Article Published Date: 11/9/2026
The latest 17.0 LTS version is https://github.com/lf-edge/eve/releases/tag/17.0.1-lts
For a change log between 17.0.0-lts and 17.0.1-lts, see: https://github.com/lf-edge/eve/compare/17.0.0-lts...17.0.1-lts
Enhancements
Linux OS and kernel updates
- Updated the 17.0 kernels to the latest eve-kernel branch heads, so all supported platforms receive the current upstream stable fixes.
- Upgraded the amd64 kernel from Linux 6.12.49 to 6.12.96, which brings in KVM fixes and other upstream stable fixes.
- Enabled Pressure Stall Information (PSI) on NVIDIA Jetson (JP6 and JP7) kernels, which improves resource pressure visibility.
- Enabled the memory and cpuset cgroups and PSI in the riscv64 kernel configuration.
- Updated Intel and AMD CPU microcode.
Security
- Fixed CVE-2026-98432, a TCP/IPv4 use-after-free of IP options, on all kernels.
- Fixed CVE-2026-99999, an IPv4 out-of-bounds write caused by an unvalidated
checksum_startvalue, on all kernels. - Fixed CVE-2026-46113 and CVE-2026-53359, two KVM x86 shadow paging use-after-free issues, on the amd64 kernel.
- Confirmed that every 17.0 kernel now includes the fix for CVE-2026-31431 (AF_ALG AEAD splice write to the page cache).
- Hardened path handling in image verification, cloud-config file writes, and EdgeView to prevent path traversal and argument injection attacks.
- Required a proxy-auth token, TLS, and a destination policy on the management proxy listener used by EVE-k.
- Bounded HTTP response reads and error strings to protect edge nodes from resource exhaustion.
- Hardened virtual TPM state handling and migrated legacy encryption.
- Hardened 802.1X (PNAC) re-authentication and removal handling.
- Surfaced how the vault was unlocked and reported the key derivation method, which improves auditability of edge node encryption.
- Updated the CA certificate bundle to stop trusting the compromised eMudhra certificate authorities.
Networking
- Made Local network instances resilient to uplink port flapping.
- Added a configurable DPC test-fail interval and lowered the minimum test interval, which gives administrators finer control over connectivity checks.
- Improved tracking of network interface and bridge identity so that interface changes are handled reliably.
- Fixed reporting of application IP addresses and flows.
Hardware and device support
- Added the ability to change an adapter assignment with an application restart instead of a purge.
- Added support for disabling the console at runtime.
- Reported USB-attached UARTs as COM ports.
- Improved iGPU passthrough by placing guest stolen memory at the host base and always detaching the host framebuffer for an assigned iGPU.
- Matched device ports by PCI address instead of interface name alone, and resolved interface names to the deepest PCI address.
- Fixed the CDI file for NVIDIA Jetson JP7 and the udev rules for NVIDIA devices.
EVE-k and clustering
- Added support for opting in to native Kubernetes orchestration on an edge node cluster.
- Allowed a healthy peer node to act for a downed node's designated application.
- Surfaced insufficient disk and memory conditions for Longhorn, and reported storage readiness accurately.
- Reduced idle logging in the CSI handler status path.
- Accepted a cluster PVC for volumes instead of downloading the source again.
- Used a stable latest tag for the external boot image in VMIRSes.
Observability
- Reported the provisioned size (
volsize) of ZFS zvol volumes in volume metrics. - Captured cgo aborts and full tracebacks in logs.
- Reduced log noise by no longer logging routine QMP events at warning level.
- Honored the hostname declared in an application instance cloud-init configuration.
Resolved Issues and Bug Fixes
- Fixed a vTPM issue so that it detects the SHA256 PCR bank.
- Fixed an issue where failed volume deletions leaked volumes. EVE-OS now retries them.
- Fixed an out-of-memory condition in the pillar service caused by the USB manager.
- Fixed an issue where work was lost when the volume manager worker pool was saturated.
- Fixed a node reboot that occurred when an I/O bundle did not belong to the edge node.
- Fixed a reboot loop and several out-of-range panics related to PAC results.
- Fixed a zedrouter crash that occurred when a network instance was deleted.
- Fixed a panic caused by concurrent ZFS namespace enumeration.
- Fixed a watchdog reset caused by the disk metrics collection.
- Fixed a proto registry panic in the TPM recovery package.
- Fixed goroutine leaks in the zedUpload and nettrace libraries.
- Fixed downloads aborting when the declared size is unknown, and downloads hanging on a refused request.
- Fixed unbounded growth of download metrics counters.
- Fixed newlog space reservation so it uses bytes instead of megabytes.
- Fixed an issue where the configuration was not re-read after a maintenance mode change or clear.
- Fixed bootstrap SSH authorized keys losing newlines between entries.
- Fixed the installer post-install verification stage.
- Fixed an issue where escrow data could be lost during attestation.
- Fixed a failed decryption being reported as a success.
- Fixed file permission gaps in the vault and attestation components.
- Fixed patch envelope artifacts backed by zvols being served with the wrong size.
- Fixed USB passthrough behavior when device models change at runtime.
- Fixed PCI address extraction for USB devices behind PCIe bridges.
- Fixed a graceful stop wait that could run without a bound, and made the force flag work when stopping a KVM domain.
- Fixed two adapter assignment regressions introduced by the PCI port matching change.
- Fixed the portmap overlap check so it ignores Switch network instances.
- Fixed a case where a VMIRS remained stranded with zero replicas.
- Fixed EVE-k issues with stale VMIRS generations after a purge, stuck kubelet volume mounts, Multus NAD apply ordering, the Longhorn disk reservation, and the Longhorn disk path readiness gate.
- Fixed EVE-k issues with the SR-IOV device plugin gate, CDI HTTP DataVolume imports through the proxy, and PVC upload cancellation and failure reporting.
- Fixed EVE-k container volume user name resolution, and added diagnostics for stuck application bring-up.
- Fixed stale state after a cluster DNID reassignment, and cleared the retired pod subnet when a node joins.
- Fixed controller pod recycling after a single-node to cluster datastore migration.
- Fixed containerd CPU confinement so it respects the configured maximum vCPU count.
- Fixed the debug container exec scheduler and the debug package device model fields.
- Fixed decimal literals used as file modes in pillar.
- Fixed installer media sharing disk and ESP GUIDs by giving each installer its own.