X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=584c8c147c49de53bb2014ad9bee82fce730797d;hp=d7eb0e11b31b38236f79f5f1cfe001005a5df35f;hb=6df76d3357fc5896b6902399ed7ce6d7c7835f58;hpb=546993d587c5fc380e9745eae98f863e02e68575

diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c
index d7eb0e1..584c8c1 100644
--- a/lustre/osd-ldiskfs/osd_handler.c
+++ b/lustre/osd-ldiskfs/osd_handler.c
@@ -39,12 +39,11 @@
 
 #define DEBUG_SUBSYSTEM S_OSD
 
+#include <linux/fs_struct.h>
 #include <linux/kallsyms.h>
 #include <linux/module.h>
 #include <linux/user_namespace.h>
-#ifdef HAVE_UIDGID_HEADER
-# include <linux/uidgid.h>
-#endif
+#include <linux/uidgid.h>
 
 /* prerequisite for linux/xattr.h */
 #include <linux/types.h>
@@ -211,14 +210,14 @@ osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 		i = oti->oti_ins_cache_size * 2;
 		if (i == 0)
 			i = OSD_INS_CACHE_SIZE;
-		OBD_ALLOC(idc, sizeof(*idc) * i);
+		OBD_ALLOC_PTR_ARRAY(idc, i);
 		if (idc == NULL)
 			return ERR_PTR(-ENOMEM);
 		if (oti->oti_ins_cache != NULL) {
 			memcpy(idc, oti->oti_ins_cache,
 			       oti->oti_ins_cache_used * sizeof(*idc));
-			OBD_FREE(oti->oti_ins_cache,
-				 oti->oti_ins_cache_used * sizeof(*idc));
+			OBD_FREE_PTR_ARRAY(oti->oti_ins_cache,
+					   oti->oti_ins_cache_used);
 		}
 		oti->oti_ins_cache = idc;
 		oti->oti_ins_cache_size = i;
@@ -286,6 +285,76 @@ osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 	return idc;
 }
 
+static void osd_idc_dump_lma(const struct lu_env *env,
+				struct osd_device *osd,
+				unsigned long ino,
+				bool check_in_oi)
+{
+	struct osd_thread_info *info = osd_oti_get(env);
+	struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
+	const struct lu_fid *fid;
+	struct osd_inode_id lid;
+	struct inode *inode;
+	int rc;
+
+	inode = osd_ldiskfs_iget(osd_sb(osd), ino);
+	if (IS_ERR(inode)) {
+		CERROR("%s: can't get inode %lu: rc = %d\n",
+		       osd->od_svname, ino, (int)PTR_ERR(inode));
+		return;
+	}
+	if (is_bad_inode(inode)) {
+		CERROR("%s: bad inode %lu\n", osd->od_svname, ino);
+		goto put;
+	}
+	rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
+	if (rc) {
+		CERROR("%s: can't get LMA for %lu: rc = %d\n",
+		       osd->od_svname, ino, rc);
+		goto put;
+	}
+	fid = &loa->loa_lma.lma_self_fid;
+	LCONSOLE(D_INFO, "%s: "DFID" in inode %lu/%u\n", osd->od_svname,
+		      PFID(fid), ino, (unsigned)inode->i_generation);
+	if (!check_in_oi)
+		goto put;
+	rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
+	if (rc) {
+		CERROR("%s: can't lookup "DFID": rc = %d\n",
+		       osd->od_svname, PFID(fid), rc);
+		goto put;
+	}
+	LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n", osd->od_svname,
+		      PFID(fid), lid.oii_ino, lid.oii_gen);
+put:
+	iput(inode);
+}
+
+static void osd_idc_dump_debug(const struct lu_env *env,
+				struct osd_device *osd,
+				const struct lu_fid *fid,
+				unsigned long ino1,
+				unsigned long ino2)
+{
+	struct osd_inode_id lid;
+
+	int rc;
+
+	rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
+	if (!rc) {
+		LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n",
+			osd->od_svname, PFID(fid), lid.oii_ino, lid.oii_gen);
+		osd_idc_dump_lma(env, osd, lid.oii_ino, false);
+	} else {
+		CERROR("%s: can't lookup "DFID": rc = %d\n",
+		       osd->od_svname, PFID(fid), rc);
+	}
+	if (ino1)
+		osd_idc_dump_lma(env, osd, ino1, true);
+	if (ino2)
+		osd_idc_dump_lma(env, osd, ino2, true);
+}
+
 /*
  * lookup mapping for given FID and fill it from the given object.
  * the object is lolcal by definition.
@@ -303,7 +372,12 @@ static int osd_idc_find_and_init(const struct lu_env *env,
 		if (obj->oo_inode == NULL)
 			return 0;
 		if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
-			LASSERT(idc->oic_lid.oii_ino == 0);
+			if (idc->oic_lid.oii_ino) {
+				osd_idc_dump_debug(env, osd, fid,
+						   idc->oic_lid.oii_ino,
+						   obj->oo_inode->i_ino);
+				return -EINVAL;
+			}
 			idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 			idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 		}
@@ -393,12 +467,11 @@ int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 		lustre_loa_swab(loa, true);
 		/* Check LMA compatibility */
 		if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
-			CWARN("%s: unsupported incompat LMA feature(s) %#x "
-			      "for fid = "DFID", ino = %lu\n",
+			rc = -EOPNOTSUPP;
+			CWARN("%s: unsupported incompat LMA feature(s) %#x for fid = "DFID", ino = %lu: rc = %d\n",
 			      osd_ino2name(inode),
 			      lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
-			      PFID(&lma->lma_self_fid), inode->i_ino);
-			rc = -EOPNOTSUPP;
+			      PFID(&lma->lma_self_fid), inode->i_ino, rc);
 		}
 	} else if (rc == 0) {
 		rc = -ENODATA;
@@ -443,10 +516,11 @@ struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 		iput(inode);
 		inode = ERR_PTR(-ESTALE);
 	} else if (is_bad_inode(inode)) {
-		CWARN("%s: bad inode: ino = %u\n",
-		      osd_dev2name(dev), id->oii_ino);
+		rc = -ENOENT;
+		CWARN("%s: bad inode: ino = %u: rc = %d\n",
+		      osd_dev2name(dev), id->oii_ino, rc);
 		iput(inode);
-		inode = ERR_PTR(-ENOENT);
+		inode = ERR_PTR(rc);
 	} else if ((rc = osd_attach_jinode(inode))) {
 		iput(inode);
 		inode = ERR_PTR(rc);
@@ -479,6 +553,7 @@ int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 		struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 		struct inode *parent = child->d_parent->d_inode;
 		struct lu_fid *fid = NULL;
+		char fidstr[FID_LEN + 1] = "unknown";
 
 		rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 		if (!rc2) {
@@ -495,19 +570,18 @@ int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 		}
 
 		if (fid != NULL)
-			/* below message is checked in sanity.sh test_129 */
-			CWARN("%s: directory (inode: %lu, FID: "DFID") %s maximum entry limit\n",
-			      osd_name(osd), parent->i_ino, PFID(fid),
-			      rc == -ENOSPC ? "has reached" : "is approaching");
-		else
-			/* below message is checked in sanity.sh test_129 */
-			CWARN("%s: directory (inode: %lu, FID: unknown) %s maximum entry limit\n",
-			      osd_name(osd), parent->i_ino,
-			      rc == -ENOSPC ? "has reached" : "is approaching");
+			snprintf(fidstr, sizeof(fidstr), DFID, PFID(fid));
+
+		/* below message is checked in sanity.sh test_129 */
+		if (rc == -ENOSPC) {
+			CWARN("%s: directory (inode: %lu, FID: %s) has reached max size limit\n",
+			      osd_name(osd), parent->i_ino, fidstr);
+		} else {
+			rc = 0;	/* ignore such error now */
+			CWARN("%s: directory (inode: %lu, FID: %s) is approaching max size limit\n",
+			      osd_name(osd), parent->i_ino, fidstr);
+		}
 
-		/* ignore such error now */
-		if (rc == -ENOBUFS)
-			rc = 0;
 	}
 
 	return rc;
@@ -855,10 +929,8 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 }
 
 struct osd_check_lmv_buf {
-#ifdef HAVE_DIR_CONTEXT
 	/* please keep it as first member */
 	struct dir_context ctx;
-#endif
 	struct osd_thread_info *oclb_info;
 	struct osd_device *oclb_dev;
 	struct osd_idmap_cache *oclb_oic;
@@ -867,7 +939,7 @@ struct osd_check_lmv_buf {
 };
 
 /**
- * It is called internally by ->readdir() to filter out the
+ * It is called internally by ->iterate*() to filter out the
  * local slave object's FID of the striped directory.
  *
  * \retval	1 found the local slave's FID
@@ -960,13 +1032,10 @@ static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 {
 	struct lu_buf *buf = &oti->oti_big_buf;
 	struct dentry *dentry = &oti->oti_obj_dentry;
-	struct file *filp = &oti->oti_file;
-	const struct file_operations *fops;
+	struct file *filp;
 	struct lmv_mds_md_v1 *lmv1;
 	struct osd_check_lmv_buf oclb = {
-#ifdef HAVE_DIR_CONTEXT
 		.ctx.actor = osd_stripe_dir_filldir,
-#endif
 		.oclb_info = oti,
 		.oclb_dev = dev,
 		.oclb_oic = oic,
@@ -1008,35 +1077,23 @@ again:
 	if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
 		GOTO(out, rc = 0);
 
-	fops = inode->i_fop;
-	dentry->d_inode = inode;
-	dentry->d_sb = inode->i_sb;
-	filp->f_pos = 0;
-	filp->f_path.dentry = dentry;
-	filp->f_mode = FMODE_64BITHASH;
-	filp->f_mapping = inode->i_mapping;
-	filp->f_op = fops;
-	filp->private_data = NULL;
-	set_file_inode(filp, inode);
+	filp = osd_quasi_file(oti->oti_env, inode);
+	rc = osd_security_file_alloc(filp);
+	if (rc)
+		goto out;
 
 	do {
 		oclb.oclb_items = 0;
-#ifdef HAVE_DIR_CONTEXT
-		oclb.ctx.pos = filp->f_pos;
-		rc = fops->iterate_shared(filp, &oclb.ctx);
-		filp->f_pos = oclb.ctx.pos;
-#else
-		rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
-#endif
+		rc = iterate_dir(filp, &oclb.ctx);
 	} while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
 		 filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
-	fops->release(inode, filp);
+	inode->i_fop->release(inode, filp);
 
 out:
 	if (rc < 0)
-		CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
-		       DFID": rc = %d\n", osd_ino2name(inode),
-		       inode->i_ino, inode->i_generation,
+		CDEBUG(D_LFSCK,
+		       "%s: cannot check LMV, ino = %lu/%u "DFID": rc = %d\n",
+		       osd_ino2name(inode), inode->i_ino, inode->i_generation,
 		       PFID(&oic->oic_fid), rc);
 	else
 		rc = 0;
@@ -1070,7 +1127,13 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 
 	LINVRNT(osd_invariant(obj));
 	LASSERT(obj->oo_inode == NULL);
-	LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
+
+	if (fid_is_sane(fid) == 0) {
+		CERROR("%s: invalid FID "DFID"\n", ldev->ld_obd->obd_name,
+		       PFID(fid));
+		dump_stack();
+		RETURN(-EINVAL);
+	}
 
 	dev = osd_dev(ldev);
 	scrub = &dev->od_scrub.os_scrub;
@@ -1185,17 +1248,19 @@ trigger:
 		}
 	}
 
-	if (thread_is_running(&scrub->os_thread)) {
+	if (scrub->os_running) {
 		if (scrub->os_partial_scan && !scrub->os_in_join)
 			goto join;
 
-		if (IS_ERR_OR_NULL(inode) || result)
+		osd_add_oi_cache(info, dev, id, fid);
+		if (IS_ERR_OR_NULL(inode) || result) {
+			osd_oii_insert(dev, oic, result == -ENOENT);
 			GOTO(out, result = -EINPROGRESS);
+		}
 
 		LASSERT(remote);
 		LASSERT(obj->oo_inode == inode);
 
-		osd_add_oi_cache(info, dev, id, fid);
 		osd_oii_insert(dev, oic, true);
 		goto found;
 	}
@@ -1218,13 +1283,15 @@ join:
 	if (rc1 && rc1 != -EALREADY)
 		GOTO(out, result = -EREMCHG);
 
-	if (IS_ERR_OR_NULL(inode) || result)
+	osd_add_oi_cache(info, dev, id, fid);
+	if (IS_ERR_OR_NULL(inode) || result) {
+		osd_oii_insert(dev, oic, result == -ENOENT);
 		GOTO(out, result = -EINPROGRESS);
+	}
 
 	LASSERT(remote);
 	LASSERT(obj->oo_inode == inode);
 
-	osd_add_oi_cache(info, dev, id, fid);
 	osd_oii_insert(dev, oic, true);
 	goto found;
 
@@ -1407,6 +1474,13 @@ static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 
 	LINVRNT(osd_invariant(obj));
 
+	if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
+	    cfs_fail_val == 2) {
+		struct osd_thread_info *info = osd_oti_get(env);
+		struct osd_idmap_cache *oic = &info->oti_cache;
+		/* invalidate thread cache */
+		memset(&oic->oic_fid, 0, sizeof(oic->oic_fid));
+	}
 	if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 		obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 		l->lo_header->loh_attr |= LOHA_EXISTS;
@@ -1436,6 +1510,7 @@ static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 			result = 0;
 		}
 	}
+	obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
 
 	LINVRNT(osd_invariant(obj));
 	return result;
@@ -1462,8 +1537,6 @@ static int osd_oxc_get(struct osd_object *obj, const char *name,
 	size_t namelen = strlen(name);
 	int rc;
 
-	ENTRY;
-
 	rcu_read_lock();
 	list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
 		if (namelen == tmp->oxe_namelen &&
@@ -1490,7 +1563,6 @@ static int osd_oxc_get(struct osd_object *obj, const char *name,
 		GOTO(out, rc = -ERANGE);
 
 	memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
-	EXIT;
 out:
 	rcu_read_unlock();
 
@@ -1590,11 +1662,10 @@ static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 	dt_object_fini(&obj->oo_dt);
 	if (obj->oo_hl_head != NULL)
 		ldiskfs_htree_lock_head_free(obj->oo_hl_head);
+	/* obj doesn't contain an lu_object_header, so we don't need call_rcu */
 	OBD_FREE_PTR(obj);
-	if (unlikely(h)) {
-		lu_object_header_fini(h);
-		OBD_FREE_PTR(h);
-	}
+	if (unlikely(h))
+		lu_object_header_free(h);
 }
 
 /*
@@ -1616,16 +1687,6 @@ static void osd_index_fini(struct osd_object *o)
 	}
 }
 
-/*
- * Concurrency: no concurrent access is possible that late in object
- * life-cycle (for all existing callers, that is. New callers have to provide
- * their own locking.)
- */
-static int osd_inode_unlinked(const struct inode *inode)
-{
-	return inode->i_nlink == 0;
-}
-
 enum {
 	OSD_TXN_OI_DELETE_CREDITS    = 20,
 	OSD_TXN_INODE_DELETE_CREDITS = 20
@@ -1712,9 +1773,10 @@ static int osd_param_is_not_sane(const struct osd_device *dev,
 static void osd_trans_commit_cb(struct super_block *sb,
 				struct ldiskfs_journal_cb_entry *jcb, int error)
 {
-	struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
+	struct osd_thandle *oh = container_of(jcb, struct osd_thandle, ot_jcb);
 	struct thandle *th = &oh->ot_super;
 	struct lu_device *lud = &th->th_dev->dd_lu_dev;
+	struct osd_device *osd = osd_dev(lud);
 	struct dt_txn_commit_cb *dcb, *tmp;
 
 	LASSERT(oh->ot_handle == NULL);
@@ -1734,17 +1796,13 @@ static void osd_trans_commit_cb(struct super_block *sb,
 	}
 
 	lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
-	lu_device_put(lud);
+	if (atomic_dec_and_test(&osd->od_commit_cb_in_flight))
+		wake_up(&osd->od_commit_cb_done);
 	th->th_dev = NULL;
 
 	OBD_FREE_PTR(oh);
 }
 
-#ifndef HAVE_SB_START_WRITE
-# define sb_start_write(sb) do {} while (0)
-# define sb_end_write(sb) do {} while (0)
-#endif
-
 static struct thandle *osd_trans_create(const struct lu_env *env,
 					struct dt_device *d)
 {
@@ -1780,6 +1838,7 @@ static struct thandle *osd_trans_create(const struct lu_env *env,
 	th->th_dev = d;
 	th->th_result = 0;
 	oh->ot_credits = 0;
+	oh->oh_declared_ext = 0;
 	INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
 	INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
 	INIT_LIST_HEAD(&oh->ot_trunc_locks);
@@ -1802,7 +1861,7 @@ void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
 	struct osd_thread_info *oti = osd_oti_get(env);
 	struct osd_thandle *oh;
 
-	oh = container_of0(th, struct osd_thandle, ot_super);
+	oh = container_of(th, struct osd_thandle, ot_super);
 	LASSERT(oh != NULL);
 
 	CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
@@ -1861,7 +1920,7 @@ static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 
 	LASSERT(current->journal_info == NULL);
 
-	oh = container_of0(th, struct osd_thandle, ot_super);
+	oh = container_of(th, struct osd_thandle, ot_super);
 	LASSERT(oh != NULL);
 	LASSERT(oh->ot_handle == NULL);
 
@@ -1922,7 +1981,7 @@ static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 		oh->ot_handle = jh;
 		LASSERT(oti->oti_txns == 0);
 
-		lu_device_get(&d->dd_lu_dev);
+		atomic_inc(&dev->od_commit_cb_in_flight);
 		lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
 			      "osd-tx", th);
 		oti->oti_txns++;
@@ -1990,7 +2049,7 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
 
 	ENTRY;
 
-	oh = container_of0(th, struct osd_thandle, ot_super);
+	oh = container_of(th, struct osd_thandle, ot_super);
 
 	remove_agents = oh->ot_remove_agents;
 
@@ -2039,7 +2098,7 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
 		OBD_FREE_PTR(oh);
 	}
 
-	osd_trunc_unlock_all(&truncates);
+	osd_trunc_unlock_all(env, &truncates);
 
 	/* inform the quota slave device that the transaction is stopping */
 	qsd_op_end(env, qsd, qtrans);
@@ -2063,6 +2122,7 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
 	if (unlikely(remove_agents != 0))
 		osd_process_scheduled_agent_removals(env, osd);
 
+	LASSERT(oti->oti_ins_cache_depth > 0);
 	oti->oti_ins_cache_depth--;
 	/* reset OI cache for safety */
 	if (oti->oti_ins_cache_depth == 0)
@@ -2075,8 +2135,8 @@ static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
 
 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
 {
-	struct osd_thandle *oh = container_of0(th, struct osd_thandle,
-					       ot_super);
+	struct osd_thandle *oh = container_of(th, struct osd_thandle,
+					      ot_super);
 
 	LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
 	LASSERT(&dcb->dcb_func != NULL);
@@ -2115,6 +2175,9 @@ static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 	if (!inode)
 		return;
 
+	if (osd_has_index(obj) &&  obj->oo_dt.do_index_ops == &osd_index_iam_ops)
+		ldiskfs_set_inode_flag(inode, LDISKFS_INODE_JOURNAL_DATA);
+
 	uid = i_uid_read(inode);
 	gid = i_gid_read(inode);
 	projid = i_projid_read(inode);
@@ -2212,9 +2275,9 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d,
 
 	statfs_pack(sfs, ksfs);
 	if (unlikely(sb->s_flags & SB_RDONLY))
-		sfs->os_state |= OS_STATE_READONLY;
+		sfs->os_state |= OS_STATFS_READONLY;
 
-	sfs->os_state |= osd->od_nonrotational ? OS_STATE_NONROT : 0;
+	sfs->os_state |= osd->od_nonrotational ? OS_STATFS_NONROT : 0;
 
 	if (ldiskfs_has_feature_extents(sb))
 		sfs->os_maxbytes = sb->s_maxbytes;
@@ -2229,7 +2292,7 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d,
 	 *
 	 * Reserve 0.78% of total space, at least 8MB for small filesystems.
 	 */
-	CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
+	BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
 	reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
 	if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
 		reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
@@ -2376,6 +2439,8 @@ static void osd_conf_get(const struct lu_env *env,
 			       d->od_svname, name);
 		}
 	}
+
+	param->ddp_has_lseek_data_hole = true;
 }
 
 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
@@ -2430,10 +2495,14 @@ static int osd_commit_async(const struct lu_env *env,
 	RETURN(rc);
 }
 
-/* Our own copy of the set readonly functions if present, or NU if not. */
-static int (*priv_dev_set_rdonly)(struct block_device *bdev);
-static int (*priv_dev_check_rdonly)(struct block_device *bdev);
-/* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
+static int (*priv_security_file_alloc)(struct file *file);
+
+int osd_security_file_alloc(struct file *file)
+{
+	if (priv_security_file_alloc)
+		return priv_security_file_alloc(file);
+	return 0;
+}
 
 /*
  * Concurrency: shouldn't matter.
@@ -2446,35 +2515,8 @@ static int osd_ro(const struct lu_env *env, struct dt_device *d)
 
 	ENTRY;
 
-	if (priv_dev_set_rdonly) {
-		struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
-
-		rc = 0;
-		CERROR("*** setting %s read-only ***\n",
-		       osd_dt_dev(d)->od_svname);
-
-		if (sb->s_op->freeze_fs) {
-			rc = sb->s_op->freeze_fs(sb);
-			if (rc)
-				goto out;
-		}
-
-		if (jdev && (jdev != dev)) {
-			CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
-			       (long)jdev);
-			priv_dev_set_rdonly(jdev);
-		}
-		CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
-		priv_dev_set_rdonly(dev);
-
-		if (sb->s_op->unfreeze_fs)
-			sb->s_op->unfreeze_fs(sb);
-	}
-
-out:
-	if (rc)
-		CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
-		       osd_dt_dev(d)->od_svname, (long)dev, rc);
+	CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
+	       osd_dt_dev(d)->od_svname, (long)dev, rc);
 
 	RETURN(rc);
 }
@@ -2623,11 +2665,12 @@ static void osd_inode_getattr(const struct lu_env *env,
 	attr->la_valid	|= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
 			   LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
 			   LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
-			   LA_BLKSIZE | LA_TYPE;
+			   LA_BLKSIZE | LA_TYPE | LA_BTIME;
 
 	attr->la_atime = inode->i_atime.tv_sec;
 	attr->la_mtime = inode->i_mtime.tv_sec;
 	attr->la_ctime = inode->i_ctime.tv_sec;
+	attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
 	attr->la_mode	 = inode->i_mode;
 	attr->la_size	 = i_size_read(inode);
 	attr->la_blocks	 = inode->i_blocks;
@@ -2648,10 +2691,71 @@ static void osd_inode_getattr(const struct lu_env *env,
 		attr->la_flags |= LUSTRE_PROJINHERIT_FL;
 }
 
+static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt,
+			    u64 *count)
+{
+	struct osd_object *obj = osd_dt_obj(dt);
+	const struct dt_it_ops *iops;
+	struct dt_it *it;
+	int rc;
+
+	ENTRY;
+
+	LASSERT(S_ISDIR(obj->oo_inode->i_mode));
+	LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)));
+
+	if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
+		*count = obj->oo_dirent_count;
+		RETURN(0);
+	}
+
+	/* directory not initialized yet */
+	if (!dt->do_index_ops) {
+		*count = 0;
+		RETURN(0);
+	}
+
+	iops = &dt->do_index_ops->dio_it;
+	it = iops->init(env, dt, LUDA_64BITHASH);
+	if (IS_ERR(it))
+		RETURN(PTR_ERR(it));
+
+	rc = iops->load(env, it, 0);
+	if (rc < 0) {
+		if (rc == -ENODATA) {
+			rc = 0;
+			*count = 0;
+		}
+		GOTO(out, rc);
+	}
+	if (rc > 0)
+		rc = iops->next(env, it);
+
+	for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) {
+		if (rc == -ESTALE)
+			continue;
+
+		if (iops->key_size(env, it) == 0)
+			continue;
+
+		(*count)++;
+	}
+	if (rc == 1) {
+		obj->oo_dirent_count = *count;
+		rc = 0;
+	}
+out:
+	iops->put(env, it);
+	iops->fini(env, it);
+
+	RETURN(rc);
+}
+
 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);
+	int rc = 0;
 
 	if (unlikely(!dt_object_exists(dt)))
 		return -ENOENT;
@@ -2667,9 +2771,17 @@ static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
 		attr->la_valid |= LA_FLAGS;
 		attr->la_flags |= LUSTRE_ORPHAN_FL;
 	}
+	if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
+		attr->la_valid |= LA_FLAGS;
+		attr->la_flags |= LUSTRE_ENCRYPT_FL;
+	}
 	spin_unlock(&obj->oo_guard);
 
-	return 0;
+	if (S_ISDIR(obj->oo_inode->i_mode) &&
+	    fid_is_namespace_visible(lu_object_fid(&dt->do_lu)))
+		rc = osd_dirent_count(env, dt, &attr->la_dirent_count);
+
+	return rc;
 }
 
 static int osd_declare_attr_qid(const struct lu_env *env,
@@ -2752,7 +2864,7 @@ static int osd_declare_attr_set(const struct lu_env *env,
 	obj = osd_dt_obj(dt);
 	LASSERT(osd_invariant(obj));
 
-	oh = container_of0(handle, struct osd_thandle, ot_super);
+	oh = container_of(handle, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle == NULL);
 
 	osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
@@ -2779,9 +2891,9 @@ static int osd_declare_attr_set(const struct lu_env *env,
 		bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
 
 		if (!ignore_edquot)
-			CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
-			       "(the quota space is %lld)\n",
-			       obj->oo_inode->i_sb->s_id, attr->la_uid,
+			CDEBUG(D_QUOTA,
+			       "%s: enforce quota on UID %u, GID %u (quota space is %lld)\n",
+			       osd_ino2name(obj->oo_inode), attr->la_uid,
 			       attr->la_gid, bspace);
 
 		/* USERQUOTA */
@@ -2794,9 +2906,10 @@ static int osd_declare_attr_set(const struct lu_env *env,
 			RETURN(rc);
 
 		gid = i_gid_read(obj->oo_inode);
+		CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
+		       attr->la_uid, gid, attr->la_gid);
 		enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
-		rc = osd_declare_attr_qid(env, obj, oh, bspace,
-					  i_gid_read(obj->oo_inode),
+		rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
 					  attr->la_gid, enforce, GRPQUOTA,
 					  ignore_edquot);
 		if (rc)
@@ -2866,6 +2979,13 @@ static int osd_inode_setattr(const struct lu_env *env,
 		/* always keep S_NOCMTIME */
 		inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
 				 S_NOCMTIME;
+#if defined(S_ENCRYPTED)
+		/* Always remove S_ENCRYPTED, because ldiskfs must not be
+		 * aware of encryption status. It is just stored into LMA
+		 * so that it can be forwared to client side.
+		 */
+		inode->i_flags &= ~S_ENCRYPTED;
+#endif
 		/*
 		 * Ext4 did not transfer inherit flags from
 		 * @inode->i_flags to raw inode i_flags when writing
@@ -2880,7 +3000,8 @@ static int osd_inode_setattr(const struct lu_env *env,
 }
 
 #ifdef HAVE_PROJECT_QUOTA
-static int osd_transfer_project(struct inode *inode, __u32 projid)
+static int osd_transfer_project(struct inode *inode, __u32 projid,
+				struct thandle *handle)
 {
 	struct super_block *sb = inode->i_sb;
 	struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
@@ -2912,9 +3033,18 @@ static int osd_transfer_project(struct inode *inode, __u32 projid)
 
 	raw_inode = ldiskfs_raw_inode(&iloc);
 	if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
-		err = -EOVERFLOW;
-		brelse(iloc.bh);
-		return err;
+		struct osd_thandle *oh = container_of(handle,
+						      struct osd_thandle,
+						      ot_super);
+		/**
+		 * try to expand inode size automatically.
+		 */
+		ldiskfs_mark_inode_dirty(oh->ot_handle, inode);
+		if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
+			err = -EOVERFLOW;
+			brelse(iloc.bh);
+			return err;
+		}
 	}
 	brelse(iloc.bh);
 
@@ -2931,7 +3061,8 @@ static int osd_transfer_project(struct inode *inode, __u32 projid)
 }
 #endif
 
-static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
+static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr,
+			      struct thandle *handle)
 {
 	int rc;
 
@@ -2939,6 +3070,11 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
 	    (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
 		struct iattr iattr;
 
+		CDEBUG(D_QUOTA,
+		       "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
+		       inode->i_ino, i_uid_read(inode), attr->la_uid,
+		       i_gid_read(inode), attr->la_gid);
+
 		dquot_initialize(inode);
 		iattr.ia_valid = 0;
 		if (attr->la_valid & LA_UID)
@@ -2950,9 +3086,8 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
 
 		rc = dquot_transfer(inode, &iattr);
 		if (rc) {
-			CERROR("%s: quota transfer failed: rc = %d. Is quota "
-			       "enforcement enabled on the ldiskfs "
-			       "filesystem?\n", inode->i_sb->s_id, rc);
+			CERROR("%s: quota transfer failed. Is quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
+			       osd_ino2name(inode), rc);
 			return rc;
 		}
 	}
@@ -2961,14 +3096,13 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
 	if (attr->la_valid & LA_PROJID &&
 	    attr->la_projid != i_projid_read(inode)) {
 #ifdef HAVE_PROJECT_QUOTA
-		rc = osd_transfer_project(inode, attr->la_projid);
+		rc = osd_transfer_project(inode, attr->la_projid, handle);
 #else
 		rc = -ENOTSUPP;
 #endif
 		if (rc) {
-			CERROR("%s: quota transfer failed: rc = %d. Is project "
-			       "enforcement enabled on the ldiskfs "
-			       "filesystem?\n", inode->i_sb->s_id, rc);
+			CERROR("%s: quota transfer failed. Is project enforcement enabled on the ldiskfs filesystem? rc = %d\n",
+			       osd_ino2name(inode), rc);
 			return rc;
 		}
 	}
@@ -3012,7 +3146,7 @@ static int osd_attr_set(const struct lu_env *env,
 		if (unlikely(ipd == NULL))
 			RETURN(-ENOMEM);
 
-		oh = container_of0(handle, struct osd_thandle, ot_super);
+		oh = container_of(handle, struct osd_thandle, ot_super);
 		rc = iam_update(oh->ot_handle, bag,
 				(const struct iam_key *)fid1,
 				(const struct iam_rec *)id, ipd);
@@ -3022,7 +3156,7 @@ static int osd_attr_set(const struct lu_env *env,
 
 	inode = obj->oo_inode;
 
-	rc = osd_quota_transfer(inode, attr);
+	rc = osd_quota_transfer(inode, attr, handle);
 	if (rc)
 		return rc;
 
@@ -3032,7 +3166,9 @@ static int osd_attr_set(const struct lu_env *env,
 	if (rc != 0)
 		GOTO(out, rc);
 
-	ll_dirty_inode(inode, I_DIRTY_DATASYNC);
+	osd_dirty_inode(inode, I_DIRTY_DATASYNC);
+
+	osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
 
 	if (!(attr->la_valid & LA_FLAGS))
 		GOTO(out, rc);
@@ -3052,6 +3188,9 @@ static int osd_attr_set(const struct lu_env *env,
 		lma->lma_incompat |=
 			lustre_to_lma_flags(attr->la_flags);
 		lustre_lma_swab(lma);
+
+		osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
+
 		rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
 				     lma, sizeof(*lma), XATTR_REPLACE);
 		if (rc != 0) {
@@ -3067,7 +3206,6 @@ static int osd_attr_set(const struct lu_env *env,
 		osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
 	}
 out:
-	osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
 
 	return rc;
 }
@@ -3158,6 +3296,8 @@ static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
 
 	oth = container_of(th, struct osd_thandle, ot_super);
 	LASSERT(oth->ot_handle->h_transaction != NULL);
+	if (fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)))
+		obj->oo_dirent_count = 0;
 	result = osd_mkfile(info, obj, mode, hint, th, attr);
 
 	return result;
@@ -3304,7 +3444,8 @@ static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
 }
 
 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
-			  struct lu_attr *attr, struct dt_object_format *dof)
+			  struct lu_attr *attr, struct dt_object_format *dof,
+			  struct thandle *handle)
 {
 	struct inode *inode = obj->oo_inode;
 	__u64 valid = attr->la_valid;
@@ -3321,7 +3462,7 @@ static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
 	if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
 		attr->la_valid &= ~LA_MTIME;
 
-	result = osd_quota_transfer(inode, attr);
+	result = osd_quota_transfer(inode, attr, handle);
 	if (result)
 		return;
 
@@ -3335,7 +3476,7 @@ static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
 		 * enabled on ldiskfs (lquota takes care of it).
 		 */
 		LASSERTF(result == 0, "%d\n", result);
-		ll_dirty_inode(inode, I_DIRTY_DATASYNC);
+		osd_dirty_inode(inode, I_DIRTY_DATASYNC);
 	}
 
 	attr->la_valid = valid;
@@ -3372,7 +3513,7 @@ static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
 	}
 
 	if (likely(result == 0)) {
-		osd_attr_init(info, obj, attr, dof);
+		osd_attr_init(info, obj, attr, dof, th);
 		osd_object_init0(obj);
 	}
 
@@ -3400,7 +3541,10 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
 
 	LASSERT(obj->oo_inode != NULL);
 
-	oh = container_of0(th, struct osd_thandle, ot_super);
+	if (CFS_FAIL_CHECK(OBD_FAIL_OSD_OI_ENOSPC))
+		return -ENOSPC;
+
+	oh = container_of(th, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle);
 	osd_trans_exec_op(env, th, OSD_OT_INSERT);
 
@@ -3466,7 +3610,7 @@ static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
 
 	LASSERT(handle != NULL);
 
-	oh = container_of0(handle, struct osd_thandle, ot_super);
+	oh = container_of(handle, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle == NULL);
 
 	/*
@@ -3524,7 +3668,7 @@ static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
 	if (inode == NULL)
 		RETURN(-ENOENT);
 
-	oh = container_of0(th, struct osd_thandle, ot_super);
+	oh = container_of(th, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle == NULL);
 
 	osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
@@ -3575,7 +3719,7 @@ static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
 
 	ENTRY;
 
-	oh = container_of0(th, struct osd_thandle, ot_super);
+	oh = container_of(th, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle);
 	LASSERT(inode);
 	LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
@@ -3591,13 +3735,15 @@ static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
 	}
 
 	if (S_ISDIR(inode->i_mode)) {
-		LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
-			inode->i_nlink == 2);
+		if (inode->i_nlink > 2)
+			CERROR("%s: directory "DFID" ino %lu link count is %u at unlink. run e2fsck to repair\n",
+			       osd_name(osd), PFID(fid), inode->i_ino,
+			       inode->i_nlink);
 
 		spin_lock(&obj->oo_guard);
 		clear_nlink(inode);
 		spin_unlock(&obj->oo_guard);
-		ll_dirty_inode(inode, I_DIRTY_DATASYNC);
+		osd_dirty_inode(inode, I_DIRTY_DATASYNC);
 	}
 
 	osd_trans_exec_op(env, th, OSD_OT_DESTROY);
@@ -3756,6 +3902,9 @@ static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
 	__u32 saved_nlink = dir->i_nlink;
 	int rc;
 
+	if (OBD_FAIL_CHECK(OBD_FAIL_OSD_DOTDOT_ENOSPC))
+		return -ENOSPC;
+
 	dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
 	osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
 
@@ -3796,6 +3945,7 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
 	struct osd_thread_info *info = osd_oti_get(env);
 	struct inode *local;
 	struct osd_thandle *oh;
+	uid_t own[2] = {0, 0};
 	int rc;
 
 	ENTRY;
@@ -3804,8 +3954,7 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
 	oh = container_of(th, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle->h_transaction != NULL);
 
-	local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
-				     NULL);
+	local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type, own);
 	if (IS_ERR(local)) {
 		CERROR("%s: create local error %d\n", osd_name(osd),
 		       (int)PTR_ERR(local));
@@ -3824,7 +3973,7 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
 	 * debugging if we need to determine where this symlink came from.
 	 */
 	if (S_ISLNK(type)) {
-		CLASSERT(LDISKFS_N_BLOCKS * 4 >= FID_LEN + 1);
+		BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
 		rc = snprintf((char *)LDISKFS_I(local)->i_data,
 			      LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
 
@@ -3837,11 +3986,10 @@ static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
 #ifdef	HAVE_PROJECT_QUOTA
 	if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
 	    i_projid_read(pobj->oo_inode) != 0) {
-		rc = osd_transfer_project(local, 0);
+		rc = osd_transfer_project(local, 0, th);
 		if (rc) {
-			CERROR("%s: quota transfer failed: rc = %d. Is project "
-			       "quota enforcement enabled on the ldiskfs "
-			       "filesystem?\n", local->i_sb->s_id, rc);
+			CERROR("%s: quota transfer failed:. Is project quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
+			       osd_ino2name(local), rc);
 			RETURN(ERR_PTR(rc));
 		}
 	}
@@ -3908,20 +4056,16 @@ static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 	struct osd_thread_info *info = osd_oti_get(env);
 	struct osd_obj_orphan *oor, *tmp;
 	struct osd_inode_id id;
-	struct list_head list;
+	LIST_HEAD(list);
 	struct inode *inode;
 	struct lu_fid fid;
 	handle_t *jh;
 	__u32 ino;
 
-	INIT_LIST_HEAD(&list);
-
 	spin_lock(&osd->od_osfs_lock);
 	list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
-		if (oor->oor_env == env) {
-			list_del(&oor->oor_list);
-			list_add(&oor->oor_list, &list);
-		}
+		if (oor->oor_env == env)
+			list_move(&oor->oor_list, &list);
 	}
 	spin_unlock(&osd->od_osfs_lock);
 
@@ -4009,8 +4153,21 @@ static int osd_create(const struct lu_env *env, struct dt_object *dt,
 			obj->oo_dt.do_body_ops = &osd_body_ops;
 	}
 
-	if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
+	if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY)) {
+		struct inode *inode = obj->oo_inode;
+
 		result = __osd_oi_insert(env, obj, fid, th);
+		if (result && inode) {
+			spin_lock(&obj->oo_guard);
+			clear_nlink(inode);
+			spin_unlock(&obj->oo_guard);
+			osd_dirty_inode(inode, I_DIRTY_DATASYNC);
+			ldiskfs_set_inode_state(inode,
+						LDISKFS_STATE_LUSTRE_DESTROY);
+			iput(inode);
+			obj->oo_inode = NULL;
+		}
+	}
 
 	/*
 	 * a small optimization - dt_insert() isn't usually applied
@@ -4034,19 +4191,21 @@ static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
 			       struct thandle *handle)
 {
 	struct osd_thandle *oh;
+	int rc;
 
 	/* it's possible that object doesn't exist yet */
 	LASSERT(handle != NULL);
 
-	oh = container_of0(handle, struct osd_thandle, ot_super);
+	oh = container_of(handle, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle == NULL);
 
 	osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
 			     osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
 
-	osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
+	rc = osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev),
+				   osd_dt_obj(dt));
 
-	return 0;
+	return rc;
 }
 
 /*
@@ -4068,7 +4227,7 @@ static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
 	LASSERT(osd_is_write_locked(env, obj));
 	LASSERT(th != NULL);
 
-	oh = container_of0(th, struct osd_thandle, ot_super);
+	oh = container_of(th, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle != NULL);
 
 	osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
@@ -4097,7 +4256,7 @@ static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
 	}
 	spin_unlock(&obj->oo_guard);
 
-	ll_dirty_inode(inode, I_DIRTY_DATASYNC);
+	osd_dirty_inode(inode, I_DIRTY_DATASYNC);
 	LINVRNT(osd_invariant(obj));
 
 	osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
@@ -4116,7 +4275,7 @@ static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
 	LASSERT(!dt_object_remote(dt));
 	LASSERT(handle != NULL);
 
-	oh = container_of0(handle, struct osd_thandle, ot_super);
+	oh = container_of(handle, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle == NULL);
 
 	osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
@@ -4144,7 +4303,10 @@ static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
 	LASSERT(osd_is_write_locked(env, obj));
 	LASSERT(th != NULL);
 
-	oh = container_of0(th, struct osd_thandle, ot_super);
+	if (OBD_FAIL_CHECK(OBD_FAIL_OSD_REF_DEL))
+		return -EIO;
+
+	oh = container_of(th, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle != NULL);
 
 	osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
@@ -4170,7 +4332,7 @@ static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
 	ldiskfs_dec_count(oh->ot_handle, inode);
 	spin_unlock(&obj->oo_guard);
 
-	ll_dirty_inode(inode, I_DIRTY_DATASYNC);
+	osd_dirty_inode(inode, I_DIRTY_DATASYNC);
 	LINVRNT(osd_invariant(obj));
 
 	osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
@@ -4295,7 +4457,7 @@ static int osd_declare_xattr_set(const struct lu_env *env,
 
 	LASSERT(handle != NULL);
 
-	oh = container_of0(handle, struct osd_thandle, ot_super);
+	oh = container_of(handle, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle == NULL);
 
 	if (strcmp(name, XATTR_NAME_LMA) == 0) {
@@ -4434,7 +4596,7 @@ static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
 
 		/* Remove old PFID EA entry firstly. */
 		dquot_initialize(inode);
-		rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
+		rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
 		if (rc == -ENODATA) {
 			if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
 				RETURN(rc);
@@ -4514,7 +4676,7 @@ static int osd_xattr_handle_linkea(const struct lu_env *env,
 
 	ENTRY;
 
-	oh = container_of0(handle, struct osd_thandle, ot_super);
+	oh = container_of(handle, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle != NULL);
 
 	rc = linkea_init_with_rec(&ldata);
@@ -4587,7 +4749,7 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
 		 * Version is set after all inode operations are finished,
 		 * so we should mark it dirty here
 		 */
-		ll_dirty_inode(inode, I_DIRTY_DATASYNC);
+		osd_dirty_inode(inode, I_DIRTY_DATASYNC);
 
 		RETURN(0);
 	}
@@ -4683,7 +4845,7 @@ static int osd_declare_xattr_del(const struct lu_env *env,
 	LASSERT(!dt_object_remote(dt));
 	LASSERT(handle != NULL);
 
-	oh = container_of0(handle, struct osd_thandle, ot_super);
+	oh = container_of(handle, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle == NULL);
 
 	osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
@@ -4740,7 +4902,7 @@ static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
 		dquot_initialize(inode);
 		dentry->d_inode = inode;
 		dentry->d_sb = inode->i_sb;
-		rc = osd_removexattr(dentry, inode, name);
+		rc = ll_vfs_removexattr(dentry, inode, name);
 	}
 
 	osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
@@ -4758,20 +4920,11 @@ static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
 {
 	struct osd_object *obj = osd_dt_obj(dt);
 	struct inode *inode = obj->oo_inode;
-	struct osd_thread_info *info = osd_oti_get(env);
-	struct dentry *dentry = &info->oti_obj_dentry;
-	struct file *file = &info->oti_file;
+	struct file *file = osd_quasi_file(env, inode);
 	int rc;
 
 	ENTRY;
 
-	dentry->d_inode = inode;
-	dentry->d_sb = inode->i_sb;
-	file->f_path.dentry = dentry;
-	file->f_mapping = inode->i_mapping;
-	file->f_op = inode->i_fop;
-	set_file_inode(file, inode);
-
 	rc = vfs_fsync_range(file, start, end, 0);
 
 	RETURN(rc);
@@ -4984,7 +5137,7 @@ static int osd_index_declare_iam_delete(const struct lu_env *env,
 {
 	struct osd_thandle *oh;
 
-	oh = container_of0(handle, struct osd_thandle, ot_super);
+	oh = container_of(handle, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle == NULL);
 
 	/* Recycle  may cause additional three blocks to be changed. */
@@ -5032,7 +5185,7 @@ static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
 	if (unlikely(ipd == NULL))
 		RETURN(-ENOMEM);
 
-	oh = container_of0(handle, struct osd_thandle, ot_super);
+	oh = container_of(handle, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle != NULL);
 	LASSERT(oh->ot_handle->h_transaction != NULL);
 
@@ -5063,7 +5216,7 @@ static int osd_index_declare_ea_delete(const struct lu_env *env,
 	LASSERT(!dt_object_remote(dt));
 	LASSERT(handle != NULL);
 
-	oh = container_of0(handle, struct osd_thandle, ot_super);
+	oh = container_of(handle, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle == NULL);
 
 	credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
@@ -5225,11 +5378,21 @@ static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
 	} else {
 		rc = PTR_ERR(bh);
 	}
+
+	if (!rc && fid_is_namespace_visible(lu_object_fid(&dt->do_lu)) &&
+	    obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
+		/* NB, dirent count may not be accurate, because it's counted
+		 * without lock.
+		 */
+		if (obj->oo_dirent_count)
+			obj->oo_dirent_count--;
+		else
+			obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
+	}
 	if (hlock != NULL)
 		ldiskfs_htree_unlock(hlock);
 	else
 		up_write(&obj->oo_ext_idx_sem);
-
 	GOTO(out, rc);
 out:
 	LASSERT(osd_invariant(obj));
@@ -5316,7 +5479,7 @@ static int osd_index_declare_iam_insert(const struct lu_env *env,
 
 	LASSERT(handle != NULL);
 
-	oh = container_of0(handle, struct osd_thandle, ot_super);
+	oh = container_of(handle, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle == NULL);
 
 	osd_trans_declare_op(env, oh, OSD_OT_INSERT,
@@ -5366,7 +5529,7 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
 	if (unlikely(ipd == NULL))
 		RETURN(-ENOMEM);
 
-	oh = container_of0(th, struct osd_thandle, ot_super);
+	oh = container_of(th, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle != NULL);
 	LASSERT(oh->ot_handle->h_transaction != NULL);
 	if (S_ISDIR(obj->oo_inode->i_mode)) {
@@ -5564,6 +5727,10 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
 					      hlock, th);
 		}
 	}
+	if (!rc && fid_is_namespace_visible(lu_object_fid(&pobj->oo_dt.do_lu))
+	    && pobj->oo_dirent_count != LU_DIRENT_COUNT_UNSET)
+		pobj->oo_dirent_count++;
+
 	if (hlock != NULL)
 		ldiskfs_htree_unlock(hlock);
 	else
@@ -5589,8 +5756,7 @@ osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
 	if (!fid_is_norm(fid) && !fid_is_igif(fid))
 		RETURN(0);
 
-	if (thread_is_running(&scrub->os_thread) &&
-	    scrub->os_pos_current > id->oii_ino)
+	if (scrub->os_running && scrub->os_pos_current > id->oii_ino)
 		RETURN(0);
 
 	if (dev->od_auto_scrub_interval == AS_NEVER ||
@@ -5631,7 +5797,7 @@ again:
 	insert = false;
 
 trigger:
-	if (thread_is_running(&scrub->os_thread)) {
+	if (scrub->os_running) {
 		if (inode == NULL) {
 			inode = osd_iget(oti, dev, id);
 			/* The inode has been removed (by race maybe). */
@@ -5664,8 +5830,7 @@ trigger:
 		rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
 				     SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
 		CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
-		       "%s: trigger partial OI scrub for RPC inconsistency "
-		       "checking FID "DFID": rc = %d\n",
+		       "%s: trigger partial OI scrub for RPC inconsistency checking FID "DFID": rc = %d\n",
 		       osd_dev2name(dev), PFID(fid), rc);
 		if (rc == 0 || rc == -EALREADY)
 			goto again;
@@ -5957,7 +6122,7 @@ static int osd_index_declare_ea_insert(const struct lu_env *env,
 	LASSERT(fid != NULL);
 	LASSERT(rec1->rec_type != 0);
 
-	oh = container_of0(handle, struct osd_thandle, ot_super);
+	oh = container_of(handle, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle == NULL);
 
 	credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
@@ -6117,6 +6282,7 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
 		iput(child_inode);
 	LASSERT(osd_invariant(obj));
 	osd_trans_exec_check(env, th, OSD_OT_INSERT);
+
 	RETURN(rc);
 }
 
@@ -6453,7 +6619,7 @@ static struct dt_it *osd_it_ea_init(const struct lu_env *env,
 	file->f_path.dentry	= obj_dentry;
 	file->f_mapping		= obj->oo_inode->i_mapping;
 	file->f_op		= obj->oo_inode->i_fop;
-	set_file_inode(file, obj->oo_inode);
+	file->f_inode = obj->oo_inode;
 
 	lu_object_get(lo);
 	RETURN((struct dt_it *)oie);
@@ -6515,13 +6681,11 @@ static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
 }
 
 struct osd_filldir_cbs {
-#ifdef HAVE_DIR_CONTEXT
 	struct dir_context ctx;
-#endif
 	struct osd_it_ea  *it;
 };
 /**
- * It is called internally by ->readdir(). It fills the
+ * It is called internally by ->iterate*(). It fills the
  * iterator's in-memory data structure with required
  * information i.e. name, namelen, rec_size etc.
  *
@@ -6588,7 +6752,7 @@ static int osd_ldiskfs_filldir(void *buf,
 }
 
 /**
- * Calls ->readdir() to load a directory entry at a time
+ * Calls ->iterate*() to load a directory entry at a time
  * and stored it in iterator's in-memory data structure.
  *
  * \param di iterator's in memory structure
@@ -6607,9 +6771,7 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env,
 	struct file *filp = &it->oie_file;
 	int rc = 0;
 	struct osd_filldir_cbs buf = {
-#ifdef HAVE_DIR_CONTEXT
 		.ctx.actor = osd_ldiskfs_filldir,
-#endif
 		.it = it
 	};
 
@@ -6625,18 +6787,16 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env,
 		down_read(&obj->oo_ext_idx_sem);
 	}
 
-#ifdef HAVE_DIR_CONTEXT
-	buf.ctx.pos = filp->f_pos;
-	rc = inode->i_fop->iterate_shared(filp, &buf.ctx);
-	filp->f_pos = buf.ctx.pos;
-#else
-	rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
-#endif
+	filp->f_cred = current_cred();
+	rc = osd_security_file_alloc(filp);
+	if (rc)
+		GOTO(unlock, rc);
 
-	if (hlock != NULL)
-		ldiskfs_htree_unlock(hlock);
-	else
-		up_read(&obj->oo_ext_idx_sem);
+	filp->f_flags |= O_NOATIME;
+	filp->f_mode |= FMODE_NONOTIFY;
+	rc = iterate_dir(filp, &buf.ctx);
+	if (rc)
+		GOTO(unlock, rc);
 
 	if (it->oie_rd_dirent == 0) {
 		/*
@@ -6650,12 +6810,17 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env,
 		it->oie_dirent = it->oie_buf;
 		it->oie_it_dirent = 1;
 	}
+unlock:
+	if (hlock != NULL)
+		ldiskfs_htree_unlock(hlock);
+	else
+		up_read(&obj->oo_ext_idx_sem);
 
 	RETURN(rc);
 }
 
 /**
- * It calls osd_ldiskfs_it_fill() which will use ->readdir()
+ * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
  * to load a directory entry at a time and stored it in
  * iterator's in-memory data structure.
  *
@@ -6718,6 +6883,11 @@ static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
 	return it->oie_dirent->oied_namelen;
 }
 
+#if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
+#undef LDISKFS_DIR_REC_LEN
+#define LDISKFS_DIR_REC_LEN(de)		LDISKFS_DIR_ENTRY_LEN_((de))
+#endif
+
 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
 {
 	if (LDISKFS_DIR_REC_LEN(de) >=
@@ -6793,11 +6963,10 @@ osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
 	 * That means we lose it!
 	 */
 	if (rc != 0)
-		CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
-		       "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
-		       osd_ino2name(inode),
-		       dir->i_ino, dir->i_generation, namelen,
-		       dentry->d_name.name, PFID(fid), rc);
+		CDEBUG(D_LFSCK,
+		       "%s: fail to reinsert the dirent, dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
+		       osd_ino2name(inode), dir->i_ino, dir->i_generation,
+		       namelen, dentry->d_name.name, PFID(fid), rc);
 
 	RETURN(rc);
 }
@@ -7208,8 +7377,6 @@ static inline int osd_it_ea_rec(const struct lu_env *env,
 
 				rc = osd_ea_fid_get(env, obj, ino, fid, id);
 			}
-		} else {
-			osd_id_gen(id, ino, OSD_OII_NOGEN);
 		}
 	}
 
@@ -7219,15 +7386,6 @@ static inline int osd_it_ea_rec(const struct lu_env *env,
 			   it->oie_dirent->oied_namelen,
 			   it->oie_dirent->oied_type, attr);
 
-	if (rc < 0)
-		RETURN(rc);
-
-	if (osd_remote_fid(env, dev, fid))
-		RETURN(0);
-
-	if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
-		osd_add_oi_cache(oti, dev, id, fid);
-
 	RETURN(rc > 0 ? 0 : rc);
 }
 
@@ -7268,7 +7426,7 @@ static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
 }
 
 /**
- * It calls osd_ldiskfs_it_fill() which will use ->readdir()
+ * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
  * to load a directory entry at a time and stored it i inn,
  * in iterator's in-memory data structure.
  *
@@ -7384,11 +7542,17 @@ static void osd_key_fini(const struct lu_context *ctx,
 	if (info->oti_dio_pages) {
 		int i;
 		for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
-			if (info->oti_dio_pages[i])
-				__free_page(info->oti_dio_pages[i]);
+			struct page *page = info->oti_dio_pages[i];
+			if (page) {
+				LASSERT(PagePrivate2(page));
+				LASSERT(PageLocked(page));
+				ClearPagePrivate2(page);
+				unlock_page(page);
+				__free_page(page);
+			}
 		}
-		OBD_FREE(info->oti_dio_pages,
-			 sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
+		OBD_FREE_PTR_ARRAY_LARGE(info->oti_dio_pages,
+					 PTLRPC_MAX_BRW_PAGES);
 	}
 
 	if (info->oti_inode != NULL)
@@ -7402,7 +7566,7 @@ static void osd_key_fini(const struct lu_context *ctx,
 	lu_buf_free(&info->oti_big_buf);
 	if (idc != NULL) {
 		LASSERT(info->oti_ins_cache_size > 0);
-		OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
+		OBD_FREE_PTR_ARRAY(idc, info->oti_ins_cache_size);
 		info->oti_ins_cache = NULL;
 		info->oti_ins_cache_size = 0;
 	}
@@ -7444,7 +7608,7 @@ static int osd_device_init(const struct lu_env *env, struct lu_device *d,
 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
 {
 	struct seq_server_site *ss = osd_seq_site(osd);
-	int rc;
+	int rc = 0;
 
 	ENTRY;
 
@@ -7458,13 +7622,8 @@ static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
 	if (osd->od_cl_seq == NULL)
 		RETURN(-ENOMEM);
 
-	rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
-			     osd->od_svname, ss->ss_server_seq);
-	if (rc != 0) {
-		OBD_FREE_PTR(osd->od_cl_seq);
-		osd->od_cl_seq = NULL;
-		RETURN(rc);
-	}
+	seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
+			osd->od_svname, ss->ss_server_seq);
 
 	if (ss->ss_node_id == 0) {
 		/*
@@ -7521,6 +7680,8 @@ static void osd_umount(const struct lu_env *env, struct osd_device *o)
 	if (o->od_mnt != NULL) {
 		shrink_dcache_sb(osd_sb(o));
 		osd_sync(env, &o->od_dt_dev);
+		wait_event(o->od_commit_cb_done,
+			  !atomic_read(&o->od_commit_cb_in_flight));
 
 		mntput(o->od_mnt);
 		o->od_mnt = NULL;
@@ -7543,7 +7704,7 @@ static int osd_mount(const struct lu_env *env,
 	struct osd_thread_info *info = osd_oti_get(env);
 	struct lu_fid *fid = &info->oti_fid;
 	struct inode *inode;
-	int rc = 0, force_over_512tb = 0;
+	int rc = 0, force_over_1024tb = 0;
 
 	ENTRY;
 
@@ -7570,23 +7731,22 @@ static int osd_mount(const struct lu_env *env,
 #endif
 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
 	if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
-		CWARN("force_over_128tb option is deprecated. "
-		      "Filesystems less than 512TB can be created without any "
-		      "force options. Use force_over_512tb option for "
-		      "filesystems greater than 512TB.\n");
+		CWARN("force_over_128tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force option. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
 	}
 #endif
 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
 	if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
-		CWARN("force_over_256tb option is deprecated. "
-		      "Filesystems less than 512TB can be created without any "
-		      "force options. Use force_over_512tb option for "
-		      "filesystems greater than 512TB.\n");
+		CWARN("force_over_256tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
+	}
+#endif
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
+	if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
+		CWARN("force_over_512tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
 	}
 #endif
 
-	if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
-		force_over_512tb = 1;
+	if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
+		force_over_1024tb = 1;
 
 	__page = alloc_page(GFP_KERNEL);
 	if (__page == NULL)
@@ -7596,7 +7756,7 @@ static int osd_mount(const struct lu_env *env,
 	*options = '\0';
 	if (opts != NULL) {
 		/* strip out the options for back compatiblity */
-		static char *sout[] = {
+		static const char * const sout[] = {
 			"mballoc",
 			"iopen",
 			"noiopen",
@@ -7608,6 +7768,7 @@ static int osd_mount(const struct lu_env *env,
 			"force_over_128tb",
 			"force_over_256tb",
 			"force_over_512tb",
+			"force_over_1024tb",
 			NULL
 		};
 		strncat(options, opts, PAGE_SIZE);
@@ -7655,33 +7816,18 @@ static int osd_mount(const struct lu_env *env,
 	}
 
 	if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
-				 osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
-				 force_over_512tb == 0) {
-		CERROR("%s: device %s LDISKFS does not support filesystems "
-		       "greater than 512TB and can cause data corruption. "
-		       "Use \"force_over_512tb\" mount option to override.\n",
+				 osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
+				 force_over_1024tb == 0) {
+		CERROR("%s: device %s LDISKFS has not been tested on filesystems larger than 1024TB and may cause data corruption. Use 'force_over_1024tb' mount option to override.\n",
 		       name, dev);
 		GOTO(out_mnt, rc = -EINVAL);
 	}
 
 	if (lmd_flags & LMD_FLG_DEV_RDONLY) {
-		if (priv_dev_set_rdonly) {
-			priv_dev_set_rdonly(osd_sb(o)->s_bdev);
-			o->od_dt_dev.dd_rdonly = 1;
-			LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
-				      name);
-		} else {
-			LCONSOLE_WARN("%s: not support dev_rdonly on this device",
-				      name);
-
-			GOTO(out_mnt, rc = -EOPNOTSUPP);
-		}
-	} else if (priv_dev_check_rdonly &&
-		   priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
-		CERROR("%s: underlying device %s is marked as "
-		       "read-only. Setup failed\n", name, dev);
+		LCONSOLE_WARN("%s: not support dev_rdonly on this device",
+			      name);
 
-		GOTO(out_mnt, rc = -EROFS);
+		GOTO(out_mnt, rc = -EOPNOTSUPP);
 	}
 
 	if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
@@ -7738,6 +7884,10 @@ static struct lu_device *osd_device_fini(const struct lu_env *env,
 	osd_index_backup(env, o, false);
 	osd_shutdown(env, o);
 	osd_procfs_fini(o);
+	if (o->od_oi_table != NULL)
+		osd_oi_fini(osd_oti_get(env), o);
+	if (o->od_extent_bytes_percpu)
+		free_percpu(o->od_extent_bytes_percpu);
 	osd_obj_map_fini(o);
 	osd_umount(env, o);
 
@@ -7771,12 +7921,16 @@ static int osd_device_init0(const struct lu_env *env,
 	spin_lock_init(&o->od_lock);
 	o->od_index_backup_policy = LIBP_NONE;
 	o->od_t10_type = 0;
+	init_waitqueue_head(&o->od_commit_cb_done);
 
 	o->od_read_cache = 1;
 	o->od_writethrough_cache = 1;
 	o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
-
+	o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
+	o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
 	o->od_auto_scrub_interval = AS_DEFAULT;
+	/* default fallocate to unwritten extents: LU-14326/LU-14333 */
+	o->od_fallocate_zero_blocks = 0;
 
 	cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
 			sizeof(o->od_svname));
@@ -7859,6 +8013,12 @@ static int osd_device_init0(const struct lu_env *env,
 		GOTO(out_procfs, rc);
 	}
 
+	o->od_extent_bytes_percpu = alloc_percpu(unsigned int);
+	if (!o->od_extent_bytes_percpu) {
+		rc = -ENOMEM;
+		GOTO(out_procfs, rc);
+	}
+
 	RETURN(0);
 
 out_procfs:
@@ -7914,10 +8074,8 @@ static struct lu_device *osd_device_free(const struct lu_env *env,
 	/* XXX: make osd top device in order to release reference */
 	d->ld_site->ls_top_dev = d;
 	lu_site_purge(env, d->ld_site, -1);
-	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_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
+		      D_ERROR, lu_cdebug_printer);
 	lu_site_fini(&o->od_site);
 	dt_device_fini(&o->od_dt_dev);
 	OBD_FREE_PTR(o);
@@ -8089,10 +8247,18 @@ static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
 	RETURN(result);
 }
 
-static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
-			 struct lu_fid *fid, struct md_op_data *op_data)
+/**
+ * Implementation of lu_device_operations::ldo_fid_alloc() for OSD
+ *
+ * Allocate FID.
+ *
+ * see include/lu_object.h for the details.
+ */
+static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d,
+			 struct lu_fid *fid, struct lu_object *parent,
+			 const struct lu_name *name)
 {
-	struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
+	struct osd_device *osd = osd_dev(d);
 
 	return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
 }
@@ -8111,6 +8277,7 @@ const struct lu_device_operations osd_lu_ops = {
 	.ldo_process_config    = osd_process_config,
 	.ldo_recovery_complete = osd_recovery_complete,
 	.ldo_prepare           = osd_prepare,
+	.ldo_fid_alloc	       = osd_fid_alloc,
 };
 
 static const struct lu_device_type_operations osd_device_type_ops = {
@@ -8145,11 +8312,10 @@ static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
 /*
  * lprocfs legacy support.
  */
-static struct obd_ops osd_obd_device_ops = {
+static const struct obd_ops osd_obd_device_ops = {
 	.o_owner = THIS_MODULE,
 	.o_connect	= osd_obd_connect,
 	.o_disconnect	= osd_obd_disconnect,
-	.o_fid_alloc	= osd_fid_alloc,
 	.o_health_check = osd_health_check,
 };
 
@@ -8182,10 +8348,11 @@ static int __init osd_init(void)
 	struct kobject *kobj;
 	int rc;
 
-	CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
+	BUILD_BUG_ON(BH_DXLock >=
+		     sizeof(((struct buffer_head *)0)->b_state) * 8);
 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
 	/* please, try to keep osd_thread_info smaller than a page */
-	CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
+	BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
 #endif
 
 	osd_oi_mod_init();
@@ -8195,9 +8362,8 @@ static int __init osd_init(void)
 		return rc;
 
 #ifdef CONFIG_KALLSYMS
-	priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
-	priv_dev_check_rdonly =
-		(void *)kallsyms_lookup_name("dev_check_rdonly");
+	priv_security_file_alloc =
+		(void *)kallsyms_lookup_name("security_file_alloc");
 #endif
 
 	rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,