Proxmox VE 9.2.22 → 9.2.26
4 releases · 0 breaking changes · 0 security fixes · 0 advisories fixed
0security0breaking0deprecated6feature8fix
Breaking changes
None in the release notes.
Security changes
None in the release notes.
Every release in between
9.2.26 Nov 4, 2026
featureapi: backup: extract config: also allow it with the Datastore.Allocate permission on the storage, like restoring a backup does.
fixfix #8169: api: backup: extract config: allow extracting the configuration of guest backups without a guest ID again, for example of manually renamed backups.
otherceph: osd: create: rewrite an outdated bootstrap-osd keyring with the key from the auth database. A stale key, for example on a node restored from a backup taken before a key rotation, made OSD creation fail with "RADOS permission denied".
otherapi: backup: only delete backup jobs. Passing the ID of another job, like a realm sync job, deleted that job.
featurejobs: keep jobs of unknown types in jobs.cfg. During a rolling upgrade, editing any job on a node that was not upgraded yet removed jobs of newly added types for the whole cluster.
otherapi: replication: describe the returned job status in the API schema, so API clients, like the Proxmox Datacenter Manager, and the API viewer can show the job states.
fixfix #7513: certificates: check the permissions on the '/nodes//certificates' path instead of the whole node, and offer it in the permission path selector, so certificate management can be delegated with least privileges.
otherfirewall: vnet: only list the VNets of the running SDN configuration in the VNet firewall dialog, as the permissions are checked against it.
fixui: fix showing "(undefined)" in the title of the tag context menu in clusters, and drop a redundant '#' from the ASN label of SDN controllers.
fixfix #5001: ui: add a context menu with guest bulk actions to pools in the pool view of the resource tree, like the tag view has.
featurenotifications: support nested match rules through the new 'expression' option of matchers, which combines calendar, field and severity rules with all-of, any-of, one-of and their negations at any depth:
otherthe matcher editor of the web interface uses it, and shows existing matchers converted to an equivalent expression.
otherthe API refuses to save matchers with an expression while a cluster node reports an older version, as such nodes cannot parse them. Nodes without a reported version are not checked, so upgrade all nodes first.
9.2.25 Oct 27, 2026
featureceph: allow the installation wizard to select any Ceph release supported by the target node, even when the node serving the web interface has a different architecture or Proxmox VE major version.
otherui: virtual machines: only offer TPM 2.0 for aarch64 guests, as TPM 1.2 is not supported by their QEMU device and firmware.
othercertificates:
otherrenew the cluster CA during the daily update before its remaining validity shortens node certificates, and log the expiry of renewed node certificates. Clients trusting the CA directly must import the renewed CA certificate before the previous one expires.
otherre-issue node certificates signed by a replaced cluster CA, so nodes that were offline during CA rotation recover on their next daily update.
othermark expired and near-expiry certificates in the web interface, with a shorter warning window for short-lived certificates to avoid warning before their normal renewal time.
fixfix #8195: ceph: lock configuration updates when creating or removing metadata servers, preventing concurrent tasks from overwriting each other's changes or leaving stale MDS sections behind.
fixfix #8184: ui: settings: allow Remote Resizing as the default noVNC scaling mode, so the guest display can follow the console window size when supported.
otherceph: order all ceph-* services, RBD mappings and CephFS mirroring after the cluster filesystem at boot, avoiding failures from reading their configuration before /etc/pve is mounted.
fixfix #8006: acme: base automatic renewal and the API renewal threshold on the certificate lifetime instead of a fixed 30 days. This avoids unnecessary daily renewals and CA rate limits for shorter-lived certificates while retaining at least three days for retries.
9.2.24 Oct 23, 2026
otherapi: node shell: harden the console against crafted parameters by only passing the known Ceph version and repository options to the command of the Ceph installation wizard, rejecting any other parameter.
fixui: consent banner: fix loading the web interface when the consent text contains a single quote, which broke the JavaScript of the page. This only affected texts set outside the web interface, for example through the API, as the web interface stores them base64-encoded.
9.2.23 Oct 19, 2026
featureceph: add Ceph 21.2 Umbrella as an experimental release, which 'pveceph install' and the installation wizard offer for testing. This also lets the wizard recognize a cluster that already runs Umbrella when adding a node, instead of reporting that it found no Ceph installation.
otherceph: OSD flags: offer the 'noautoscale' flag, which pauses the PG autoscaler for all pools. This helps during maintenance and upgrades, for example to Ceph 21 Umbrella, which doubles the autoscaler's target of PGs per OSD and could otherwise start splitting PGs while daemons of both versions run.…
otherceph: rolling restart: let a missing active manager (MGR_DOWN) only block OSD restarts, as only their safety check needs the manager. During an upgrade, managers still running the old version can stop reporting to the upgraded monitors, which left no manager active and blocked restarting the…
otherceph: rolling restart: do not let BLUESTORE_BLUEFS_OVERSIZED warnings block. Ceph 21 Umbrella OSDs can raise this space usage advisory right after their upgrade, which made the restart refuse every further node unless forced.
otherapi: resource pools: require the Pool.Audit privilege on a pool when querying it with 'GET /pools?poolid='. The permission was checked, but its result was ignored, so any user could list the members of every pool.
otherapi: network: reject line breaks in the OVS options of an interface, as the value is written as is into /etc/network/interfaces.
featureui: ceph: service list: allow restarting a monitor, manager or MDS shown as stopped. The shown state comes from the cluster, so a daemon whose service runs but does not report in, such as a manager not reporting to upgraded monitors, only offered Start, which did nothing.
otherceph: Cephx key migration: support Ceph daemons on hosts outside this Proxmox VE cluster, such as the tiebreaker monitor of a stretch cluster. Previously, such a monitor made the helper refuse staging client keys and switching the service cipher, and made the monitor key rotation fail:
otherquery such monitors with 'ceph tell' instead of over SSH
otherjudge their cipher support by the version they run, and refuse if they are not reported as running
otherrotate the 'mon.' key without writing their keyring or restarting them, print the commands to refresh their emergency keyring, and check that they remain in the quorum
otherwarn about key copies such hosts keep for their own use, for example of 'client.admin', and print the commands to refresh them