X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosd-zfs%2Fosd_handler.c;h=8a0329851f3238e79800e59d0e1299ac703c11e0;hb=79d06b6fac3d18f2e5755a940e8afa42e70f3ba2;hp=fedf1e99851e6806832f6d35530783cdfca91694;hpb=35920b759ed78441db0cd9de6ac8ec66da862f22;p=fs%2Flustre-release.git

diff --git a/lustre/osd-zfs/osd_handler.c b/lustre/osd-zfs/osd_handler.c
index fedf1e9..8a03298 100644
--- a/lustre/osd-zfs/osd_handler.c
+++ b/lustre/osd-zfs/osd_handler.c
@@ -28,7 +28,7 @@
  * Use is subject to license terms.
  */
 /*
- * Copyright (c) 2011, 2012 Whamcloud, Inc.
+ * Copyright (c) 2012, Intel Corporation.
  * Use is subject to license terms.
  *
  */
@@ -131,6 +131,7 @@ static void osd_trans_commit_cb(void *cb_data, int error)
 {
 	struct osd_thandle	*oh = cb_data;
 	struct thandle		*th = &oh->ot_super;
+	struct osd_device	*osd = osd_dt_dev(th->th_dev);
 	struct lu_device	*lud = &th->th_dev->dd_lu_dev;
 	struct dt_txn_commit_cb	*dcb, *tmp;
 
@@ -151,6 +152,14 @@ static void osd_trans_commit_cb(void *cb_data, int error)
 	cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage)
 		dcb->dcb_func(NULL, th, dcb, error);
 
+	/* Unlike ldiskfs, zfs updates space accounting at commit time.
+	 * As a consequence, op_end is called only now to inform the quota slave
+	 * component that reserved quota space is now accounted in usage and
+	 * should be released. Quota space won't be adjusted at this point since
+	 * we can't provide a suitable environment. It will be performed
+	 * asynchronously by a lquota thread. */
+	qsd_op_end(NULL, osd->od_quota_slave, &oh->ot_quota_trans);
+
 	lu_device_put(lud);
 	th->th_dev = NULL;
 	lu_context_exit(&th->th_ctx);
@@ -200,11 +209,10 @@ static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 		if (!lu_device_is_md(&d->dd_lu_dev) && rc == -ENOSPC)
 			CERROR("%s: failed to start transaction due to ENOSPC. "
 			       "Metadata overhead is underestimated or "
-			       "grant_ratio is too low.\n",
-			       osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name);
+			       "grant_ratio is too low.\n", osd->od_svname);
 		else
 			CERROR("%s: can't assign tx: rc = %d\n",
-			       osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name, rc);
+			       osd->od_svname, rc);
 	} else {
 		/* add commit callback */
 		dmu_tx_callback_register(oh->ot_tx, osd_trans_commit_cb, oh);
@@ -234,10 +242,23 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
 		LASSERT(oh->ot_tx);
 		dmu_tx_abort(oh->ot_tx);
 		osd_object_sa_dirty_rele(oh);
+		/* there won't be any commit, release reserved quota space now,
+		 * if any */
+		qsd_op_end(env, osd->od_quota_slave, &oh->ot_quota_trans);
 		OBD_FREE_PTR(oh);
 		RETURN(0);
 	}
 
+	/* When doing our own inode accounting, the ZAPs storing per-uid/gid
+	 * usage are updated at operation execution time, so we should call
+	 * qsd_op_end() straight away. Otherwise (for blk accounting maintained
+	 * by ZFS and when #inode is estimated from #blks) accounting is updated
+	 * at commit time and the call to qsd_op_end() must be delayed */
+	if (oh->ot_quota_trans.lqt_id_cnt > 0 &&
+			!oh->ot_quota_trans.lqt_ids[0].lqi_is_blk &&
+			!osd->od_quota_iused_est)
+		qsd_op_end(env, osd->od_quota_slave, &oh->ot_quota_trans);
+
 	rc = dt_txn_hook_stop(env, th);
 	if (rc != 0)
 		CDEBUG(D_OTHER, "%s: transaction hook failed: rc = %d\n",
@@ -278,7 +299,8 @@ static struct thandle *osd_trans_create(const struct lu_env *env,
 	oh->ot_tx = tx;
 	CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
 	CFS_INIT_LIST_HEAD(&oh->ot_sa_list);
-	cfs_sema_init(&oh->ot_sa_lock, 1);
+	sema_init(&oh->ot_sa_lock, 1);
+	memset(&oh->ot_quota_trans, 0, sizeof(oh->ot_quota_trans));
 	th = &oh->ot_super;
 	th->th_dev = dt;
 	th->th_result = 0;
@@ -361,6 +383,7 @@ static int osd_commit_async(const struct lu_env *env, struct dt_device *dev)
 	tx_state_t	  *tx = &dmu_objset_pool(osd->od_objset.os)->dp_tx;
 	uint64_t	   txg;
 
+	mutex_enter(&tx->tx_sync_lock);
 	txg = tx->tx_open_txg + 1;
 	if (tx->tx_quiesce_txg_waiting < txg) {
 		tx->tx_quiesce_txg_waiting = txg;
@@ -482,9 +505,17 @@ static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
 	RETURN(0);
 }
 
+static void osd_xattr_changed_cb(void *arg, uint64_t newval)
+{
+	struct osd_device *osd = arg;
+
+	osd->od_xattr_in_sa = (newval == ZFS_XATTR_SA);
+}
+
 static int osd_mount(const struct lu_env *env,
 		     struct osd_device *o, struct lustre_cfg *cfg)
 {
+	struct dsl_dataset *ds;
 	char	  *dev  = lustre_cfg_string(cfg, 1);
 	dmu_buf_t *rootdb;
 	int	   rc;
@@ -506,6 +537,13 @@ static int osd_mount(const struct lu_env *env,
 		RETURN(rc);
 	}
 
+	ds = dmu_objset_ds(o->od_objset.os);
+	LASSERT(ds);
+	rc = dsl_prop_register(ds, "xattr", osd_xattr_changed_cb, o);
+	if (rc)
+		CERROR("%s: cat not register xattr callback, ignore: %d\n",
+		       o->od_svname, rc);
+
 	rc = __osd_obj2dbuf(env, o->od_objset.os, o->od_objset.root,
 				&rootdb, root_tag);
 	if (rc) {
@@ -522,6 +560,15 @@ static int osd_mount(const struct lu_env *env,
 	if (rc)
 		GOTO(err, rc);
 
+	rc = lu_site_init(&o->od_site, osd2lu_dev(o));
+	if (rc)
+		GOTO(err, rc);
+	o->od_site.ls_bottom_dev = osd2lu_dev(o);
+
+	rc = lu_site_init_finish(&o->od_site);
+	if (rc)
+		GOTO(err, rc);
+
 	/* Use our own ZAP for inode accounting by default, this can be changed
 	 * via procfs to estimate the inode usage from the block usage */
 	o->od_quota_iused_est = 0;
@@ -532,6 +579,14 @@ static int osd_mount(const struct lu_env *env,
 
 	o->arc_prune_cb = arc_add_prune_callback(arc_prune_func, o);
 
+	/* initialize quota slave instance */
+	o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
+				     o->od_proc_entry);
+	if (IS_ERR(o->od_quota_slave)) {
+		rc = PTR_ERR(o->od_quota_slave);
+		o->od_quota_slave = NULL;
+		GOTO(err, rc);
+	}
 err:
 	RETURN(rc);
 }
@@ -619,9 +674,13 @@ static struct lu_device *osd_device_free(const struct lu_env *env,
 
 	cleanup_capa_hash(o->od_capa_hash);
 	/* XXX: make osd top device in order to release reference */
-	/*d->ld_site->ls_top_dev = d;
+	d->ld_site->ls_top_dev = d;
 	lu_site_purge(env, d->ld_site, -1);
-	lu_site_fini(&o->od_site);*/
+	if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
+		LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
+		lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
+	}
+	lu_site_fini(&o->od_site);
 	dt_device_fini(&o->od_dt_dev);
 	OBD_FREE_PTR(o);
 
@@ -632,13 +691,20 @@ static struct lu_device *osd_device_fini(const struct lu_env *env,
 					 struct lu_device *d)
 {
 	struct osd_device *o = osd_dev(d);
+	struct dsl_dataset *ds;
 	int		   rc;
 	ENTRY;
 
 
+	osd_shutdown(env, o);
 	osd_oi_fini(env, o);
 
 	if (o->od_objset.os) {
+		ds = dmu_objset_ds(o->od_objset.os);
+		rc = dsl_prop_unregister(ds, "xattr", osd_xattr_changed_cb, o);
+		if (rc)
+			CERROR("%s: dsl_prop_unregister xattr error %d\n",
+				o->od_svname, rc);
 		arc_remove_prune_callback(o->arc_prune_cb);
 		o->arc_prune_cb = NULL;
 		osd_sync(env, lu2dt_dev(d));
@@ -690,8 +756,17 @@ static int osd_process_config(const struct lu_env *env,
 
 static int osd_recovery_complete(const struct lu_env *env, struct lu_device *d)
 {
+	struct osd_device	*osd = osd_dev(d);
+	int			 rc = 0;
 	ENTRY;
-	RETURN(0);
+
+	if (osd->od_quota_slave == NULL)
+		RETURN(0);
+
+	/* start qsd instance on recovery completion, this notifies the quota
+	 * slave code that we are about to process new requests now */
+	rc = qsd_start(env, osd->od_quota_slave);
+	RETURN(rc);
 }
 
 /*
@@ -714,9 +789,9 @@ static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
 
 	*exp = class_conn2export(&conn);
 
-	cfs_spin_lock(&osd->od_objset.lock);
+	spin_lock(&osd->od_objset.lock);
 	osd->od_connects++;
-	cfs_spin_unlock(&osd->od_objset.lock);
+	spin_unlock(&osd->od_objset.lock);
 
 	RETURN(0);
 }
@@ -733,11 +808,11 @@ static int osd_obd_disconnect(struct obd_export *exp)
 	ENTRY;
 
 	/* Only disconnect the underlying layers on the final disconnect. */
-	cfs_spin_lock(&osd->od_objset.lock);
+	spin_lock(&osd->od_objset.lock);
 	osd->od_connects--;
 	if (osd->od_connects == 0)
 		release = 1;
-	cfs_spin_unlock(&osd->od_objset.lock);
+	spin_unlock(&osd->od_objset.lock);
 
 	rc = class_disconnect(exp); /* bz 9811 */
 
@@ -753,14 +828,19 @@ static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
 	int			 rc = 0;
 	ENTRY;
 
-	/* initialize quota slave instance */
-	osd->od_quota_slave = qsd_init(env, osd->od_svname, &osd->od_dt_dev,
-				       osd->od_proc_entry);
-	if (IS_ERR(osd->od_quota_slave)) {
-		rc = PTR_ERR(osd->od_quota_slave);
-		osd->od_quota_slave = NULL;
+	if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) {
+		/* MDT/MDD still use old infrastructure to create
+		 * special files */
+		rc = llo_local_objects_setup(env, lu2md_dev(pdev),
+					     lu2dt_dev(dev));
+		if (rc)
+			RETURN(rc);
 	}
 
+	if (osd->od_quota_slave != NULL)
+		/* set up quota slave objects */
+		rc = qsd_prepare(env, osd->od_quota_slave);
+
 	RETURN(rc);
 }