vrr/cozystack
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fix(linstor): preserve TCP ports during toggle-disk operations

Browse source 
Update fix-duplicate-tcp-ports patch to preserve existing TCP ports when
DrbdRscData is recreated during toggle-disk operations. Without this,
removeLayerData() frees ports and ensureStackDataExists() may allocate
different ones, causing port mismatches between controller and satellites
if the satellite misses the update.

Also add dh_strip_nondeterminism override in Dockerfile to fix build
failures on some JAR files.

Upstream: https://github.com/LINBIT/linstor-server/pull/476#issuecomment-4147527442

Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
This commit is contained in:
Andrei Kvapil 2026-03-28 08:31:56 +01:00
parent 5da23f42f7
commit 812d4138bb
No known key found for this signature in database
GPG key ID: 931CF7FEACEAF765
3 changed files with 112 additions and 59 deletions
Download patch file Download diff file Expand all files Collapse all files

2
packages/system/linstor/images/piraeus-server/Dockerfile
View file

@ -61,6 +61,8 @@ RUN test -d .gradlehome && echo ".gradlehome found in tarball" || (echo ".gradle
# Build DEB packages from tarball
# Override GRADLE_FLAGS to remove --offline flag, allowing Gradle to download missing dependencies
RUN sed -i 's/GRADLE_FLAGS = --offline/GRADLE_FLAGS =/' debian/rules || true
# Skip dh_strip_nondeterminism to avoid failures on some JAR files (logback-core)
RUN printf '\noverride_dh_strip_nondeterminism:\n\ttrue\n' >> debian/rules
RUN LD_LIBRARY_PATH='' dpkg-buildpackage -rfakeroot -b -uc
# Copy built .deb packages to a location accessible from final image

4
packages/system/linstor/images/piraeus-server/patches/README.md
View file

@ -8,7 +8,7 @@ Custom patches for piraeus-server (linstor-server) v1.32.3.
- Upstream: [#475](https://github.com/LINBIT/linstor-server/pull/475)
- **force-metadata-check-on-disk-add.diff** — Create metadata during toggle-disk from diskless to diskful
- Upstream: [#474](https://github.com/LINBIT/linstor-server/pull/474)
- **fix-duplicate-tcp-ports.diff** — Prevent duplicate TCP ports after toggle-disk operations
- Upstream: [#476](https://github.com/LINBIT/linstor-server/pull/476)
- **fix-duplicate-tcp-ports.diff** — Preserve TCP ports during toggle-disk to prevent port mismatch between controller and satellites
- Upstream: [#476](https://github.com/LINBIT/linstor-server/pull/476) (superseded by this expanded fix)
- **skip-adjust-when-device-inaccessible.diff** — Fix resources stuck in StandAlone after reboot, Unknown state race condition, and encrypted resource deletion
- Upstream: [#477](https://github.com/LINBIT/linstor-server/pull/477)

165
packages/system/linstor/images/piraeus-server/patches/fix-duplicate-tcp-ports.diff
View file

@ -1,80 +1,131 @@
From 1250abe99d64a0501795e37d3b6af62410002239 Mon Sep 17 00:00:00 2001
From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: Andrei Kvapil <kvapss@gmail.com>
Date: Mon, 12 Jan 2026 13:44:46 +0100
Subject: [PATCH] fix(drbd): prevent duplicate TCP ports after toggle-disk
operations
Date: Fri, 28 Mar 2026 13:00:00 +0100
Subject: [PATCH] fix(drbd): preserve TCP ports during toggle-disk operations
Remove redundant ensureStackDataExists() call with empty payload from
resetStoragePools() method that was causing TCP port conflicts after
toggle-disk operations.
Prevent TCP port mismatches after toggle-disk operations by preserving
existing TCP ports when rebuilding DrbdRscData.
Root Cause:
-----------
The resetStoragePools() method, introduced in 2019 (commit 95cc17d0b8),
calls ensureStackDataExists() with an empty LayerPayload. This worked
correctly when TCP ports were stored at RscDfn level.
During toggle-disk operations, removeLayerData() deletes DrbdRscData
(freeing its TCP ports from the number pool), then ensureStackDataExists()
creates new DrbdRscData. Since the payload has no explicit tcpPorts,
the controller allocates new ports from the pool -- which may differ from
the old ports if other resources claimed them in the meantime.
After the TCP port migration to per-node level (commit f754943463, May
2025), this empty payload results in DrbdRscData being created without
TCP ports assigned. The controller then sends a Pojo with an empty port
Set to satellites.
The controller correctly avoids collisions in its own number pool, but
the satellite may miss the update (e.g. during controller restart or
network issues). When this happens, the satellite keeps the old ports
while peers receive the new ones, causing DRBD connection failures
(StandAlone/Connecting state).
On satellites, when DrbdRscData is initialized with an empty port list,
initPorts() uses preferredNewPortsRef from peer resources. Since
SatelliteDynamicNumberPool.tryAllocate() always returns true (no-op),
any port from preferredNewPortsRef is accepted without conflict checking,
leading to duplicate TCP port assignments.
Impact:
-------
This regression affects toggle-disk operations, particularly:
- Snapshot creation/restore operations
- Manual toggle-disk operations
- Any operation calling resetStoragePools()
Symptoms include:
- DRBD resources failing to adjust with "port is also used" errors
- Resources stuck in StandAlone or Connecting states
- Multiple resources on the same node using identical TCP ports
Additionally, remove the redundant ensureStackDataExists() call from
resetStoragePools() -- the caller already invokes it with the correct
payload.
Solution:
---------
Remove the ensureStackDataExists() call from resetStoragePools() as it
is redundant. The calling code (e.g., CtrlRscToggleDiskApiCallHandler
line 1071) already invokes ensureStackDataExists() with the correct
payload immediately after resetStoragePools().
1. Add copyDrbdTcpPortsIfExists() to save existing TCP ports into the
LayerPayload before removeLayerData() deletes them.
2. Call it from copyDrbdNodeIdIfExists() (covers both toggle-disk paths)
and from the needsDeactivate path (shared storage pool case).
3. Remove the redundant ensureStackDataExists() from resetStoragePools().
This fix ensures:
1. resetStoragePools() only resets storage pool assignments
2. Layer data creation with proper TCP ports happens via the caller's
ensureStackDataExists() with correct payload
3. No DrbdRscData objects are created without TCP port assignments
Related Issues:
---------------
Fixes #454 - Duplicate TCP ports after backup/restore operations
Related to user reports of resources stuck in StandAlone after node
reboots when toggle-disk or backup operations were in progress.
Testing:
--------
Verified that:
- Toggle-disk operations no longer create resources without TCP ports
- Backup/restore operations complete without TCP port conflicts
- Resources maintain unique TCP ports across toggle-disk cycles
This ensures the same TCP ports are reused when DrbdRscData is recreated,
eliminating the window for port mismatch between controller and satellites.
Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: Andrei Kvapil <kvapss@gmail.com>
---
.../linbit/linstor/layer/resource/CtrlRscLayerDataFactory.java | 2 --
1 file changed, 2 deletions(-)
.../controller/CtrlRscToggleDiskApiCallHandler.java | 40 +++++++++++++++++++--
.../linstor/layer/resource/CtrlRscLayerDataFactory.java | 2 --
2 files changed, 38 insertions(+), 4 deletions(-)
diff --git a/controller/src/main/java/com/linbit/linstor/core/apicallhandler/controller/CtrlRscToggleDiskApiCallHandler.java b/controller/src/main/java/com/linbit/linstor/core/apicallhandler/controller/CtrlRscToggleDiskApiCallHandler.java
index ccdb0cee5..b0554c2ec 100644
--- a/controller/src/main/java/com/linbit/linstor/core/apicallhandler/controller/CtrlRscToggleDiskApiCallHandler.java
+++ b/controller/src/main/java/com/linbit/linstor/core/apicallhandler/controller/CtrlRscToggleDiskApiCallHandler.java
@@ -58,6 +58,7 @@ import com.linbit.linstor.stateflags.StateFlags;
import com.linbit.linstor.storage.StorageException;
import com.linbit.linstor.storage.data.adapter.drbd.DrbdRscData;
import com.linbit.linstor.storage.interfaces.categories.resource.AbsRscLayerObject;
+import com.linbit.linstor.core.types.TcpPortNumber;
import com.linbit.linstor.storage.interfaces.categories.resource.VlmProviderObject;
import com.linbit.linstor.storage.kinds.DeviceLayerKind;
import com.linbit.linstor.storage.kinds.DeviceProviderKind;
@@ -88,6 +89,7 @@ import java.util.LinkedHashMap;
import java.util.List;
import java.util.Map.Entry;
import java.util.Set;
+import java.util.TreeSet;
import org.reactivestreams.Publisher;
import reactor.core.publisher.Flux;
@@ -587,8 +589,9 @@ public class CtrlRscToggleDiskApiCallHandler implements CtrlSatelliteConnectionL
/*
* We also have to remove the currently diskless DrbdRscData and free up the node-id as now we must
- * use the shared resource's node-id
+ * use the shared resource's node-id. We still need to preserve TCP ports though.
*/
+ copyDrbdTcpPortsIfExists(rsc, payload);
}
else
{
@@ -726,7 +729,7 @@ public class CtrlRscToggleDiskApiCallHandler implements CtrlSatelliteConnectionL
/**
* Although we need to rebuild the layerData as the layerList might have changed, if we do not
* deactivate (i.e. down) the current resource, we need to make sure that deleting DrbdRscData
- * and recreating a new DrbdRscData ends up with the same node-id as before.
+ * and recreating a new DrbdRscData ends up with the same node-id and TCP ports as before.
*/
private void copyDrbdNodeIdIfExists(Resource rsc, LayerPayload payload) throws ImplementationError
{
@@ -743,6 +746,37 @@ public class CtrlRscToggleDiskApiCallHandler implements CtrlSatelliteConnectionL
DrbdRscData<Resource> drbdRscData = (DrbdRscData<Resource>) drbdRscDataSet.iterator().next();
payload.drbdRsc.nodeId = drbdRscData.getNodeId().value;
}
+ copyDrbdTcpPortsIfExists(rsc, payload);
+ }
+
+ /**
+ * Preserves existing TCP ports during toggle-disk operations.
+ *
+ * When removeLayerData() deletes DrbdRscData, the TCP ports are freed from the number pool.
+ * If ensureStackDataExists() then allocates different ports, and the satellite misses the update
+ * (e.g. due to controller restart or connectivity issues), the satellite keeps the old ports
+ * while peers get the new ones, causing DRBD connections to fail with StandAlone state.
+ */
+ private void copyDrbdTcpPortsIfExists(Resource rsc, LayerPayload payload) throws ImplementationError
+ {
+ Set<AbsRscLayerObject<Resource>> drbdRscDataSet = LayerRscUtils.getRscDataByLayer(
+ getLayerData(apiCtx, rsc),
+ DeviceLayerKind.DRBD
+ );
+ if (!drbdRscDataSet.isEmpty())
+ {
+ DrbdRscData<Resource> drbdRscData = (DrbdRscData<Resource>) drbdRscDataSet.iterator().next();
+ Collection<TcpPortNumber> tcpPorts = drbdRscData.getTcpPortList();
+ if (tcpPorts != null && !tcpPorts.isEmpty())
+ {
+ Set<Integer> portInts = new TreeSet<>();
+ for (TcpPortNumber port : tcpPorts)
+ {
+ portInts.add(port.value);
+ }
+ payload.drbdRsc.tcpPorts = portInts;
+ }
+ }
}
private List<DeviceLayerKind> removeLayerData(Resource rscRef)
diff --git a/controller/src/main/java/com/linbit/linstor/layer/resource/CtrlRscLayerDataFactory.java b/controller/src/main/java/com/linbit/linstor/layer/resource/CtrlRscLayerDataFactory.java
index 3538b380c..4f589145e 100644
--- a/controller/src/main/java/com/linbit/linstor/layer/resource/CtrlRscLayerDataFactory.java
+++ b/controller/src/main/java/com/linbit/linstor/layer/resource/CtrlRscLayerDataFactory.java
@@ -276,8 +276,6 @@ public class CtrlRscLayerDataFactory
rscDataToProcess.addAll(rscData.getChildren());
}
-
@ -82,6 +133,6 @@ index 3538b380c..4f589145e 100644
}
catch (AccessDeniedException exc)
{
--
--
2.39.5 (Apple Git-154)