X-Git-Url: https://git.liburcu.org/?p=urcu.git;a=blobdiff_plain;f=rculfhash.c;h=034c7f9053cfbd881496b592cbb3bcddf92a1d8f;hp=074a6f4ea8c4c40f7b2f7374d3ecdc8387e34bff;hb=adc0de68ccd78689659ed49d4b3d5d36c6720e20;hpb=860d07e812c3001e9d6ce261c8b861dabd3203b3

diff --git a/rculfhash.c b/rculfhash.c
index 074a6f4..034c7f9 100644
--- a/rculfhash.c
+++ b/rculfhash.c
@@ -193,12 +193,16 @@
 
 /*
  * The removed flag needs to be updated atomically with the pointer.
+ * It indicates that no node must attach to the node scheduled for
+ * removal. The gc flag also needs to be updated atomically with the
+ * pointer. It indicates that node garbage collection must be performed.
  * The dummy flag does not require to be updated atomically with the
  * pointer, but it is added as a pointer low bit flag to save space.
  */
 #define REMOVED_FLAG		(1UL << 0)
-#define DUMMY_FLAG		(1UL << 1)
-#define FLAGS_MASK		((1UL << 2) - 1)
+#define GC_FLAG			(1UL << 1)
+#define DUMMY_FLAG		(1UL << 2)
+#define FLAGS_MASK		((1UL << 3) - 1)
 
 /* Value of the end pointer. Should not interact with flags. */
 #define END_VALUE		NULL
@@ -261,11 +265,22 @@ struct partition_resize_work {
 		    unsigned long start, unsigned long len);
 };
 
+enum add_mode {
+	ADD_DEFAULT = 0,
+	ADD_UNIQUE = 1,
+	ADD_REPLACE = 2,
+};
+
 static
 struct cds_lfht_node *_cds_lfht_add(struct cds_lfht *ht,
 				unsigned long size,
 				struct cds_lfht_node *node,
-				int unique, int dummy);
+				enum add_mode mode, int dummy);
+
+static
+int _cds_lfht_del(struct cds_lfht *ht, unsigned long size,
+		struct cds_lfht_node *node,
+		int dummy_removal, int do_gc);
 
 /*
  * Algorithm to reverse bits in a word by lookup table, extended to
@@ -661,6 +676,18 @@ struct cds_lfht_node *flag_removed(struct cds_lfht_node *node)
 	return (struct cds_lfht_node *) (((unsigned long) node) | REMOVED_FLAG);
 }
 
+static
+int is_gc(struct cds_lfht_node *node)
+{
+	return ((unsigned long) node) & GC_FLAG;
+}
+
+static
+struct cds_lfht_node *flag_gc(struct cds_lfht_node *node)
+{
+	return (struct cds_lfht_node *) (((unsigned long) node) | GC_FLAG);
+}
+
 static
 int is_dummy(struct cds_lfht_node *node)
 {
@@ -716,8 +743,10 @@ void _cds_lfht_gc_bucket(struct cds_lfht_node *dummy, struct cds_lfht_node *node
 	struct cds_lfht_node *iter_prev, *iter, *next, *new_next;
 
 	assert(!is_dummy(dummy));
+	assert(!is_gc(dummy));
 	assert(!is_removed(dummy));
 	assert(!is_dummy(node));
+	assert(!is_gc(node));
 	assert(!is_removed(node));
 	for (;;) {
 		iter_prev = dummy;
@@ -737,16 +766,18 @@ void _cds_lfht_gc_bucket(struct cds_lfht_node *dummy, struct cds_lfht_node *node
 			if (likely(clear_flag(iter)->p.reverse_hash > node->p.reverse_hash))
 				return;
 			next = rcu_dereference(clear_flag(iter)->p.next);
-			if (likely(is_removed(next)))
+			if (likely(is_gc(next)))
 				break;
 			iter_prev = clear_flag(iter);
 			iter = next;
 		}
-		assert(!is_removed(iter));
+		assert(!is_gc(iter));
 		if (is_dummy(iter))
 			new_next = flag_dummy(clear_flag(next));
 		else
 			new_next = clear_flag(next);
+		if (is_removed(iter))
+			new_next = flag_removed(new_next);
 		(void) uatomic_cmpxchg(&iter_prev->p.next, iter, new_next);
 	}
 	return;
@@ -756,14 +787,15 @@ static
 struct cds_lfht_node *_cds_lfht_add(struct cds_lfht *ht,
 				unsigned long size,
 				struct cds_lfht_node *node,
-				int unique, int dummy)
+				enum add_mode mode, int dummy)
 {
 	struct cds_lfht_node *iter_prev, *iter, *next, *new_node, *new_next,
-			*dummy_node;
+			*dummy_node, *return_node;
 	struct _cds_lfht_node *lookup;
 	unsigned long hash, index, order;
 
 	assert(!is_dummy(node));
+	assert(!is_gc(node));
 	assert(!is_removed(node));
 	if (!size) {
 		assert(dummy);
@@ -791,14 +823,19 @@ struct cds_lfht_node *_cds_lfht_add(struct cds_lfht *ht,
 			if (likely(clear_flag(iter)->p.reverse_hash > node->p.reverse_hash))
 				goto insert;
 			next = rcu_dereference(clear_flag(iter)->p.next);
-			if (unlikely(is_removed(next)))
+			if (unlikely(is_gc(next)))
 				goto gc_node;
-			if (unique
+			assert(!is_removed(next));
+			if ((mode == ADD_UNIQUE || mode == ADD_REPLACE)
 			    && !is_dummy(next)
 			    && !ht->compare_fct(node->key, node->key_len,
 						clear_flag(iter)->key,
-						clear_flag(iter)->key_len))
-				return clear_flag(iter);
+						clear_flag(iter)->key_len)) {
+				if (mode == ADD_UNIQUE)
+					return clear_flag(iter);
+				else /* mode == ADD_REPLACE */
+					goto replace;
+			}
 			/* Only account for identical reverse hash once */
 			if (iter_prev->p.reverse_hash != clear_flag(iter)->p.reverse_hash
 			    && !is_dummy(next))
@@ -806,10 +843,13 @@ struct cds_lfht_node *_cds_lfht_add(struct cds_lfht *ht,
 			iter_prev = clear_flag(iter);
 			iter = next;
 		}
+
 	insert:
 		assert(node != clear_flag(iter));
 		assert(!is_removed(iter_prev));
 		assert(!is_removed(iter));
+		assert(!is_gc(iter_prev));
+		assert(!is_gc(iter));
 		assert(iter_prev != node);
 		if (!dummy)
 			node->p.next = clear_flag(iter);
@@ -820,12 +860,55 @@ struct cds_lfht_node *_cds_lfht_add(struct cds_lfht *ht,
 		else
 			new_node = node;
 		if (uatomic_cmpxchg(&iter_prev->p.next, iter,
-				    new_node) != iter)
+				    new_node) != iter) {
 			continue;	/* retry */
+		} else {
+			if (mode == ADD_REPLACE)
+				return_node = NULL;
+			else	/* ADD_DEFAULT and ADD_UNIQUE */
+				return_node = node;
+			goto gc_end;
+		}
+
+	replace:
+		/* Insert after node to be replaced */
+		iter_prev = clear_flag(iter);
+		iter = next;
+		assert(node != clear_flag(iter));
+		assert(!is_removed(iter_prev));
+		assert(!is_removed(iter));
+		assert(!is_gc(iter_prev));
+		assert(!is_gc(iter));
+		assert(iter_prev != node);
+		assert(!dummy);
+		node->p.next = clear_flag(iter);
+		if (is_dummy(iter))
+			new_node = flag_dummy(node);
 		else
+			new_node = node;
+		/*
+		 * Here is the whole trick for lock-free replace: we add
+		 * the replacement node _after_ the node we want to
+		 * replace by atomically setting its next pointer at the
+		 * same time we set its removal and gc flags. Given that
+		 * the lookups/get next use an iterator aware of the
+		 * next pointer, they will either skip the old node due
+		 * to the removal/gc flag and see the new node, or use
+		 * the old new, but will not see the new one.
+		 */
+		new_node = flag_removed(new_node);
+		new_node = flag_gc(new_node);
+		if (uatomic_cmpxchg(&iter_prev->p.next,
+			      iter, new_node) != iter) {
+			continue;	/* retry */
+		} else {
+			return_node = iter_prev;
 			goto gc_end;
+		}
+
 	gc_node:
 		assert(!is_removed(iter));
+		assert(!is_gc(iter));
 		if (is_dummy(iter))
 			new_next = flag_dummy(clear_flag(next));
 		else
@@ -840,13 +923,13 @@ gc_end:
 	lookup = &ht->t.tbl[order]->nodes[index & (!order ? 0 : ((1UL << (order - 1)) - 1))];
 	dummy_node = (struct cds_lfht_node *) lookup;
 	_cds_lfht_gc_bucket(dummy_node, node);
-	return node;
+	return return_node;
 }
 
 static
 int _cds_lfht_del(struct cds_lfht *ht, unsigned long size,
 		struct cds_lfht_node *node,
-		int dummy_removal)
+		int dummy_removal, int do_gc)
 {
 	struct cds_lfht_node *dummy, *next, *old;
 	struct _cds_lfht_node *lookup;
@@ -855,9 +938,12 @@ int _cds_lfht_del(struct cds_lfht *ht, unsigned long size,
 
 	/* logically delete the node */
 	assert(!is_dummy(node));
+	assert(!is_gc(node));
 	assert(!is_removed(node));
 	old = rcu_dereference(node->p.next);
 	do {
+		struct cds_lfht_node *new_next;
+
 		next = old;
 		if (unlikely(is_removed(next)))
 			goto end;
@@ -865,13 +951,18 @@ int _cds_lfht_del(struct cds_lfht *ht, unsigned long size,
 			assert(is_dummy(next));
 		else
 			assert(!is_dummy(next));
-		old = uatomic_cmpxchg(&node->p.next, next,
-				      flag_removed(next));
+		new_next = flag_removed(next);
+		if (do_gc)
+			new_next = flag_gc(new_next);
+		old = uatomic_cmpxchg(&node->p.next, next, new_next);
 	} while (old != next);
 
 	/* We performed the (logical) deletion. */
 	flagged = 1;
 
+	if (!do_gc)
+		goto end;
+
 	/*
 	 * Ensure that the node is not visible to readers anymore: lookup for
 	 * the node, and remove it (along with any other logically removed node)
@@ -975,7 +1066,7 @@ void init_table_populate_partition(struct cds_lfht *ht, unsigned long i,
 		new_node->p.reverse_hash =
 			bit_reverse_ulong(!i ? 0 : (1UL << (i - 1)) + j);
 		(void) _cds_lfht_add(ht, !i ? 0 : (1UL << (i - 1)),
-				new_node, 0, 1);
+				new_node, ADD_DEFAULT, 1);
 		if (CMM_LOAD_SHARED(ht->in_progress_destroy))
 			break;
 	}
@@ -1078,7 +1169,7 @@ void remove_table_partition(struct cds_lfht *ht, unsigned long i,
 		fini_node->p.reverse_hash =
 			bit_reverse_ulong(!i ? 0 : (1UL << (i - 1)) + j);
 		(void) _cds_lfht_del(ht, !i ? 0 : (1UL << (i - 1)),
-				fini_node, 1);
+				fini_node, 1, 1);
 		if (CMM_LOAD_SHARED(ht->in_progress_destroy))
 			break;
 	}
@@ -1196,7 +1287,8 @@ struct cds_lfht *_cds_lfht_new(cds_lfht_hash_fct hash_fct,
 	return ht;
 }
 
-struct cds_lfht_node *cds_lfht_lookup(struct cds_lfht *ht, void *key, size_t key_len)
+void cds_lfht_lookup(struct cds_lfht *ht, void *key, size_t key_len,
+		struct cds_lfht_iter *iter)
 {
 	struct cds_lfht_node *node, *next, *dummy_node;
 	struct _cds_lfht_node *lookup;
@@ -1233,21 +1325,22 @@ struct cds_lfht_node *cds_lfht_lookup(struct cds_lfht *ht, void *key, size_t key
 		node = clear_flag(next);
 	}
 	assert(!node || !is_dummy(rcu_dereference(node->p.next)));
-	return node;
+	iter->node = node;
+	iter->next = next;
 }
 
-struct cds_lfht_node *cds_lfht_next(struct cds_lfht *ht,
-				struct cds_lfht_node *node)
+void cds_lfht_next(struct cds_lfht *ht, struct cds_lfht_iter *iter)
 {
-	struct cds_lfht_node *next;
+	struct cds_lfht_node *node, *next;
 	unsigned long reverse_hash;
 	void *key;
 	size_t key_len;
 
+	node = iter->node;
 	reverse_hash = node->p.reverse_hash;
 	key = node->key;
 	key_len = node->key_len;
-	next = rcu_dereference(node->p.next);
+	next = iter->next;
 	node = clear_flag(next);
 
 	for (;;) {
@@ -1268,7 +1361,8 @@ struct cds_lfht_node *cds_lfht_next(struct cds_lfht *ht,
 		node = clear_flag(next);
 	}
 	assert(!node || !is_dummy(rcu_dereference(node->p.next)));
-	return node;
+	iter->node = node;
+	iter->next = next;
 }
 
 void cds_lfht_add(struct cds_lfht *ht, struct cds_lfht_node *node)
@@ -1279,12 +1373,12 @@ void cds_lfht_add(struct cds_lfht *ht, struct cds_lfht_node *node)
 	node->p.reverse_hash = bit_reverse_ulong((unsigned long) hash);
 
 	size = rcu_dereference(ht->t.size);
-	(void) _cds_lfht_add(ht, size, node, 0, 0);
+	(void) _cds_lfht_add(ht, size, node, ADD_DEFAULT, 0);
 	ht_count_add(ht, size);
 }
 
 struct cds_lfht_node *cds_lfht_add_unique(struct cds_lfht *ht,
-					struct cds_lfht_node *node)
+				struct cds_lfht_node *node)
 {
 	unsigned long hash, size;
 	struct cds_lfht_node *ret;
@@ -1293,19 +1387,35 @@ struct cds_lfht_node *cds_lfht_add_unique(struct cds_lfht *ht,
 	node->p.reverse_hash = bit_reverse_ulong((unsigned long) hash);
 
 	size = rcu_dereference(ht->t.size);
-	ret = _cds_lfht_add(ht, size, node, 1, 0);
+	ret = _cds_lfht_add(ht, size, node, ADD_UNIQUE, 0);
 	if (ret == node)
 		ht_count_add(ht, size);
 	return ret;
 }
 
+struct cds_lfht_node *cds_lfht_replace(struct cds_lfht *ht,
+				struct cds_lfht_node *node)
+{
+	unsigned long hash, size;
+	struct cds_lfht_node *ret;
+
+	hash = ht->hash_fct(node->key, node->key_len, ht->hash_seed);
+	node->p.reverse_hash = bit_reverse_ulong((unsigned long) hash);
+
+	size = rcu_dereference(ht->t.size);
+	ret = _cds_lfht_add(ht, size, node, ADD_REPLACE, 0);
+	if (ret == NULL)
+		ht_count_add(ht, size);
+	return ret;
+}
+
 int cds_lfht_del(struct cds_lfht *ht, struct cds_lfht_node *node)
 {
 	unsigned long size;
 	int ret;
 
 	size = rcu_dereference(ht->t.size);
-	ret = _cds_lfht_del(ht, size, node, 0);
+	ret = _cds_lfht_del(ht, size, node, 0, 1);
 	if (!ret)
 		ht_count_del(ht, size);
 	return ret;
@@ -1326,6 +1436,7 @@ int cds_lfht_delete_dummy(struct cds_lfht *ht)
 		if (!is_dummy(node))
 			return -EPERM;
 		assert(!is_removed(node));
+		assert(!is_gc(node));
 	} while (!is_end(node));
 	/*
 	 * size accessed without rcu_dereference because hash table is
@@ -1386,7 +1497,7 @@ void cds_lfht_count_nodes(struct cds_lfht *ht,
 	node = (struct cds_lfht_node *) lookup;
 	do {
 		next = rcu_dereference(node->p.next);
-		if (is_removed(next)) {
+		if (is_removed(next) || is_gc(next)) {
 			assert(!is_dummy(next));
 			(*removed)++;
 		} else if (!is_dummy(next))