QEMU security fixes and Postgres feature rollback debate

QEMU work led the day with virtio-gpu dmabuf ownership fixes tied to a recent CVE and continued pushes to make the human monitor optional at build time. PostgreSQL hackers meanwhile argued over reverting several complex PG19 features after elevated post-freeze bug counts.

Virtio-gpu dmabuf and scanout fixes

An 18-patch RFC series on qemu-devel addresses virtio-gpu and dmabuf ownership bugs in QEMU. The changes were uncovered while handling CVE-2026-66020 and focus on scanout ownership. Virtualization stacks that expose GPU or shared-buffer paths need to track the corrected ownership model.

Making QEMU HMP compile-time optional

Several related series, including 49- and 50-patch versions and a maintainer pull, propose making the Human Monitor Protocol optional via --disable-hmp or a CONFIG_HMP build flag. The work separates HMP from QMP so deployments can ship smaller QMP-only binaries. Embedders and minimal guests that no longer use the legacy interactive monitor are the primary audience.

Postgres debate on reverting PG19 features

A heated pgsql-hackers thread examined whether to revert three complex features slated for PG19: RI foreign-key batching, REPACK CONCURRENTLY, and online checksums. Participants cited high post-freeze bug counts surfaced in part through LLM analysis. The decision will affect what ships in the next major release and how much risk the project accepts after feature freeze.

Upstream of remaining bsd-user syscalls

Warner Losh posted v4 of a 40-patch series completing most remaining system calls for QEMU bsd-user. The series notes heavy assistance from Claude with explicit attribution. Fuller syscall coverage improves fidelity of FreeBSD user-mode emulation.

Vhost-user isolated memory mode

An RFC v2 series of 13 patches adds an optional memory-isolation mode to QEMU vhost-user. The design relies on bounce buffers and shadow virtqueues. The option reduces shared-memory attack surface between guests and device backends.

Logical replication conflict log table

Review of a proposed conflict-log history table for PostgreSQL logical replication continued, surfacing buffer-size and replica-identity bugs. The patch aims to retain durable conflict records rather than transient logs. Operators running logical replication for failover or multi-node topologies have a direct interest in the fix quality.