ceph: periodically trim stale dentries

Previous commit make VFS delete stale dentry when last reference is
dropped. Lease also can become invalid when corresponding dentry has
no reference. This patch make cephfs periodically scan lease list,
delete corresponding dentry if lease is invalid.

There are two types of lease, dentry lease and dir lease. dentry lease
has life time and applies to singe dentry. Dentry lease is added to tail
of a list when it's updated, leases at front of the list will expire
first. Dir lease is CEPH_CAP_FILE_SHARED on directory inode, it applies
to all dentries in the directory. Dentries have dir leases are added to
another list. Dentries in the list are periodically checked in a round
robin manner.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

1 files changed, 257 insertions, 50 deletions

diff --git a/fs/ceph/dir.c b/fs/ceph/dir.c
index f34a57f56c72..eba283557653 100644
--- a/fs/ceph/dir.c
+++ b/fs/ceph/dir.c
@@ -29,6 +29,9 @@
 
 const struct dentry_operations ceph_dentry_ops;
 
+static bool __dentry_lease_is_valid(struct ceph_dentry_info *di);
+static int __dir_lease_try_check(const struct dentry *dentry);
+
 /*
  * Initialize ceph dentry state.
  */
@@ -44,7 +47,7 @@ static int ceph_d_init(struct dentry *dentry)
 	di->lease_session = NULL;
 	di->time = jiffies;
 	dentry->d_fsdata = di;
-	ceph_dentry_lru_add(dentry);
+	INIT_LIST_HEAD(&di->lease_list);
 	return 0;
 }
 
@@ -241,6 +244,7 @@ static int __dcache_readdir(struct file *file,  struct dir_context *ctx,
 			goto out;
 		}
 		if (fpos_cmp(ctx->pos, di->offset) <= 0) {
+			__ceph_dentry_dir_lease_touch(di);
 			emit_dentry = true;
 		}
 		spin_unlock(&dentry->d_lock);
@@ -1125,13 +1129,259 @@ static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
 }
 
 /*
+ * Move dentry to tail of mdsc->dentry_leases list when lease is updated.
+ * Leases at front of the list will expire first. (Assume all leases have
+ * similar duration)
+ *
+ * Called under dentry->d_lock.
+ */
+void __ceph_dentry_lease_touch(struct ceph_dentry_info *di)
+{
+	struct dentry *dn = di->dentry;
+	struct ceph_mds_client *mdsc;
+
+	dout("dentry_lease_touch %p %p '%pd'\n", di, dn, dn);
+
+	di->flags |= CEPH_DENTRY_LEASE_LIST;
+	if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
+		di->flags |= CEPH_DENTRY_REFERENCED;
+		return;
+	}
+
+	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
+	spin_lock(&mdsc->dentry_list_lock);
+	list_move_tail(&di->lease_list, &mdsc->dentry_leases);
+	spin_unlock(&mdsc->dentry_list_lock);
+}
+
+static void __dentry_dir_lease_touch(struct ceph_mds_client* mdsc,
+				     struct ceph_dentry_info *di)
+{
+	di->flags &= ~(CEPH_DENTRY_LEASE_LIST | CEPH_DENTRY_REFERENCED);
+	di->lease_gen = 0;
+	di->time = jiffies;
+	list_move_tail(&di->lease_list, &mdsc->dentry_dir_leases);
+}
+
+/*
+ * When dir lease is used, add dentry to tail of mdsc->dentry_dir_leases
+ * list if it's not in the list, otherwise set 'referenced' flag.
+ *
+ * Called under dentry->d_lock.
+ */
+void __ceph_dentry_dir_lease_touch(struct ceph_dentry_info *di)
+{
+	struct dentry *dn = di->dentry;
+	struct ceph_mds_client *mdsc;
+
+	dout("dentry_dir_lease_touch %p %p '%pd' (offset %lld)\n",
+	     di, dn, dn, di->offset);
+
+	if (!list_empty(&di->lease_list)) {
+		if (di->flags & CEPH_DENTRY_LEASE_LIST) {
+			/* don't remove dentry from dentry lease list
+			 * if its lease is valid */
+			if (__dentry_lease_is_valid(di))
+				return;
+		} else {
+			di->flags |= CEPH_DENTRY_REFERENCED;
+			return;
+		}
+	}
+
+	if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
+		di->flags |= CEPH_DENTRY_REFERENCED;
+		di->flags &= ~CEPH_DENTRY_LEASE_LIST;
+		return;
+	}
+
+	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
+	spin_lock(&mdsc->dentry_list_lock);
+	__dentry_dir_lease_touch(mdsc, di),
+	spin_unlock(&mdsc->dentry_list_lock);
+}
+
+static void __dentry_lease_unlist(struct ceph_dentry_info *di)
+{
+	struct ceph_mds_client *mdsc;
+	if (di->flags & CEPH_DENTRY_SHRINK_LIST)
+		return;
+	if (list_empty(&di->lease_list))
+		return;
+
+	mdsc = ceph_sb_to_client(di->dentry->d_sb)->mdsc;
+	spin_lock(&mdsc->dentry_list_lock);
+	list_del_init(&di->lease_list);
+	spin_unlock(&mdsc->dentry_list_lock);
+}
+
+enum {
+	KEEP	= 0,
+	DELETE	= 1,
+	TOUCH	= 2,
+	STOP	= 4,
+};
+
+struct ceph_lease_walk_control {
+	bool dir_lease;
+	unsigned long nr_to_scan;
+	unsigned long dir_lease_ttl;
+};
+
+static unsigned long
+__dentry_leases_walk(struct ceph_mds_client *mdsc,
+		     struct ceph_lease_walk_control *lwc,
+		     int (*check)(struct dentry*, void*))
+{
+	struct ceph_dentry_info *di, *tmp;
+	struct dentry *dentry, *last = NULL;
+	struct list_head* list;
+        LIST_HEAD(dispose);
+	unsigned long freed = 0;
+	int ret = 0;
+
+	list = lwc->dir_lease ? &mdsc->dentry_dir_leases : &mdsc->dentry_leases;
+	spin_lock(&mdsc->dentry_list_lock);
+	list_for_each_entry_safe(di, tmp, list, lease_list) {
+		if (!lwc->nr_to_scan)
+			break;
+		--lwc->nr_to_scan;
+
+		dentry = di->dentry;
+		if (last == dentry)
+			break;
+
+		if (!spin_trylock(&dentry->d_lock))
+			continue;
+
+		if (dentry->d_lockref.count < 0) {
+			list_del_init(&di->lease_list);
+			goto next;
+		}
+
+		ret = check(dentry, lwc);
+		if (ret & TOUCH) {
+			/* move it into tail of dir lease list */
+			__dentry_dir_lease_touch(mdsc, di);
+			if (!last)
+				last = dentry;
+		}
+		if (ret & DELETE) {
+			/* stale lease */
+			di->flags &= ~CEPH_DENTRY_REFERENCED;
+			if (dentry->d_lockref.count > 0) {
+				/* update_dentry_lease() will re-add
+				 * it to lease list, or
+				 * ceph_d_delete() will return 1 when
+				 * last reference is dropped */
+				list_del_init(&di->lease_list);
+			} else {
+				di->flags |= CEPH_DENTRY_SHRINK_LIST;
+				list_move_tail(&di->lease_list, &dispose);
+				dget_dlock(dentry);
+			}
+		}
+next:
+		spin_unlock(&dentry->d_lock);
+		if (ret & STOP)
+			break;
+	}
+	spin_unlock(&mdsc->dentry_list_lock);
+
+	while (!list_empty(&dispose)) {
+		di = list_first_entry(&dispose, struct ceph_dentry_info,
+				      lease_list);
+		dentry = di->dentry;
+		spin_lock(&dentry->d_lock);
+
+		list_del_init(&di->lease_list);
+		di->flags &= ~CEPH_DENTRY_SHRINK_LIST;
+		if (di->flags & CEPH_DENTRY_REFERENCED) {
+			spin_lock(&mdsc->dentry_list_lock);
+			if (di->flags & CEPH_DENTRY_LEASE_LIST) {
+				list_add_tail(&di->lease_list,
+					      &mdsc->dentry_leases);
+			} else {
+				__dentry_dir_lease_touch(mdsc, di);
+			}
+			spin_unlock(&mdsc->dentry_list_lock);
+		} else {
+			freed++;
+		}
+
+		spin_unlock(&dentry->d_lock);
+		/* ceph_d_delete() does the trick */
+		dput(dentry);
+	}
+	return freed;
+}
+
+static int __dentry_lease_check(struct dentry *dentry, void *arg)
+{
+	struct ceph_dentry_info *di = ceph_dentry(dentry);
+	int ret;
+
+	if (__dentry_lease_is_valid(di))
+		return STOP;
+	ret = __dir_lease_try_check(dentry);
+	if (ret == -EBUSY)
+		return KEEP;
+	if (ret > 0)
+		return TOUCH;
+	return DELETE;
+}
+
+static int __dir_lease_check(struct dentry *dentry, void *arg)
+{
+	struct ceph_lease_walk_control *lwc = arg;
+	struct ceph_dentry_info *di = ceph_dentry(dentry);
+
+	int ret = __dir_lease_try_check(dentry);
+	if (ret == -EBUSY)
+		return KEEP;
+	if (ret > 0) {
+		if (time_before(jiffies, di->time + lwc->dir_lease_ttl))
+			return STOP;
+		/* Move dentry to tail of dir lease list if we don't want
+		 * to delete it. So dentries in the list are checked in a
+		 * round robin manner */
+		return TOUCH;
+	}
+	return DELETE;
+}
+
+int ceph_trim_dentries(struct ceph_mds_client *mdsc)
+{
+	struct ceph_lease_walk_control lwc;
+	unsigned long freed;
+
+	lwc.dir_lease = false;
+	lwc.nr_to_scan  = CEPH_CAPS_PER_RELEASE * 2;
+	freed = __dentry_leases_walk(mdsc, &lwc, __dentry_lease_check);
+	if (!lwc.nr_to_scan) /* more invalid leases */
+		return -EAGAIN;
+
+	if (lwc.nr_to_scan < CEPH_CAPS_PER_RELEASE)
+		lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE;
+
+	lwc.dir_lease = true;
+	freed +=__dentry_leases_walk(mdsc, &lwc, __dir_lease_check);
+	if (!lwc.nr_to_scan) /* more to check */
+		return -EAGAIN;
+
+	return freed > 0 ? 1 : 0;
+}
+
+/*
  * Ensure a dentry lease will no longer revalidate.
  */
 void ceph_invalidate_dentry_lease(struct dentry *dentry)
 {
+	struct ceph_dentry_info *di = ceph_dentry(dentry);
 	spin_lock(&dentry->d_lock);
-	ceph_dentry(dentry)->time = jiffies;
-	ceph_dentry(dentry)->lease_shared_gen = 0;
+	di->time = jiffies;
+	di->lease_shared_gen = 0;
+	__dentry_lease_unlist(di);
 	spin_unlock(&dentry->d_lock);
 }
 
@@ -1251,6 +1501,8 @@ static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
 	if (atomic_read(&ci->i_shared_gen) == di->lease_shared_gen)
 		valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
 	spin_unlock(&ci->i_ceph_lock);
+	if (valid)
+		__ceph_dentry_dir_lease_touch(di);
 	dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
 	     dir, (unsigned)atomic_read(&ci->i_shared_gen),
 	     dentry, (unsigned)di->lease_shared_gen, valid);
@@ -1343,11 +1595,8 @@ static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
 	}
 
 	dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
-	if (valid) {
-		ceph_dentry_lru_touch(dentry);
-	} else {
+	if (!valid)
 		ceph_dir_clear_complete(dir);
-	}
 
 	if (!(flags & LOOKUP_RCU))
 		dput(parent);
@@ -1387,9 +1636,9 @@ static void ceph_d_release(struct dentry *dentry)
 	struct ceph_dentry_info *di = ceph_dentry(dentry);
 
 	dout("d_release %p\n", dentry);
-	ceph_dentry_lru_del(dentry);
 
 	spin_lock(&dentry->d_lock);
+	__dentry_lease_unlist(di);
 	dentry->d_fsdata = NULL;
 	spin_unlock(&dentry->d_lock);
 
@@ -1490,49 +1739,7 @@ static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
 	return size - left;
 }
 
-/*
- * We maintain a private dentry LRU.
- *
- * FIXME: this needs to be changed to a per-mds lru to be useful.
- */
-void ceph_dentry_lru_add(struct dentry *dn)
-{
-	struct ceph_dentry_info *di = ceph_dentry(dn);
-	struct ceph_mds_client *mdsc;
 
-	dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
-	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
-	spin_lock(&mdsc->dentry_lru_lock);
-	list_add_tail(&di->lru, &mdsc->dentry_lru);
-	mdsc->num_dentry++;
-	spin_unlock(&mdsc->dentry_lru_lock);
-}
-
-void ceph_dentry_lru_touch(struct dentry *dn)
-{
-	struct ceph_dentry_info *di = ceph_dentry(dn);
-	struct ceph_mds_client *mdsc;
-
-	dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
-	     di->offset);
-	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
-	spin_lock(&mdsc->dentry_lru_lock);
-	list_move_tail(&di->lru, &mdsc->dentry_lru);
-	spin_unlock(&mdsc->dentry_lru_lock);
-}
-
-void ceph_dentry_lru_del(struct dentry *dn)
-{
-	struct ceph_dentry_info *di = ceph_dentry(dn);
-	struct ceph_mds_client *mdsc;
-
-	dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
-	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
-	spin_lock(&mdsc->dentry_lru_lock);
-	list_del_init(&di->lru);
-	mdsc->num_dentry--;
-	spin_unlock(&mdsc->dentry_lru_lock);
-}
 
 /*
  * Return name hash for a given dentry.  This is dependent on