every row uses the same captured module and configured target. The qualification does not reproduce the original race or compare a native-fixed kernel.

For deployment onto an older running build, follow the older-boot/newer-userspace procedure. Build from the newer final configuration and verify explicit loading, NFS operation and reversal on the older boot. Different store paths, kernel-image bytes or ELF notes do not by themselves prevent loading. Module integrity, kernel compatibility, transition and workload checks still apply.

Activation

Activate only a realization whose capture and applicable qualification are accepted, after rollout approval. This applies to automatic and manual loading.

With services.live-patches.enable = true, the defaults are:

services.live-patches.loadOnBoot = true;
services.live-patches.loadOnSwitch = false;

A boot into the configured generation attempts loading once before runit starts. Deploying that generation to a running node leaves loading manual by default. Set both options to false to keep boot loading manual as well. An explicit loadOnSwitch = true also attempts loading during live test or switch, including a same-generation switch or a switch back to the booted generation. See the automatic-loading policy for all four combinations and action boundaries.

Each attempt uses the same manager and admission checks as live-patches load. A failure is logged without aborting boot/switch or retrying automatically. Creating or restarting the service does not load or unload a patch. After automatic loading, verify enabled=1, transition=0, replace=0 and uname -r equal to 6.12.113.1, then check NFS operation and data integrity before returning the node to service. Service liveness alone is insufficient.

For manual loading during an approved rollout:

  1. Require a healthy native 6.12.113 boot covered by the compatibility qualification. Confirm live-patches status and live-patches list show no active numbered release.
  2. Confirm the captured module, manifest, inspection and closure still match the approved realization.
  3. Quiesce NFS mount creation, final unmount and writes to NFS shutdown attributes. Keep established I/O bounded and monitored.
  4. Run live-patches load 1 once. The command waits for convergence. Require enabled=1, transition=0, replace=0 and uname -r equal to 6.12.113.1.
  5. Run every active safety check, including NFS data-integrity and continued-I/O checks. Stop on a mismatch, timeout, workload regression or kernel warning.

Bare live-patches load loads all batches shipped by the configuration in order. This target contains only batch 1. Qualification covers native to 6.12.113.1; it does not establish support for a deeper stack. Never force a transition or toggle the patch repeatedly.

Rollback

Reversal is supported only after qualification and rollout approval. Keep the same NFS creation, final-unmount and shutdown-sysfs quiescence, then run live-patches unload 1 once. Require reverse convergence, removal of the module and its metadata, and uname -r equal to native 6.12.113. Recheck existing and fresh serial NFS operation and data integrity. Do not force removal if reverse convergence is incomplete.

Disabling loadOnBoot or loadOnSwitch does not unload the patch. Rolling back the system configuration does not restore live kernel state. Check live-patches status for modules left loaded but absent from the current configuration. Restore a generation that configures the complete loaded stack before managed removal, or follow the approved reboot procedure. Bare live-patches unload removes all loaded configured batches newest first; unload 1 refuses while a descendant remains loaded. Older generations retain their own activation policy and manager capabilities.

Recovery

This livepatch is preventive. It cannot repair an already-freed client, interrupted shutdown work, uncertain RPC/NFS state, application state or filesystem state.

If a boot has already hit the failure or an unexplained Oops, fail over or fence it according to the service procedure. Reboot into an approved native-fixed generation and reconcile uncertain operations before returning the node to service. Linux 6.12.117 contains the v2 source fix, but it is not an approved recovery generation until its operator qualification is recorded.