Remove dummy znode from zvol_state

struct zvol_state contains a dummy znode, which is around 1KB on x64,
only for zfs_range_lock. But in reality, other than z_range_lock and
z_range_avl, zfs_range_lock only need znode on regular file, which
means we add 1KB on a structure and gain nothing.

In this patch, we remove the dummy znode for zvol_state. In order to
do that, we also need to refactor zfs_range_lock a bit. We move
z_range_lock and z_range_avl pair out of znode_t to form zfs_rlock_t.
This new struct replaces znode_t as the main handle inside the range
lock functions.

We also add pointers to z_size, z_blksz, and z_max_blksz so range lock
code doesn't depend on znode_t.  This allows non-ZPL consumers like
Lustre to use the range locks with their equivalent znode_t structure.

Signed-off-by: Chunwei Chen <[email protected]>
Signed-off-by: Boris Protopopov <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>
Closes #4510

3 files changed, 35 insertions, 10 deletions

diff --git a/include/sys/trace_acl.h b/include/sys/trace_acl.h
index 87ffecb4e..230894229 100644
--- a/include/sys/trace_acl.h
+++ b/include/sys/trace_acl.h
@@ -63,7 +63,6 @@ DECLARE_EVENT_CLASS(zfs_ace_class,
 	    __field(uint32_t,		z_sync_cnt)
 	    __field(mode_t,		z_mode)
 	    __field(boolean_t,		z_is_sa)
-	    __field(boolean_t,		z_is_zvol)
 	    __field(boolean_t,		z_is_mapped)
 	    __field(boolean_t,		z_is_ctldir)
 	    __field(boolean_t,		z_is_stale)
@@ -101,7 +100,6 @@ DECLARE_EVENT_CLASS(zfs_ace_class,
 	    __entry->z_sync_cnt		= zn->z_sync_cnt;
 	    __entry->z_mode		= zn->z_mode;
 	    __entry->z_is_sa		= zn->z_is_sa;
-	    __entry->z_is_zvol		= zn->z_is_zvol;
 	    __entry->z_is_mapped	= zn->z_is_mapped;
 	    __entry->z_is_ctldir	= zn->z_is_ctldir;
 	    __entry->z_is_stale		= zn->z_is_stale;
@@ -125,7 +123,7 @@ DECLARE_EVENT_CLASS(zfs_ace_class,
 	    "zn_prefetch %u moved %u blksz %u seq %u "
 	    "mapcnt %llu gen %llu size %llu "
 	    "links %llu pflags %llu uid %llu gid %llu "
-	    "sync_cnt %u mode 0x%x is_sa %d is_zvol %d "
+	    "sync_cnt %u mode 0x%x is_sa %d "
 	    "is_mapped %d is_ctldir %d is_stale %d inode { "
 	    "ino %lu nlink %u version %llu size %lli blkbits %u "
 	    "bytes %u mode 0x%x generation %x } } ace { type %u "
@@ -136,7 +134,7 @@ DECLARE_EVENT_CLASS(zfs_ace_class,
 	    __entry->z_size,
 	    __entry->z_links, __entry->z_pflags, __entry->z_uid,
 	    __entry->z_gid, __entry->z_sync_cnt, __entry->z_mode,
-	    __entry->z_is_sa, __entry->z_is_zvol, __entry->z_is_mapped,
+	    __entry->z_is_sa, __entry->z_is_mapped,
 	    __entry->z_is_ctldir, __entry->z_is_stale, __entry->i_ino,
 	    __entry->i_nlink, __entry->i_version, __entry->i_size,
 	    __entry->i_blkbits, __entry->i_bytes, __entry->i_mode,
diff --git a/include/sys/zfs_rlock.h b/include/sys/zfs_rlock.h
index ea5e40369..5322f3bc7 100644
--- a/include/sys/zfs_rlock.h
+++ b/include/sys/zfs_rlock.h
@@ -32,7 +32,9 @@ extern "C" {
 
 #ifdef _KERNEL
 
-#include <sys/zfs_znode.h>
+#include <sys/list.h>
+#include <sys/avl.h>
+#include <sys/condvar.h>
 
 typedef enum {
 	RL_READER,
@@ -40,8 +42,16 @@ typedef enum {
 	RL_APPEND
 } rl_type_t;
 
+typedef struct zfs_rlock {
+	kmutex_t zr_mutex;	/* protects changes to zr_avl */
+	avl_tree_t zr_avl;	/* avl tree of range locks */
+	uint64_t *zr_size;	/* points to znode->z_size */
+	uint_t *zr_blksz;	/* points to znode->z_blksz */
+	uint64_t *zr_max_blksz; /* points to zsb->z_max_blksz */
+} zfs_rlock_t;
+
 typedef struct rl {
-	znode_t *r_zp;		/* znode this lock applies to */
+	zfs_rlock_t *r_zrl;
 	avl_node_t r_node;	/* avl node link */
 	uint64_t r_off;		/* file range offset */
 	uint64_t r_len;		/* file range length */
@@ -61,7 +71,8 @@ typedef struct rl {
  * is converted to RL_WRITER that specified to lock from the start of the
  * end of file.  Returns the range lock structure.
  */
-rl_t *zfs_range_lock(znode_t *zp, uint64_t off, uint64_t len, rl_type_t type);
+rl_t *zfs_range_lock(zfs_rlock_t *zrl, uint64_t off, uint64_t len,
+    rl_type_t type);
 
 /* Unlock range and destroy range lock structure. */
 void zfs_range_unlock(rl_t *rl);
@@ -78,6 +89,23 @@ void zfs_range_reduce(rl_t *rl, uint64_t off, uint64_t len);
  */
 int zfs_range_compare(const void *arg1, const void *arg2);
 
+static inline void
+zfs_rlock_init(zfs_rlock_t *zrl)
+{
+	mutex_init(&zrl->zr_mutex, NULL, MUTEX_DEFAULT, NULL);
+	avl_create(&zrl->zr_avl, zfs_range_compare,
+	    sizeof (rl_t), offsetof(rl_t, r_node));
+	zrl->zr_size = NULL;
+	zrl->zr_blksz = NULL;
+	zrl->zr_max_blksz = NULL;
+}
+
+static inline void
+zfs_rlock_destroy(zfs_rlock_t *zrl)
+{
+	avl_destroy(&zrl->zr_avl);
+	mutex_destroy(&zrl->zr_mutex);
+}
 #endif /* _KERNEL */
 
 #ifdef	__cplusplus
diff --git a/include/sys/zfs_znode.h b/include/sys/zfs_znode.h
index 65fa10399..30a6ec9f8 100644
--- a/include/sys/zfs_znode.h
+++ b/include/sys/zfs_znode.h
@@ -37,6 +37,7 @@
 #include <sys/rrwlock.h>
 #include <sys/zfs_sa.h>
 #include <sys/zfs_stat.h>
+#include <sys/zfs_rlock.h>
 #endif
 #include <sys/zfs_acl.h>
 #include <sys/zil.h>
@@ -187,8 +188,7 @@ typedef struct znode {
 	krwlock_t	z_parent_lock;	/* parent lock for directories */
 	krwlock_t	z_name_lock;	/* "master" lock for dirent locks */
 	zfs_dirlock_t	*z_dirlocks;	/* directory entry lock list */
-	kmutex_t	z_range_lock;	/* protects changes to z_range_avl */
-	avl_tree_t	z_range_avl;	/* avl tree of file range locks */
+	zfs_rlock_t	z_range_lock;	/* file range lock */
 	uint8_t		z_unlinked;	/* file has been unlinked */
 	uint8_t		z_atime_dirty;	/* atime needs to be synced */
 	uint8_t		z_zn_prefetch;	/* Prefetch znodes? */
@@ -212,7 +212,6 @@ typedef struct znode {
 	list_node_t	z_link_node;	/* all znodes in fs link */
 	sa_handle_t	*z_sa_hdl;	/* handle to sa data */
 	boolean_t	z_is_sa;	/* are we native sa? */
-	boolean_t	z_is_zvol;	/* are we used by the zvol */
 	boolean_t	z_is_mapped;	/* are we mmap'ed */
 	boolean_t	z_is_ctldir;	/* are we .zfs entry */
 	boolean_t	z_is_stale;	/* are we stale due to rollback? */