]> git.ipfire.org Git - thirdparty/bind9.git/commitdiff
Simplify the parent cleaning in the prune_tree() mechanism
authorOndřej Surý <ondrej@isc.org>
Wed, 21 Feb 2024 12:32:09 +0000 (13:32 +0100)
committerMichał Kępień <michal@isc.org>
Thu, 29 Feb 2024 11:06:56 +0000 (12:06 +0100)
Instead of juggling with node locks in a cycle, cleanup the node we are
just pruning and send any the parent that's also subject to the pruning
to the prune tree via normal way (e.g. enqueue pruning on the parent).

This simplifies the code and also spreads the pruning load across more
event loop ticks which is better for lock contention as less things run
in a tight loop.

(cherry picked from commit 0b32d323e07d2468111aa725b36e72b0538e0521)

lib/dns/rbtdb.c

index ad8533ebaf02dd735d05a044cf3429d43f6008c8..00bf3cef992caba2afa8369ad723f22b3cf29fa7 100644 (file)
@@ -1847,8 +1847,9 @@ delete_node(dns_rbtdb_t *rbtdb, dns_rbtnode_t *node) {
  */
 static void
 new_reference(dns_rbtdb_t *rbtdb, dns_rbtnode_t *node,
-             isc_rwlocktype_t locktype) {
-       if (locktype == isc_rwlocktype_write && ISC_LINK_LINKED(node, deadlink))
+             isc_rwlocktype_t nlocktype) {
+       if (nlocktype == isc_rwlocktype_write &&
+           ISC_LINK_LINKED(node, deadlink))
        {
                ISC_LIST_UNLINK(rbtdb->deadnodes[node->locknum], node,
                                deadlink);
@@ -1871,13 +1872,13 @@ is_leaf(dns_rbtnode_t *node) {
 
 static void
 send_to_prune_tree(dns_rbtdb_t *rbtdb, dns_rbtnode_t *node,
-                  isc_rwlocktype_t locktype) {
+                  isc_rwlocktype_t nlocktype) {
        isc_event_t *ev;
        dns_db_t *db;
 
        ev = isc_event_allocate(rbtdb->common.mctx, NULL, DNS_EVENT_RBTPRUNE,
                                prune_tree, node, sizeof(isc_event_t));
-       new_reference(rbtdb, node, locktype);
+       new_reference(rbtdb, node, nlocktype);
        db = NULL;
        attach((dns_db_t *)rbtdb, &db);
        ev->ev_sender = db;
@@ -2097,10 +2098,6 @@ decrement_reference(dns_rbtdb_t *rbtdb, dns_rbtnode_t *node,
 #undef KEEP_NODE
 
        if (write_locked) {
-               /*
-                * We can now delete the node.
-                */
-
                /*
                 * If this node is the only one in the level it's in, deleting
                 * this node may recursively make its parent the only node in
@@ -2121,6 +2118,7 @@ decrement_reference(dns_rbtdb_t *rbtdb, dns_rbtnode_t *node,
                        send_to_prune_tree(rbtdb, node, isc_rwlocktype_write);
                        no_reference = false;
                } else {
+                       /*  We can now delete the node. */
                        delete_node(rbtdb, node);
                }
        } else {
@@ -2156,62 +2154,39 @@ restore_locks:
 }
 
 /*
- * Prune the tree by recursively cleaning-up single leaves.  In the worst
- * case, the number of iteration is the number of tree levels, which is at
- * most the maximum number of domain name labels, i.e, 127.  In practice, this
- * should be much smaller (only a few times), and even the worst case would be
- * acceptable for a single event.
+ * Prune the tree by cleaning up single leaves.  A single execution of this
+ * function cleans up a single node; if the parent of the latter becomes a
+ * single leaf on its own level as a result, the parent is then also sent to
+ * this function.
  */
 static void
 prune_tree(isc_task_t *task, isc_event_t *event) {
        dns_rbtdb_t *rbtdb = event->ev_sender;
        dns_rbtnode_t *node = event->ev_arg;
-       dns_rbtnode_t *parent;
-       unsigned int locknum;
+       dns_rbtnode_t *parent = NULL;
+       unsigned int locknum = node->locknum;
 
        UNUSED(task);
 
        isc_event_free(&event);
 
        RWLOCK(&rbtdb->tree_lock, isc_rwlocktype_write);
-       locknum = node->locknum;
-       NODE_LOCK(&rbtdb->node_locks[locknum].lock, isc_rwlocktype_write);
-       do {
-               parent = node->parent;
-               decrement_reference(rbtdb, node, 0, isc_rwlocktype_write,
-                                   isc_rwlocktype_write, true);
-
-               if (parent != NULL && parent->down == NULL) {
-                       /*
-                        * node was the only down child of the parent and has
-                        * just been removed.  We'll then need to examine the
-                        * parent.  Keep the lock if possible; otherwise,
-                        * release the old lock and acquire one for the parent.
-                        */
-                       if (parent->locknum != locknum) {
-                               NODE_UNLOCK(&rbtdb->node_locks[locknum].lock,
-                                           isc_rwlocktype_write);
-                               locknum = parent->locknum;
-                               NODE_LOCK(&rbtdb->node_locks[locknum].lock,
-                                         isc_rwlocktype_write);
-                       }
 
-                       /*
-                        * We need to gain a reference to the node before
-                        * decrementing it in the next iteration.
-                        */
-                       if (ISC_LINK_LINKED(parent, deadlink)) {
-                               ISC_LIST_UNLINK(rbtdb->deadnodes[locknum],
-                                               parent, deadlink);
-                       }
-                       new_reference(rbtdb, parent, isc_rwlocktype_write);
-               } else {
-                       parent = NULL;
-               }
+       parent = node->parent;
 
-               node = parent;
-       } while (node != NULL);
+       NODE_LOCK(&rbtdb->node_locks[locknum].lock, isc_rwlocktype_write);
+       decrement_reference(rbtdb, node, 0, isc_rwlocktype_write,
+                           isc_rwlocktype_write, true);
        NODE_UNLOCK(&rbtdb->node_locks[locknum].lock, isc_rwlocktype_write);
+
+       if (parent != NULL && is_leaf(parent)) {
+               NODE_LOCK(&rbtdb->node_locks[parent->locknum].lock,
+                         isc_rwlocktype_write);
+               send_to_prune_tree(rbtdb, parent, isc_rwlocktype_write);
+               NODE_UNLOCK(&rbtdb->node_locks[parent->locknum].lock,
+                           isc_rwlocktype_write);
+       }
+
        RWUNLOCK(&rbtdb->tree_lock, isc_rwlocktype_write);
 
        detach((dns_db_t **)&rbtdb);