Patching systems that cannot be rebooted
Live patching covers security fixes without a restart but never all of them. Windows Server hotpatch uses four planned baselines and eight hotpatches a year; Red Hat live kernel patches cover selected Important and Critical CVEs for 6 months after the errata date. A policy must name what still needs a restart. Read 19 Nov 2026.
Three live patching schemes and their limits
| Scheme | What it covers (vendor statement) | Limits stated by the vendor |
|---|---|---|
| Windows Server hotpatch (Microsoft) | Security updates without restart; baseline refreshes every three months, hotpatches in the next two months; a planned year has four baselines and eight hotpatches | Restart needed for non-security updates, .NET and non-Windows updates such as drivers and firmware; unplanned baselines (for example a zero-day that cannot be hotpatched) replace a hotpatch month and require a restart; no automatic rollback; only listed Windows Server 2022 and 2025 images; container base images unsupported |
| Canonical Livepatch (Ubuntu Pro) | Critical and high severity kernel vulnerabilities, patched “between security maintenance windows, while the system runs” | Kernel only; coverage 10 years with Ubuntu Pro and 5 more with the Legacy add-on |
| Red Hat live kernel patching | Selected Important and Critical CVEs; patches are cumulative | Supported without extra cost for 6 months after the kernel errata release date; only x86_64 and ppc64le; none released for RHEL 6 and 5 |
Upstream, the Linux kernel documentation explains the reason: “There are many situations where users are reluctant to reboot a system.” It also states a hard limit: only functions that can be traced can be patched (kernel.org). The older kpatch project says that starting with Linux 6.19 it is deprecated and in maintenance mode (GitHub); that is a statement about the upstream project, not about any vendor's product.
The policy has to say what live patching does not remove
Each scheme leaves a residual. A hotpatch calendar has months where a restart is needed; a Livepatch is limited to high and critical kernel flaws; a Red Hat live patch stops being supported after six months without extended update support, after which a restart is needed anyway. Write the restart windows into the policy: the quarterly baseline for Windows Server, and a scheduled kernel restart at least when the live patch support period ends. Hotpatch prerequisites also matter for Windows servers and Linux servers.
When nothing can be applied
NIST names a fourth scenario for assets that cannot be patched or must run without interruption: “implementation of isolation or other methods to mitigate the risk.” Its advice is to have an approved set of methods per maintenance group, reviewed in advance by security architects, such as micro-segmentation, and to reevaluate them periodically (NIST SP 800-40 Rev. 4, sections 3.3 and 3.5.4). The written exception, with its expiry, is covered in the exceptions clause.
Evidence
- The live patch identifiers applied and the kernel or build they apply to.
- The date of the last full restart per system, so the gap between restarts is visible.
- For isolated assets, the approved method, who reviewed it, and the next review date.
Sources
- Microsoft Learn, Hotpatch for Windows Server, read 19 Nov 2026
- Canonical, Livepatch, read 19 Nov 2026
- Red Hat, live kernel patching, read 19 Nov 2026
- Linux kernel documentation, livepatch, read 19 Nov 2026
- kpatch project README, read 19 Nov 2026
- NIST SP 800-40 Rev. 4, Guide to Enterprise Patch Management Planning (April 2022), read 19 Nov 2026
Last reviewed 19 Nov 2026