libxl: Fix lock manager lock ordering

The ordering of lock manager locks in the libxl driver has a flaw that was
uncovered by a migration error path. In the perform phase of migration, the
source host calls virDomainLockProcessPause to release the lock before
sending the VM to the destination host. If the send fails an attempt is made
to reacquire the lock with virDomainLockProcessResume, but that too can fail
if the destination host has not finished cleaning up the failed VM and
releasing the lock it acquired when starting to receive the VM.

This change delays calling virDomainLockProcessResume in libxlDomainStart
until the VM is successfully created, but before it is unpaused. A similar
approach is used by the qemu driver, avoiding the need to release the lock
if VM creation fails. In the migration perform phase, releasing the lock
with virDomainLockProcessPause is delayed until the VM is successfully
sent to the destination, which avoids reacquiring the lock if the send
fails.

Signed-off-by: Jim Fehlig <jfehlig@suse.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>

diff --git a/src/libxl/libxl_domain.c b/src/libxl/libxl_domain.c
index 514dbe21ac23687215da798281348923c004b397..e71faee63d2f29f056db8a5d65ca48457f496d08 100644 (file)
--- a/src/libxl/libxl_domain.c
+++ b/src/libxl/libxl_domain.c
@@ -1338,13 +1338,6 @@ libxlDomainStart(libxlDriverPrivatePtr driver,
                                   NULL) < 0)
         goto cleanup;
 
-    if (virDomainLockProcessResume(driver->lockManager,
-                                  "xen:///system",
-                                  vm,
-                                  priv->lockState) < 0)
-        goto cleanup;
-    VIR_FREE(priv->lockState);
-
     if (libxlNetworkPrepareDevices(vm->def) < 0)
         goto cleanup_dom;
 
@@ -1427,6 +1420,13 @@ libxlDomainStart(libxlDriverPrivatePtr driver,
 
     libxlLoggerOpenFile(cfg->logger, domid, vm->def->name, config_json);
 
+    if (virDomainLockProcessResume(driver->lockManager,
+                                  "xen:///system",
+                                  vm,
+                                  priv->lockState) < 0)
+        goto destroy_dom;
+    VIR_FREE(priv->lockState);
+
     /* Always enable domain death events */
     if (libxl_evenable_domain_death(cfg->ctx, vm->def->id, 0, &priv->deathW))
         goto destroy_dom;

diff --git a/src/libxl/libxl_migration.c b/src/libxl/libxl_migration.c
index b223916647707d9950221c5785ac7a1e2724ec75..7e920050c42b52d5b3243f064827648bfbc13cc2 100644 (file)
--- a/src/libxl/libxl_migration.c
+++ b/src/libxl/libxl_migration.c
@@ -1237,20 +1237,16 @@ libxlDomainMigrationSrcPerform(libxlDriverPrivatePtr driver,
     sockfd = virNetSocketDupFD(sock, true);
     virObjectUnref(sock);
 
-    if (virDomainLockProcessPause(driver->lockManager, vm, &priv->lockState) < 0)
-        VIR_WARN("Unable to release lease on %s", vm->def->name);
-    VIR_DEBUG("Preserving lock state '%s'", NULLSTR(priv->lockState));
-
     /* suspend vm and send saved data to dst through socket fd */
     virObjectUnlock(vm);
     ret = libxlDoMigrateSrcSend(driver, vm, flags, sockfd);
     virObjectLock(vm);
 
-    if (ret < 0) {
-        virDomainLockProcessResume(driver->lockManager,
-                                   "xen:///system",
-                                   vm,
-                                   priv->lockState);
+    if (ret == 0) {
+        if (virDomainLockProcessPause(driver->lockManager, vm, &priv->lockState) < 0)
+            VIR_WARN("Unable to release lease on %s", vm->def->name);
+        VIR_DEBUG("Preserving lock state '%s'", NULLSTR(priv->lockState));
+    } else {
         /*
          * Confirm phase will not be executed if perform fails. End the
          * job started in begin phase.