libceph: introduce reference counted string

The data structure is for storing namesapce string. It allows namespace
string to be shared between cephfs inodes with same layout. This data
structure can also be referenced by OSD request.

Signed-off-by: Yan, Zheng <zyan@redhat.com>

3 files changed, 114 insertions, 1 deletions

diff --git a/net/ceph/Makefile b/net/ceph/Makefile
index 958d9856..84cbed6 100644
--- a/net/ceph/Makefile
+++ b/net/ceph/Makefile
@@ -11,5 +11,5 @@ libceph-y := ceph_common.o messenger.o msgpool.o buffer.o pagelist.o \
 	crypto.o armor.o \
 	auth_x.o \
 	ceph_fs.o ceph_strings.o ceph_hash.o \
-	pagevec.o snapshot.o
+	pagevec.o snapshot.o string_table.o
 
diff --git a/net/ceph/ceph_common.c b/net/ceph/ceph_common.c
index 55d2bfe..bddfcf6 100644
--- a/net/ceph/ceph_common.c
+++ b/net/ceph/ceph_common.c
@@ -747,6 +747,8 @@ out:
 static void __exit exit_ceph_lib(void)
 {
 	dout("exit_ceph_lib\n");
+	WARN_ON(!ceph_strings_empty());
+
 	ceph_osdc_cleanup();
 	ceph_msgr_exit();
 	ceph_crypto_shutdown();
diff --git a/net/ceph/string_table.c b/net/ceph/string_table.c
new file mode 100644
index 0000000..ca53c83
--- /dev/null
+++ b/net/ceph/string_table.c
@@ -0,0 +1,111 @@
+#include <linux/slab.h>
+#include <linux/gfp.h>
+#include <linux/string.h>
+#include <linux/spinlock.h>
+#include <linux/ceph/string_table.h>
+
+static DEFINE_SPINLOCK(string_tree_lock);
+static struct rb_root string_tree = RB_ROOT;
+
+struct ceph_string *ceph_find_or_create_string(const char* str, size_t len)
+{
+	struct ceph_string *cs, *exist;
+	struct rb_node **p, *parent;
+	int ret;
+
+	exist = NULL;
+	spin_lock(&string_tree_lock);
+	p = &string_tree.rb_node;
+	while (*p) {
+		exist = rb_entry(*p, struct ceph_string, node);
+		ret = ceph_compare_string(exist, str, len);
+		if (ret > 0)
+			p = &(*p)->rb_left;
+		else if (ret < 0)
+			p = &(*p)->rb_right;
+		else
+			break;
+		exist = NULL;
+	}
+	if (exist && !kref_get_unless_zero(&exist->kref)) {
+		rb_erase(&exist->node, &string_tree);
+		RB_CLEAR_NODE(&exist->node);
+		exist = NULL;
+	}
+	spin_unlock(&string_tree_lock);
+	if (exist)
+		return exist;
+
+	cs = kmalloc(sizeof(*cs) + len + 1, GFP_NOFS);
+	if (!cs)
+		return NULL;
+
+	kref_init(&cs->kref);
+	cs->len = len;
+	memcpy(cs->str, str, len);
+	cs->str[len] = 0;
+
+retry:
+	exist = NULL;
+	parent = NULL;
+	p = &string_tree.rb_node;
+	spin_lock(&string_tree_lock);
+	while (*p) {
+		parent = *p;
+		exist = rb_entry(*p, struct ceph_string, node);
+		ret = ceph_compare_string(exist, str, len);
+		if (ret > 0)
+			p = &(*p)->rb_left;
+		else if (ret < 0)
+			p = &(*p)->rb_right;
+		else
+			break;
+		exist = NULL;
+	}
+	ret = 0;
+	if (!exist) {
+		rb_link_node(&cs->node, parent, p);
+		rb_insert_color(&cs->node, &string_tree);
+	} else if (!kref_get_unless_zero(&exist->kref)) {
+		rb_erase(&exist->node, &string_tree);
+		RB_CLEAR_NODE(&exist->node);
+		ret = -EAGAIN;
+	}
+	spin_unlock(&string_tree_lock);
+	if (ret == -EAGAIN)
+		goto retry;
+
+	if (exist) {
+		kfree(cs);
+		cs = exist;
+	}
+
+	return cs;
+}
+EXPORT_SYMBOL(ceph_find_or_create_string);
+
+static void ceph_free_string(struct rcu_head *head)
+{
+	struct ceph_string *cs = container_of(head, struct ceph_string, rcu);
+	kfree(cs);
+}
+
+void ceph_release_string(struct kref *ref)
+{
+	struct ceph_string *cs = container_of(ref, struct ceph_string, kref);
+
+	spin_lock(&string_tree_lock);
+	if (!RB_EMPTY_NODE(&cs->node)) {
+		rb_erase(&cs->node, &string_tree);
+		RB_CLEAR_NODE(&cs->node);
+	}
+	spin_unlock(&string_tree_lock);
+
+	call_rcu(&cs->rcu, ceph_free_string);
+}
+EXPORT_SYMBOL(ceph_release_string);
+
+bool ceph_strings_empty(void)
+{
+	return RB_EMPTY_ROOT(&string_tree);
+}