qemu: Cancel disk mirrors after libvirtd restart

When libvirtd is restarted during migration, we properly cancel the
ongoing migration (unless it managed to almost finished before the
restart). But if we were also migrating storage using NBD, we would
completely forget about the running disk mirrors.

Signed-off-by: Jiri Denemark <jdenemar@redhat.com>

src/qemu/qemu_domain.c

@@ -578,7 +578,27 @@ qemuDomainObjPrivateXMLFormat(virBufferPtr buf,
                               qemuDomainAsyncJobPhaseToString(
                                     priv->job.asyncJob, priv->job.phase));
         }
-        virBufferAddLit(buf, "/>\n");
+        if (priv->job.asyncJob != QEMU_ASYNC_JOB_MIGRATION_OUT) {
+            virBufferAddLit(buf, "/>\n");
+        } else {
+            size_t i;
+            virDomainDiskDefPtr disk;
+            qemuDomainDiskPrivatePtr diskPriv;
+
+            virBufferAddLit(buf, ">\n");
+            virBufferAdjustIndent(buf, 2);
+
+            for (i = 0; i < vm->def->ndisks; i++) {
+                disk = vm->def->disks[i];
+                diskPriv = QEMU_DOMAIN_DISK_PRIVATE(disk);
+                virBufferAsprintf(buf, "<disk dev='%s' migrating='%s'/>\n",
+                                  disk->dst,
+                                  diskPriv->migrating ? "yes" : "no");
+            }
+
+            virBufferAdjustIndent(buf, -2);
+            virBufferAddLit(buf, "</job>\n");
+        }
     }
     priv->job.active = job;
 
@@ -736,6 +756,29 @@ qemuDomainObjPrivateXMLParse(xmlXPathContextPtr ctxt,
         }
     }
 
+    if ((n = virXPathNodeSet("./job[1]/disk[@migrating='yes']",
+                             ctxt, &nodes)) < 0) {
+        virReportError(VIR_ERR_INTERNAL_ERROR, "%s",
+                       _("failed to parse list of disks marked for migration"));
+        goto error;
+    }
+    if (n > 0) {
+        if (priv->job.asyncJob != QEMU_ASYNC_JOB_MIGRATION_OUT) {
+            VIR_WARN("Found disks marked for migration but we were not "
+                     "migrating");
+            n = 0;
+        }
+        for (i = 0; i < n; i++) {
+            char *dst = virXMLPropString(nodes[i], "dev");
+            virDomainDiskDefPtr disk;
+
+            if (dst && (disk = virDomainDiskByName(vm->def, dst, false)))
+                QEMU_DOMAIN_DISK_PRIVATE(disk)->migrating = true;
+            VIR_FREE(dst);
+        }
+    }
+    VIR_FREE(nodes);
+
     priv->fakeReboot = virXPathBoolean("boolean(./fakereboot)", ctxt) == 1;
 
     if ((n = virXPathNodeSet("./devices/device", ctxt, &nodes)) < 0) {

src/qemu/qemu_migration.c

@@ -2026,6 +2026,7 @@ qemuMigrationDriveMirror(virQEMUDriverPtr driver,
     char *hoststr = NULL;
     unsigned int mirror_flags = VIR_DOMAIN_BLOCK_REBASE_REUSE_EXT;
     int rv;
+    virQEMUDriverConfigPtr cfg = virQEMUDriverGetConfig(driver);
 
     VIR_DEBUG("Starting drive mirrors for domain %s", vm->def->name);
 
@@ -2075,6 +2076,11 @@ qemuMigrationDriveMirror(virQEMUDriverPtr driver,
             goto cleanup;
         }
         diskPriv->migrating = true;
+
+        if (virDomainSaveStatus(driver->xmlopt, cfg->stateDir, vm) < 0) {
+            VIR_WARN("Failed to save status on vm %s", vm->def->name);
+            goto cleanup;
+        }
     }
 
     while ((rv = qemuMigrationDriveMirrorReady(driver, vm)) != 1) {
@@ -2102,6 +2108,7 @@ qemuMigrationDriveMirror(virQEMUDriverPtr driver,
     ret = 0;
 
  cleanup:
+    virObjectUnref(cfg);
     VIR_FREE(diskAlias);
     VIR_FREE(nbd_dest);
     VIR_FREE(hoststr);
@@ -5817,6 +5824,84 @@ qemuMigrationToFile(virQEMUDriverPtr driver, virDomainObjPtr vm,
     return ret;
 }
 
+
+int
+qemuMigrationCancel(virQEMUDriverPtr driver,
+                    virDomainObjPtr vm)
+{
+    qemuDomainObjPrivatePtr priv = vm->privateData;
+    virHashTablePtr blockJobs = NULL;
+    bool storage = false;
+    size_t i;
+    int ret = -1;
+
+    VIR_DEBUG("Canceling unfinished outgoing migration of domain %s",
+              vm->def->name);
+
+    for (i = 0; i < vm->def->ndisks; i++) {
+        virDomainDiskDefPtr disk = vm->def->disks[i];
+        if (QEMU_DOMAIN_DISK_PRIVATE(disk)->migrating) {
+            qemuBlockJobSyncBegin(disk);
+            storage = true;
+        }
+    }
+
+    qemuDomainObjEnterMonitor(driver, vm);
+
+    ignore_value(qemuMonitorMigrateCancel(priv->mon));
+    if (storage)
+        blockJobs = qemuMonitorGetAllBlockJobInfo(priv->mon);
+
+    if (qemuDomainObjExitMonitor(driver, vm) < 0 || (storage && !blockJobs))
+        goto endsyncjob;
+
+    if (!storage) {
+        ret = 0;
+        goto cleanup;
+    }
+
+    for (i = 0; i < vm->def->ndisks; i++) {
+        virDomainDiskDefPtr disk = vm->def->disks[i];
+        qemuDomainDiskPrivatePtr diskPriv = QEMU_DOMAIN_DISK_PRIVATE(disk);
+
+        if (!diskPriv->migrating)
+            continue;
+
+        if (virHashLookup(blockJobs, disk->info.alias)) {
+            VIR_DEBUG("Drive mirror on disk %s is still running", disk->dst);
+        } else {
+            VIR_DEBUG("Drive mirror on disk %s is gone", disk->dst);
+            qemuBlockJobSyncEnd(driver, vm, disk);
+            diskPriv->migrating = false;
+        }
+    }
+
+    if (qemuMigrationCancelDriveMirror(driver, vm, false,
+                                       QEMU_ASYNC_JOB_NONE) < 0)
+        goto endsyncjob;
+
+    ret = 0;
+
+ cleanup:
+    virHashFree(blockJobs);
+    return ret;
+
+ endsyncjob:
+    if (storage) {
+        for (i = 0; i < vm->def->ndisks; i++) {
+            virDomainDiskDefPtr disk = vm->def->disks[i];
+            qemuDomainDiskPrivatePtr diskPriv = QEMU_DOMAIN_DISK_PRIVATE(disk);
+
+            if (diskPriv->migrating) {
+                qemuBlockJobSyncEnd(driver, vm, disk);
+                diskPriv->migrating = false;
+            }
+        }
+    }
+    goto cleanup;
+}
+
+
 int
 qemuMigrationJobStart(virQEMUDriverPtr driver,
                       virDomainObjPtr vm,

src/qemu/qemu_migration.h

@@ -185,4 +185,7 @@ int qemuMigrationToFile(virQEMUDriverPtr driver, virDomainObjPtr vm,
     ATTRIBUTE_NONNULL(1) ATTRIBUTE_NONNULL(2) ATTRIBUTE_NONNULL(5)
     ATTRIBUTE_RETURN_CHECK;
 
+int qemuMigrationCancel(virQEMUDriverPtr driver,
+                        virDomainObjPtr vm);
+
 #endif /* __QEMU_MIGRATION_H__ */

src/qemu/qemu_process.c

@@ -3354,8 +3354,6 @@ qemuProcessRecoverMigration(virQEMUDriverPtr driver,
                             virDomainState state,
                             int reason)
 {
-    qemuDomainObjPrivatePtr priv = vm->privateData;
-
     if (job == QEMU_ASYNC_JOB_MIGRATION_IN) {
         switch (phase) {
         case QEMU_MIGRATION_PHASE_NONE:
@@ -3409,11 +3407,7 @@ qemuProcessRecoverMigration(virQEMUDriverPtr driver,
         case QEMU_MIGRATION_PHASE_PERFORM3:
             /* migration is still in progress, let's cancel it and resume the
              * domain */
-            VIR_DEBUG("Canceling unfinished outgoing migration of domain %s",
-                      vm->def->name);
-            qemuDomainObjEnterMonitor(driver, vm);
-            ignore_value(qemuMonitorMigrateCancel(priv->mon));
-            if (qemuDomainObjExitMonitor(driver, vm) < 0)
+            if (qemuMigrationCancel(driver, vm) < 0)
                 return -1;
             /* resume the domain but only if it was paused as a result of
              * migration */