Fix zsb->z_hold_mtx deadlock

The zfs_znode_hold_enter() / zfs_znode_hold_exit() functions are used to
serialize access to a znode and its SA buffer while the object is being
created or destroyed.  This kind of locking would normally reside in the
znode itself but in this case that's impossible because the znode and SA
buffer may not yet exist.  Therefore the locking is handled externally
with an array of mutexs and AVLs trees which contain per-object locks.

In zfs_znode_hold_enter() a per-object lock is created as needed, inserted
in to the correct AVL tree and finally the per-object lock is held.  In
zfs_znode_hold_exit() the process is reversed.  The per-object lock is
released, removed from the AVL tree and destroyed if there are no waiters.

This scheme has two important properties:

1) No memory allocations are performed while holding one of the z_hold_locks.
   This ensures evict(), which can be called from direct memory reclaim, will
   never block waiting on a z_hold_locks which just happens to have hashed
   to the same index.

2) All locks used to serialize access to an object are per-object and never
   shared.  This minimizes lock contention without creating a large number
   of dedicated locks.

On the downside it does require znode_lock_t structures to be frequently
allocated and freed.  However, because these are backed by a kmem cache
and very short lived this cost is minimal.

Signed-off-by: Brian Behlendorf <[email protected]>
Issue #4106

2 files changed, 13 insertions, 17 deletions

diff --git a/include/sys/zfs_vfsops.h b/include/sys/zfs_vfsops.h
index a59114a1a..efaefdacc 100644
--- a/include/sys/zfs_vfsops.h
+++ b/include/sys/zfs_vfsops.h
@@ -112,8 +112,9 @@ typedef struct zfs_sb {
 	uint64_t	z_groupquota_obj;
 	uint64_t	z_replay_eof;	/* New end of file - replay only */
 	sa_attr_type_t	*z_attr_table;	/* SA attr mapping->id */
-	uint64_t	z_hold_mtx_size; /* znode hold locks size */
-	kmutex_t	*z_hold_mtx;	/* znode hold locks */
+	uint64_t	z_hold_size;	/* znode hold array size */
+	avl_tree_t	*z_hold_trees;	/* znode hold trees */
+	kmutex_t	*z_hold_locks;	/* znode hold locks */
 } zfs_sb_t;
 
 #define	ZFS_SUPER_MAGIC	0x2fc12fc1
diff --git a/include/sys/zfs_znode.h b/include/sys/zfs_znode.h
index 59ca085ef..c03bef5c7 100644
--- a/include/sys/zfs_znode.h
+++ b/include/sys/zfs_znode.h
@@ -220,6 +220,12 @@ typedef struct znode {
 	struct inode	z_inode;	/* generic vfs inode */
 } znode_t;
 
+typedef struct znode_hold {
+	uint64_t	zh_obj;		/* object id */
+	kmutex_t	zh_lock;	/* lock serializing object access */
+	avl_node_t	zh_node;	/* avl tree linkage */
+	refcount_t	zh_refcount;	/* active consumer reference count */
+} znode_hold_t;
 
 /*
  * Range locking rules
@@ -273,24 +279,12 @@ typedef struct znode {
 /*
  * Macros for dealing with dmu_buf_hold
  */
-#define	ZFS_OBJ_MTX_SZ	64
-#define	ZFS_OBJ_MTX_MAX	(1024 * 1024)
+#define	ZFS_OBJ_MTX_SZ		64
+#define	ZFS_OBJ_MTX_MAX		(1024 * 1024)
+#define	ZFS_OBJ_HASH(zsb, obj)	((obj) & ((zsb->z_hold_size) - 1))
 
 extern unsigned int zfs_object_mutex_size;
 
-#define	ZFS_OBJ_HASH(zsb, obj_num)	\
-	((obj_num) & ((zsb->z_hold_mtx_size) - 1))
-#define	ZFS_OBJ_MUTEX(zsb, obj_num)	\
-	(&(zsb)->z_hold_mtx[ZFS_OBJ_HASH(zsb, obj_num)])
-#define	ZFS_OBJ_HOLD_ENTER(zsb, obj_num) \
-	mutex_enter(ZFS_OBJ_MUTEX((zsb), (obj_num)))
-#define	ZFS_OBJ_HOLD_TRYENTER(zsb, obj_num) \
-	mutex_tryenter(ZFS_OBJ_MUTEX((zsb), (obj_num)))
-#define	ZFS_OBJ_HOLD_EXIT(zsb, obj_num) \
-	mutex_exit(ZFS_OBJ_MUTEX((zsb), (obj_num)))
-#define	ZFS_OBJ_HOLD_OWNED(zsb, obj_num) \
-	mutex_owned(ZFS_OBJ_MUTEX((zsb), (obj_num)))
-
 /* Encode ZFS stored time values from a struct timespec */
 #define	ZFS_TIME_ENCODE(tp, stmp)		\
 {						\
@@ -326,6 +320,7 @@ extern void	zfs_grow_blocksize(znode_t *, uint64_t, dmu_tx_t *);
 extern int	zfs_freesp(znode_t *, uint64_t, uint64_t, int, boolean_t);
 extern void	zfs_znode_init(void);
 extern void	zfs_znode_fini(void);
+extern int	zfs_znode_hold_compare(const void *, const void *);
 extern int	zfs_zget(zfs_sb_t *, uint64_t, znode_t **);
 extern int	zfs_rezget(znode_t *);
 extern void	zfs_zinactive(znode_t *);