CVE-2026-89924: KVM: s390: Fix old_data leak in guest debug error path
In the Linux kernel, the following vulnerability has been resolved:
KVM: s390: Fix olddata leak in guest debug error path
importwpinfo() allocates a per-watchpoint olddata buffer to back up the original guest memory contents. If a later watchpoint of the same KVMSETGUESTDEBUG request fails to import, kvms390importbpdata() jumps to the error label, which frees the wpinfo array but not the olddata buffers of the entries that were imported successfully. Up to MAXBPCOUNT - 1 buffers of up to MAXWPSIZE bytes are leaked per failed request, and the request can be repeated.
Create error handling for cleaning up all created olddata memory areas.
Affected Software
Remediation
Recommended actions to resolve this vulnerability, in priority order.
- Upgrade
Upgrade
KVM (s390)to a version that resolves this vulnerability.Patch KVM: s390: Fix old_data leak in guest debug error path - Operational
After applying the kernel fix, ensure no repeated failing KVM_SET_GUEST_DEBUG requests remain in place; re-test that repeated failures no longer leak per-watchpoint old_data buffers in the guest debug error path.
Event History
Frequently Asked Questions
Who can trigger the memory leak?
An entity able to submit KVM_SET_GUEST_DEBUG requests for an s390 KVM guest can trigger it by causing a later watchpoint in a request to fail after earlier watchpoints were imported successfully.
What is the practical impact if the issue is repeatedly triggered?
Each failed request can leak up to MAX_BP_COUNT - 1 old_data buffers, each up to MAX_WP_SIZE bytes. Because the request can be repeated, memory consumption can accumulate over time.
How can I tell whether a system may be affected?
The affected error path is reached when importing multiple watchpoints in a KVM_SET_GUEST_DEBUG request: at least one watchpoint imports successfully and a later one fails. The provided information does not identify a specific log message or runtime indicator for detecting prior leaks.
What does the fix change?
The fix adds error handling that cleans up all old_data memory areas created for successfully imported watchpoints when a later watchpoint import fails.