X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-zfs%2Fosd_object.c;h=63ab904b92fceb4ba0a8d07a8449c858234d07f6;hp=06f1407b823ad11a9b2f2135549ff61797c4aeeb;hb=ec46feab365a10960f36da017b41066915b79c40;hpb=106abc184d8b57de560dc1874683ce5487dcf30a

diff --git a/lustre/osd-zfs/osd_object.c b/lustre/osd-zfs/osd_object.c
index 06f1407..63ab904 100644
--- a/lustre/osd-zfs/osd_object.c
+++ b/lustre/osd-zfs/osd_object.c
@@ -66,11 +66,8 @@ static int osd_object_sync_delay_us = -1;
 
 static struct dt_object_operations osd_obj_ops;
 static struct lu_object_operations osd_lu_obj_ops;
-extern struct dt_body_operations osd_body_ops;
 static struct dt_object_operations osd_obj_otable_it_ops;
 
-extern struct kmem_cache *osd_object_kmem;
-
 static void
 osd_object_sa_fini(struct osd_object *obj)
 {
@@ -291,6 +288,7 @@ int __osd_obj2dnode(objset_t *os, uint64_t oid, dnode_t **dnp)
 	dbi = (dmu_buf_impl_t *)db;
 	DB_DNODE_ENTER(dbi);
 	*dnp = DB_DNODE(dbi);
+	DB_DNODE_EXIT(dbi);
 	LASSERT(*dnp != NULL);
 
 	return 0;
@@ -457,7 +455,7 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 	const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 	ENTRY;
 
-	CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
+	BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
 	lma = (struct lustre_mdt_attrs *)info->oti_buf;
 	buf.lb_buf = lma;
 	buf.lb_len = sizeof(info->oti_buf);
@@ -984,8 +982,7 @@ static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
 	return rc;
 }
 
-static int osd_attr_get(const struct lu_env *env,
-			struct dt_object *dt,
+static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
 			struct lu_attr *attr)
 {
 	struct osd_object	*obj = osd_dt_obj(dt);
@@ -1006,9 +1003,16 @@ static int osd_attr_get(const struct lu_env *env,
 
 	read_lock(&obj->oo_attr_lock);
 	*attr = obj->oo_attr;
-	if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
+	if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
+		attr->la_valid |= LA_FLAGS;
 		attr->la_flags |= LUSTRE_ORPHAN_FL;
+	}
 	read_unlock(&obj->oo_attr_lock);
+	if (attr->la_valid & LA_FLAGS && attr->la_flags & LUSTRE_ORPHAN_FL)
+		CDEBUG(D_INFO, "%s: set orphan flag on "DFID" (%llx/%x)\n",
+		       osd_obj2dev(obj)->od_svname,
+		       PFID(lu_object_fid(&dt->do_lu)),
+		       attr->la_valid, obj->oo_lma_flags);
 
 	/* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
 	 * from within sa_object_size() can block on a mutex, so
@@ -1158,7 +1162,7 @@ static int osd_declare_attr_set(const struct lu_env *env,
 	if (attr && attr->la_valid & LA_UID) {
 		/* quota enforcement for user */
 		if (attr->la_uid != obj->oo_attr.la_uid) {
-			rc = qsd_transfer(env, osd->od_quota_slave,
+			rc = qsd_transfer(env, osd_def_qsd(osd),
 					  &oh->ot_quota_trans, USRQUOTA,
 					  obj->oo_attr.la_uid, attr->la_uid,
 					  bspace, &info->oti_qi, true);
@@ -1169,7 +1173,7 @@ static int osd_declare_attr_set(const struct lu_env *env,
 	if (attr && attr->la_valid & LA_GID) {
 		/* quota enforcement for group */
 		if (attr->la_gid != obj->oo_attr.la_gid) {
-			rc = qsd_transfer(env, osd->od_quota_slave,
+			rc = qsd_transfer(env, osd_def_qsd(osd),
 					  &oh->ot_quota_trans, GRPQUOTA,
 					  obj->oo_attr.la_gid, attr->la_gid,
 					  bspace, &info->oti_qi,
@@ -1181,27 +1185,29 @@ static int osd_declare_attr_set(const struct lu_env *env,
 	}
 #ifdef ZFS_PROJINHERIT
 	if (attr && attr->la_valid & LA_PROJID) {
-		if (!osd->od_projectused_dn)
-			GOTO(out, rc = -EOPNOTSUPP);
-
-		/* Usually, if project quota is upgradable for the device,
-		 * then the upgrade will be done before or when mount the
-		 * device. So when we come here, this project should have
-		 * project ID attribute already (that is zero by default).
-		 * Otherwise, there was something wrong during the former
-		 * upgrade, let's return failure to report that.
-		 *
-		 * Please note that, different from other attributes, you
-		 * can NOT simply set the project ID attribute under such
-		 * case, because adding (NOT change) project ID attribute
-		 * needs to change the object's attribute layout to match
-		 * zfs backend quota accounting requirement. */
-		if (unlikely(!obj->oo_with_projid))
-			GOTO(out, rc = -ENXIO);
-
 		/* quota enforcement for project */
 		if (attr->la_projid != obj->oo_attr.la_projid) {
-			rc = qsd_transfer(env, osd->od_quota_slave,
+			if (!osd->od_projectused_dn)
+				GOTO(out, rc = -EOPNOTSUPP);
+
+			/* Usually, if project quota is upgradable for the
+			 * device, then the upgrade will be done before or when
+			 * mount the device. So when we come here, this project
+			 * should have project ID attribute already (that is
+			 * zero by default).  Otherwise, there was something
+			 * wrong during the former upgrade, let's return failure
+			 * to report that.
+			 *
+			 * Please note that, different from other attributes,
+			 * you can NOT simply set the project ID attribute under
+			 * such case, because adding (NOT change) project ID
+			 * attribute needs to change the object's attribute
+			 * layout to match zfs backend quota accounting
+			 * requirement. */
+			if (unlikely(!obj->oo_with_projid))
+				GOTO(out, rc = -ENXIO);
+
+			rc = qsd_transfer(env, osd_def_qsd(osd),
 					  &oh->ot_quota_trans, PRJQUOTA,
 					  obj->oo_attr.la_projid,
 					  attr->la_projid, bspace,
@@ -1294,7 +1300,7 @@ static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 
 		if (la->la_flags & LUSTRE_LMA_FL_MASKS) {
 			LASSERT(!obj->oo_pfid_in_lma);
-			CLASSERT(sizeof(info->oti_buf) >= sizeof(*lma));
+			BUILD_BUG_ON(sizeof(info->oti_buf) < sizeof(*lma));
 			lma = (struct lustre_mdt_attrs *)&info->oti_buf;
 			buf.lb_buf = lma;
 			buf.lb_len = sizeof(info->oti_buf);
@@ -1332,18 +1338,15 @@ static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 
 	if (valid & LA_PROJID) {
 #ifdef ZFS_PROJINHERIT
-		/* osd_declare_attr_set() must be called firstly.
-		 * If osd::od_projectused_dn is not set, then we
-		 * can not arrive at here. */
-		LASSERT(osd->od_projectused_dn);
-		LASSERT(obj->oo_with_projid);
+		if (osd->od_projectused_dn) {
+			LASSERT(obj->oo_with_projid);
 
-		osa->projid = obj->oo_attr.la_projid = la->la_projid;
-		SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
-				 &osa->projid, 8);
-#else
-		valid &= ~LA_PROJID;
+			osa->projid = obj->oo_attr.la_projid = la->la_projid;
+			SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_PROJID(osd), NULL,
+					 &osa->projid, 8);
+		} else
 #endif
+			valid &= ~LA_PROJID;
 	}
 
 	if (valid & LA_ATIME) {
@@ -1390,7 +1393,7 @@ static int osd_attr_set(const struct lu_env *env, struct dt_object *dt,
 		 * copy */
 		obj->oo_attr.la_flags = attrs_zfs2fs(osa->flags);
 #ifdef ZFS_PROJINHERIT
-		if (obj->oo_with_projid)
+		if (obj->oo_with_projid && osd->od_projectused_dn)
 			osa->flags |= ZFS_PROJID;
 #endif
 		SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(osd), NULL,
@@ -1480,6 +1483,7 @@ static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
 	switch (dof->dof_type) {
 		case DFT_DIR:
 			dt->do_index_ops = &osd_dir_ops;
+			/* fallthrough */
 		case DFT_INDEX:
 			/* for zap create */
 			dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
@@ -1515,13 +1519,13 @@ int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
 		    struct lu_attr *la, uint64_t parent,
 		    nvlist_t *xattr)
 {
-	sa_bulk_attr_t	*bulk = osd_oti_get(env)->oti_attr_bulk;
-	struct osa_attr	*osa = &osd_oti_get(env)->oti_osa;
-	uint64_t	 gen;
-	uint64_t	 crtime[2];
-	timestruc_t	 now;
-	int		 cnt;
-	int		 rc;
+	sa_bulk_attr_t *bulk = osd_oti_get(env)->oti_attr_bulk;
+	struct osa_attr *osa = &osd_oti_get(env)->oti_osa;
+	uint64_t gen;
+	uint64_t crtime[2];
+	inode_timespec_t now;
+	int cnt;
+	int rc;
 	char *dxattr = NULL;
 	size_t sa_size;
 
@@ -1639,11 +1643,12 @@ int osd_find_new_dnode(const struct lu_env *env, dmu_tx_t *tx,
 		LASSERT(db);
 		LASSERT(dn->dn_handle);
 		DB_DNODE_ENTER(db);
-		if (refcount_add(&db->db_holds, osd_obj_tag) == 1) {
-			refcount_add(&dn->dn_holds, tag);
+		if (zfs_refcount_add(&db->db_holds, osd_obj_tag) == 1) {
+			zfs_refcount_add(&dn->dn_holds, osd_obj_tag);
 			atomic_inc_32(&dn->dn_dbufs_count);
 		}
 		*dnp = dn;
+		DB_DNODE_EXIT(db);
 		dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH);
 		break;
 	}
@@ -1811,6 +1816,14 @@ static dnode_t *osd_mkreg(const struct lu_env *env, struct osd_object *obj,
 			       osd->od_svname, rc);
 			return ERR_PTR(rc);
 		}
+	} else if ((fid_is_llog(fid))) {
+		rc = -dmu_object_set_blocksize(osd->od_os, dn->dn_object,
+					       LLOG_MIN_CHUNK_SIZE, 0, oh->ot_tx);
+		if (unlikely(rc)) {
+			CERROR("%s: can't change blocksize: %d\n",
+			       osd->od_svname, rc);
+			return ERR_PTR(rc);
+		}
 	}
 
 	return dn;
@@ -2098,15 +2111,22 @@ static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
 			   __u64 start, __u64 end)
 {
 	struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
+	struct dmu_buf_impl *db = osd_dt_obj(dt)->oo_dn->dn_dbuf;
+	uint64_t txg = 0;
 	ENTRY;
 
-	/* XXX: no other option than syncing the whole filesystem until we
-	 * support ZIL.  If the object tracked the txg that it was last
-	 * modified in, it could pass that txg here instead of "0".  Maybe
-	 * the changes are already committed, so no wait is needed at all? */
-	if (!osd->od_dt_dev.dd_rdonly) {
+	if (osd->od_dt_dev.dd_rdonly)
+		RETURN(0);
+
+	mutex_enter(&db->db_mtx);
+	if (db->db_last_dirty)
+		txg = db->db_last_dirty->dr_txg;
+	mutex_exit(&db->db_mtx);
+
+	if (txg) {
+		/* the object is dirty or being synced */
 		if (osd_object_sync_delay_us < 0)
-			txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
+			txg_wait_synced(dmu_objset_pool(osd->od_os), txg);
 		else
 			udelay(osd_object_sync_delay_us);
 	}