Rebuilding libvirt on Oracle Linux 9 with CLVM + sanlock patch

Background
Prerequisites and repository setup
Download and install source RPM
Enter build tree and install build dependencies
Add your custom patch to SOURCES
Reference patch in the spec file
Optional: adjust Release field
Build the RPM
Install rebuilt package
Restart libvirt service
Quick verification checklist

In this guide, I walk through rebuilding libvirt-daemon-driver-storage-logical on Oracle Linux 9 and applying a custom CLVM + sanlock patch.

The goal is to produce a rebuilt RPM that can be installed cleanly on Oracle Linux 9 systems while preserving the distro packaging workflow.

In clustered environments, CLVM with lvmlockd and sanlock is a common approach for shared storage. In practice, libvirt’s logical pool behavior can conflict with that model when one host encounters an LV lock held by another host. This patch targets a practical operating model: shared VG visibility with exclusive LV activation (except migration workflows coordinated by an external control plane).

Background

lvmlockd + sanlock is widely used to let multiple hosts access the same VG safely.

The issue appears when libvirt creates or activates a logical storage pool. If any LV in the shared VG is already active and locked by another host, libvirt may fail the pool operation instead of tolerating that lock state.

Typical failure example:

error: Failed to create pool from pool.xml
error: internal error: Child process (/usr/sbin/vgchange -aly vg_shared_sdb) unexpected exit status 5:   LV locked by other host: vg_shared_sdb/vm2
  Failed to lock logical volume vg_shared_sdb/vm2.

The patch in this post is designed around this goal:

Keep the VG shared and visible to all cluster hosts.
Activate VM LVs exclusively on the host currently using them.
Leave VM migration coordination to an external control plane.

Prerequisites and repository setup

dnf install rpm-build
dnf config-manager --enable ol9_codeready_builder
dnf makecache

Installs rpm-build so rpmbuild and the RPM toolchain are available for compiling the SRPM.
Enables ol9_codeready_builder so required -devel and other build dependencies can be resolved.
Refreshes DNF metadata so dependency lookups use current repository data and avoid stale package errors.

Download and install source RPM

dnf download --source libvirt-daemon-driver-storage-logical-9.0.0-10.3.0.1.el9_2.x86_64
rpm -ivh libvirt-9.0.0-10.3.0.1.el9_2.src.rpm

Downloads the matching libvirt source RPM so you rebuild from the official packaging baseline.
Installs the SRPM into ~/rpmbuild/ (including SPECS/ and SOURCES/) so the standard rpmbuild workflow can be used.

Enter build tree and install build dependencies

cd rpmbuild
rpmbuild -bb SPECS/libvirt.spec 2>&1 | grep "is needed" | awk '{print $1}' | tr "\n" " " | xargs dnf install -y

Triggers a build attempt, extracts missing package names from errors, and installs them automatically.
This quickly closes libvirt’s large dependency chain so the next real build can proceed with fewer manual iterations.

Tip: This is a practical bootstrap trick. In stricter build environments, you may prefer explicit dnf builddep workflows.

Add your custom patch to SOURCES

Copy your patch file into ~/rpmbuild/SOURCES/.

Important: the patch header must match proper git format-patch style, for example:

From b747ecf4a0bf14fa00bf3afc375fc3b2e61c523c Mon Sep 17 00:00:00 2001
From: Cloud ABC <blog@cloudabc.eu>
Date: Sat, 28 Feb 2026 14:55:22 +0100
Subject: [PATCH] CLVM: shared VG and exclusive LVs

---
 src/storage/storage_backend_logical.c | 72 ++++++++++++++++++++++++++-
 1 file changed, 71 insertions(+), 1 deletion(-)

diff --git a/src/storage/storage_backend_logical.c b/src/storage/storage_backend_logical.c
index 6acbc37f18..2de54b4574 100644
--- a/src/storage/storage_backend_logical.c
+++ b/src/storage/storage_backend_logical.c
@@ -41,15 +41,74 @@
 VIR_LOG_INIT("storage.storage_backend_logical");
 
 
+/* Add a cached cluster flag to pool struct */
+typedef struct _LogicalPoolPrivate {
+    bool vg_clustered;
+    bool vg_clustered_valid;
+} LogicalPoolPrivate;
+
+/* Helper: detect if VG uses cluster lock (sanlock/dlm) */
+static bool
+virStorageBackendLogicalVGIsClusteredCached(LogicalPoolPrivate *priv,
+                                            const char *vgname)
+{
+    g_autoptr(virCommand) cmd = NULL;
+    g_autofree char *output = NULL;
+
+    if (priv->vg_clustered_valid)
+        return priv->vg_clustered;
+
+    cmd = virCommandNewArgList("vgs",
+                               "--noheadings",
+                               "--readonly",
+                               "-o", "locktype",
+                               vgname,
+                               NULL);
+    virCommandSetOutputBuffer(cmd, &output);
+
+    if (virCommandRun(cmd, NULL) < 0 || !output) {
+        priv->vg_clustered = false;
+        priv->vg_clustered_valid = true;
+        return false;
+    }
+
+    g_strstrip(output);
+    priv->vg_clustered = (STREQ(output, "sanlock") || STREQ(output, "dlm"));
+    priv->vg_clustered_valid = true;
+    return priv->vg_clustered;
+}
+
+/* LV activation flag for VM disks */
+static const char *
+virStorageBackendLogicalGetLVActivateFlag(LogicalPoolPrivate *priv,
+                                          const char *vgname)
+{
+    if (virStorageBackendLogicalVGIsClusteredCached(priv, vgname))
+        return "-aey";   /* exclusive activation for VM disks */
+    else
+        return "-ay";    /* default for non-clustered VGs */
+}
+
 static int
 virStorageBackendLogicalSetActive(virStoragePoolObj *pool,
                                   bool on)
 {
     virStoragePoolDef *def = virStoragePoolObjGetDef(pool);
     g_autoptr(virCommand) cmd = NULL;
+    LogicalPoolPrivate priv = { 0 };
+    const char *activateFlag;
     int ret;
 
-    cmd = virStorageBackendLogicalChangeCmd(VGCHANGE, def, on);
+    if (on) {
+        if (virStorageBackendLogicalVGIsClusteredCached(&priv, def->source.name))
+            activateFlag = "-asy";
+        else
+            activateFlag = "-aly";
+    } else {
+        activateFlag = "-aln";
+    }
+
+    cmd = virCommandNewArgList(VGCHANGE, activateFlag, def->source.name, NULL);
 
     virObjectUnlock(pool);
     ret = virCommandRun(cmd, NULL);
@@ -860,7 +919,10 @@ virStorageBackendLogicalCreateVol(virStoragePoolObj *pool,
                                   virStorageVolDef *vol)
 {
     virStoragePoolDef *def = virStoragePoolObjGetDef(pool);
+    g_autoptr(virCommand) lvchange_cmd = NULL;
     virErrorPtr err;
+    LogicalPoolPrivate priv = { 0 };
+    const char *lvActivateFlag;
     struct stat sb;
     VIR_AUTOCLOSE fd = -1;
 
@@ -872,6 +934,14 @@ virStorageBackendLogicalCreateVol(virStoragePoolObj *pool,
     if (virStorageBackendLogicalLVCreate(vol, def) < 0)
         return -1;
 
+    lvActivateFlag = virStorageBackendLogicalGetLVActivateFlag(&priv, def->source.name);
+    lvchange_cmd = virCommandNewArgList(LVCHANGE,
+                                        lvActivateFlag,
+                                        vol->target.path,
+                                        NULL);
+    if (virCommandRun(lvchange_cmd, NULL) < 0)
+        goto error;
+
     if (vol->target.encryption &&
         virStorageBackendCreateVolUsingQemuImg(pool, vol, NULL, 0) < 0)
         goto error;
-- 
2.43.0

Places the patch in SOURCES/ where RPM spec processing expects it.
Keeps git format-patch style metadata so %patch application is reliable and the change remains traceable.

Reference patch in the spec file

Edit ~/rpmbuild/SPECS/libvirt.spec and add a patch declaration, for example:

Patch1003: clvm.patch

Then ensure the patch is applied in the %prep section (pattern depends on the existing spec style).

Declares and applies the patch in libvirt.spec so it becomes part of the RPM build lifecycle.
This is required because files in SOURCES/ are ignored unless referenced, and it keeps future rebuilds reproducible.

Optional: adjust Release field

Example change:

Release: 10.3.0.1%{?dist}%{?extra_release}
# to
Release: 10.3.0.1.el9_2

Changes the RPM Release tag to match your internal naming/versioning convention.
This helps distinguish custom rebuilds and reduces ambiguity when multiple variants exist.

Build the RPM

rpmbuild -bb SPECS/libvirt.spec

Runs a binary RPM build (-bb) from the spec and sources.
This produces the installable package artifact under ~/rpmbuild/RPMS/x86_64/.

Build options reference: https://linux.die.net/man/8/rpmbuild

If tests fail during build,

rpmbuild -bb SPECS/libvirt.spec --nocheck

Uses --nocheck to skip the %check phase when tests block the build.
This is useful for environment-specific failures unrelated to your patch so you can still create a package for controlled validation.

Caution: skipping tests reduces build-time validation confidence. Prefer running checks when possible.

Install rebuilt package

rpm -Uvh RPMS/x86_64/libvirt-daemon-driver-storage-logical-9.0.0-10.3.0.1.el9_2.x86_64.rpm --reinstall

Example output:

Verifying...                          ################################# [100%]
Preparing...                          ################################# [100%]
Updating / installing...
   1:libvirt-daemon-driver-storage-log################################# [ 50%]
Cleaning up / removing...
   2:libvirt-daemon-driver-storage-log################################# [100%]

Installs your rebuilt RPM with upgrade/reinstall behavior.
This deploys the patched logical storage driver, and --reinstall supports intentional same-NEVRA replacement.

Restart libvirt service

systemctl restart libvirtd

Restarts libvirtd so the runtime loads updated binaries and libraries.
This is required for your CLVM + sanlock package changes to actually take effect.

Quick verification checklist

Confirm installed package version with rpm -qa | grep libvirt-daemon-driver-storage-logical.
Inspect libvirt logs after restart: journalctl -u libvirtd -b.
Validate the logical storage workflow that depends on CLVM/sanlock behavior.

Date: February 28, 2026

Tags: Linux Libvirt