4347 ZPL can use dmu_tx_assign(TXG_WAIT)

Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Dan McDonald <danmcd@nexenta.com>
Reviewed by: Boris Protopopov <boris.protopopov@nexenta.com>
Approved by: Dan McDonald <danmcd@nexenta.com>

3 files changed, 20 insertions, 51 deletions

diff --git a/usr/src/uts/common/fs/zfs/zfs_dir.c b/usr/src/uts/common/fs/zfs/zfs_dir.c
index 439420ba41..cb3b4aca15 100644
--- a/usr/src/uts/common/fs/zfs/zfs_dir.c
+++ b/usr/src/uts/common/fs/zfs/zfs_dir.c
@@ -937,7 +937,6 @@ zfs_make_xattrdir(znode_t *zp, vattr_t *vap, vnode_t **xvpp, cred_t *cr)
 		return (SET_ERROR(EDQUOT));
 	}
 
-top:
 	tx = dmu_tx_create(zfsvfs->z_os);
 	dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
 	    ZFS_SA_BASE_ATTR_SIZE);
@@ -946,13 +945,8 @@ top:
 	fuid_dirtied = zfsvfs->z_fuid_dirty;
 	if (fuid_dirtied)
 		zfs_fuid_txhold(zfsvfs, tx);
-	error = dmu_tx_assign(tx, TXG_NOWAIT);
+	error = dmu_tx_assign(tx, TXG_WAIT);
 	if (error) {
-		if (error == ERESTART) {
-			dmu_tx_wait(tx);
-			dmu_tx_abort(tx);
-			goto top;
-		}
 		zfs_acl_ids_free(&acl_ids);
 		dmu_tx_abort(tx);
 		return (error);
diff --git a/usr/src/uts/common/fs/zfs/zfs_vnops.c b/usr/src/uts/common/fs/zfs/zfs_vnops.c
index 78e2a8cce5..6b5c6f852c 100644
--- a/usr/src/uts/common/fs/zfs/zfs_vnops.c
+++ b/usr/src/uts/common/fs/zfs/zfs_vnops.c
@@ -111,11 +111,18 @@
  *  (3)	All range locks must be grabbed before calling dmu_tx_assign(),
  *	as they can span dmu_tx_assign() calls.
  *
- *  (4)	Always pass TXG_NOWAIT as the second argument to dmu_tx_assign().
- *	This is critical because we don't want to block while holding locks.
- *	Note, in particular, that if a lock is sometimes acquired before
- *	the tx assigns, and sometimes after (e.g. z_lock), then failing to
- *	use a non-blocking assign can deadlock the system.  The scenario:
+ *  (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to
+ *      dmu_tx_assign().  This is critical because we don't want to block
+ *      while holding locks.
+ *
+ *	If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT.  This
+ *	reduces lock contention and CPU usage when we must wait (note that if
+ *	throughput is constrained by the storage, nearly every transaction
+ *	must wait).
+ *
+ *      Note, in particular, that if a lock is sometimes acquired before
+ *      the tx assigns, and sometimes after (e.g. z_lock), then failing
+ *      to use a non-blocking assign can deadlock the system.  The scenario:
  *
  *	Thread A has grabbed a lock before calling dmu_tx_assign().
  *	Thread B is in an already-assigned tx, and blocks for this lock.
@@ -728,7 +735,6 @@ zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct)
 	while (n > 0) {
 		abuf = NULL;
 		woff = uio->uio_loffset;
-again:
 		if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) ||
 		    zfs_owner_overquota(zfsvfs, zp, B_TRUE)) {
 			if (abuf != NULL)
@@ -780,13 +786,8 @@ again:
 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
 		dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz));
 		zfs_sa_upgrade_txholds(tx, zp);
-		error = dmu_tx_assign(tx, TXG_NOWAIT);
+		error = dmu_tx_assign(tx, TXG_WAIT);
 		if (error) {
-			if (error == ERESTART) {
-				dmu_tx_wait(tx);
-				dmu_tx_abort(tx);
-				goto again;
-			}
 			dmu_tx_abort(tx);
 			if (abuf != NULL)
 				dmu_return_arcbuf(abuf);
@@ -3043,12 +3044,9 @@ top:
 
 	zfs_sa_upgrade_txholds(tx, zp);
 
-	err = dmu_tx_assign(tx, TXG_NOWAIT);
-	if (err) {
-		if (err == ERESTART)
-			dmu_tx_wait(tx);
+	err = dmu_tx_assign(tx, TXG_WAIT);
+	if (err)
 		goto out;
-	}
 
 	count = 0;
 	/*
@@ -4140,19 +4138,13 @@ zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp,
 		err = SET_ERROR(EDQUOT);
 		goto out;
 	}
-top:
 	tx = dmu_tx_create(zfsvfs->z_os);
 	dmu_tx_hold_write(tx, zp->z_id, off, len);
 
 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
 	zfs_sa_upgrade_txholds(tx, zp);
-	err = dmu_tx_assign(tx, TXG_NOWAIT);
+	err = dmu_tx_assign(tx, TXG_WAIT);
 	if (err != 0) {
-		if (err == ERESTART) {
-			dmu_tx_wait(tx);
-			dmu_tx_abort(tx);
-			goto top;
-		}
 		dmu_tx_abort(tx);
 		goto out;
 	}
diff --git a/usr/src/uts/common/fs/zfs/zfs_znode.c b/usr/src/uts/common/fs/zfs/zfs_znode.c
index f9433b6b44..3a3eee52be 100644
--- a/usr/src/uts/common/fs/zfs/zfs_znode.c
+++ b/usr/src/uts/common/fs/zfs/zfs_znode.c
@@ -1469,7 +1469,6 @@ zfs_extend(znode_t *zp, uint64_t end)
 		zfs_range_unlock(rl);
 		return (0);
 	}
-top:
 	tx = dmu_tx_create(zfsvfs->z_os);
 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
 	zfs_sa_upgrade_txholds(tx, zp);
@@ -1489,13 +1488,8 @@ top:
 		newblksz = 0;
 	}
 
-	error = dmu_tx_assign(tx, TXG_NOWAIT);
+	error = dmu_tx_assign(tx, TXG_WAIT);
 	if (error) {
-		if (error == ERESTART) {
-			dmu_tx_wait(tx);
-			dmu_tx_abort(tx);
-			goto top;
-		}
 		dmu_tx_abort(tx);
 		zfs_range_unlock(rl);
 		return (error);
@@ -1592,17 +1586,11 @@ zfs_trunc(znode_t *zp, uint64_t end)
 		zfs_range_unlock(rl);
 		return (error);
 	}
-top:
 	tx = dmu_tx_create(zfsvfs->z_os);
 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
 	zfs_sa_upgrade_txholds(tx, zp);
-	error = dmu_tx_assign(tx, TXG_NOWAIT);
+	error = dmu_tx_assign(tx, TXG_WAIT);
 	if (error) {
-		if (error == ERESTART) {
-			dmu_tx_wait(tx);
-			dmu_tx_abort(tx);
-			goto top;
-		}
 		dmu_tx_abort(tx);
 		zfs_range_unlock(rl);
 		return (error);
@@ -1709,13 +1697,8 @@ log:
 	tx = dmu_tx_create(zfsvfs->z_os);
 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
 	zfs_sa_upgrade_txholds(tx, zp);
-	error = dmu_tx_assign(tx, TXG_NOWAIT);
+	error = dmu_tx_assign(tx, TXG_WAIT);
 	if (error) {
-		if (error == ERESTART) {
-			dmu_tx_wait(tx);
-			dmu_tx_abort(tx);
-			goto log;
-		}
 		dmu_tx_abort(tx);
 		return (error);
 	}