X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_handler.c;h=cb0728dc1a93c897bdf262fa5e4d2f8fb0dd57f4;hp=2b7c8224434e81b9b912f0112e17f393cc4f10eb;hb=3fc039a99f41725798d157a422174005a49200c8;hpb=5afd3300d1ded80f79dbea92d47d287ad75f463a

diff --git a/lustre/osd-ldiskfs/osd_handler.c b/lustre/osd-ldiskfs/osd_handler.c
index 2b7c822..cb0728d 100644
--- a/lustre/osd-ldiskfs/osd_handler.c
+++ b/lustre/osd-ldiskfs/osd_handler.c
@@ -41,7 +41,7 @@
  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  */
 
-#define DEBUG_SUBSYSTEM S_MDS
+#define DEBUG_SUBSYSTEM S_OSD
 
 #include <linux/module.h>
 
@@ -72,6 +72,8 @@
 #include <md_object.h>
 #include <lustre_quota.h>
 
+#include <ldiskfs/xattr.h>
+
 int ldiskfs_pdo = 1;
 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
                 "ldiskfs with parallel directory operations");
@@ -133,10 +135,6 @@ static int osd_object_invariant(const struct lu_object *l)
 /*
  * Concurrency: doesn't matter
  */
-static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
-{
-        return osd_oti_get(env)->oti_r_locks > 0;
-}
 
 /*
  * Concurrency: doesn't matter
@@ -468,9 +466,6 @@ static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 	int			 rc;
 	ENTRY;
 
-	if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
-		RETURN(0);
-
 	CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 	rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 			     info->oti_mdt_attrs_old, LMA_OLD_SIZE);
@@ -591,7 +586,7 @@ static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 	}
 
 	id = &info->oti_id;
-	if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
+	if (!list_empty(&scrub->os_inconsistent_items)) {
 		/* Search order: 2. OI scrub pending list. */
 		result = osd_oii_lookup(dev, fid, id);
 		if (result == 0)
@@ -917,11 +912,11 @@ static void osd_trans_commit_cb(struct super_block *sb,
         dt_txn_hook_commit(th);
 
 	/* call per-transaction callbacks if any */
-	cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
+	list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 		LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 			 "commit callback entry: magic=%x name='%s'\n",
 			 dcb->dcb_magic, dcb->dcb_name);
-		cfs_list_del_init(&dcb->dcb_linkage);
+		list_del_init(&dcb->dcb_linkage);
 		dcb->dcb_func(NULL, th, dcb, error);
 	}
 
@@ -931,7 +926,7 @@ static void osd_trans_commit_cb(struct super_block *sb,
 
         lu_context_exit(&th->th_ctx);
         lu_context_fini(&th->th_ctx);
-        OBD_FREE_PTR(oh);
+	thandle_put(th);
 }
 
 static struct thandle *osd_trans_create(const struct lu_env *env,
@@ -946,19 +941,21 @@ static struct thandle *osd_trans_create(const struct lu_env *env,
 	/* on pending IO in this thread should left from prev. request */
 	LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 
-        th = ERR_PTR(-ENOMEM);
-	OBD_ALLOC_GFP(oh, sizeof *oh, __GFP_IO);
-        if (oh != NULL) {
+	th = ERR_PTR(-ENOMEM);
+	OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
+	if (oh != NULL) {
 		oh->ot_quota_trans = &oti->oti_quota_trans;
 		memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
-                th = &oh->ot_super;
-                th->th_dev = d;
-                th->th_result = 0;
-                th->th_tags = LCT_TX_HANDLE;
-                oh->ot_credits = 0;
-                oti->oti_dev = osd_dt_dev(d);
-                CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
-                osd_th_alloced(oh);
+		th = &oh->ot_super;
+		th->th_dev = d;
+		th->th_result = 0;
+		th->th_tags = LCT_TX_HANDLE;
+		oh->ot_credits = 0;
+		atomic_set(&th->th_refc, 1);
+		th->th_alloc_size = sizeof(*oh);
+		oti->oti_dev = osd_dt_dev(d);
+		INIT_LIST_HEAD(&oh->ot_dcb_list);
+		osd_th_alloced(oh);
 
 		memset(oti->oti_declare_ops, 0,
 					sizeof(oti->oti_declare_ops));
@@ -967,8 +964,8 @@ static struct thandle *osd_trans_create(const struct lu_env *env,
 		memset(oti->oti_declare_ops_cred, 0,
 					sizeof(oti->oti_declare_ops_cred));
 		oti->oti_rollback = false;
-        }
-        RETURN(th);
+	}
+	RETURN(th);
 }
 
 /*
@@ -1003,11 +1000,9 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 		      LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
 		      oh->ot_credits,
 		      osd_journal(dev)->j_max_transaction_buffers);
-		CWARN("  create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
+		CWARN("  create: %u/%u, destroy: %u/%u\n",
 		      oti->oti_declare_ops[OSD_OT_CREATE],
 		      oti->oti_declare_ops_cred[OSD_OT_CREATE],
-		      oti->oti_declare_ops[OSD_OT_DELETE],
-		      oti->oti_declare_ops_cred[OSD_OT_DELETE],
 		      oti->oti_declare_ops[OSD_OT_DESTROY],
 		      oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
 		CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
@@ -1025,8 +1020,8 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 		CWARN("  insert: %u/%u, delete: %u/%u\n",
 		      oti->oti_declare_ops[OSD_OT_INSERT],
 		      oti->oti_declare_ops_cred[OSD_OT_INSERT],
-		      oti->oti_declare_ops[OSD_OT_DESTROY],
-		      oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
+		      oti->oti_declare_ops[OSD_OT_DELETE],
+		      oti->oti_declare_ops_cred[OSD_OT_DELETE]);
 		CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
 		      oti->oti_declare_ops[OSD_OT_REF_ADD],
 		      oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
@@ -1080,8 +1075,8 @@ static int osd_seq_exists(const struct lu_env *env,
 	int			rc;
 	ENTRY;
 
-	if (ss == NULL)
-		RETURN(1);
+	LASSERT(ss != NULL);
+	LASSERT(ss->ss_server_fld != NULL);
 
 	rc = osd_fld_lookup(env, osd, seq, range);
 	if (rc != 0) {
@@ -1097,21 +1092,20 @@ static int osd_seq_exists(const struct lu_env *env,
 /*
  * Concurrency: shouldn't matter.
  */
-static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
+static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
+			  struct thandle *th)
 {
-        int                     rc = 0;
-        struct osd_thandle     *oh;
-        struct osd_thread_info *oti = osd_oti_get(env);
-        struct osd_iobuf       *iobuf = &oti->oti_iobuf;
+	int                     rc = 0;
+	struct osd_thandle     *oh;
+	struct osd_thread_info *oti = osd_oti_get(env);
+	struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 	struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
-        ENTRY;
+	struct lquota_trans    *qtrans;
+	ENTRY;
 
-        oh = container_of0(th, struct osd_thandle, ot_super);
+	oh = container_of0(th, struct osd_thandle, ot_super);
 
-	if (qsd != NULL)
-		/* inform the quota slave device that the transaction is
-		 * stopping */
-		qsd_op_end(env, qsd, oh->ot_quota_trans);
+	qtrans = oh->ot_quota_trans;
 	oh->ot_quota_trans = NULL;
 
         if (oh->ot_handle != NULL) {
@@ -1140,9 +1134,12 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
                 if (rc != 0)
                         CERROR("Failure to stop transaction: %d\n", rc);
         } else {
-                OBD_FREE_PTR(oh);
+		thandle_put(&oh->ot_super);
         }
 
+	/* inform the quota slave device that the transaction is stopping */
+	qsd_op_end(env, qsd, qtrans);
+
 	/* as we want IO to journal and data IO be concurrent, we don't block
 	 * awaiting data IO completion in osd_do_bio(), instead we wait here
 	 * once transaction is submitted to the journal. all reqular requests
@@ -1154,6 +1151,7 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
 	 */
 	wait_event(iobuf->dr_wait,
 		       atomic_read(&iobuf->dr_numreqs) == 0);
+	osd_fini_iobuf(oti->oti_dev, iobuf);
 	if (!rc)
 		rc = iobuf->dr_error;
 
@@ -1167,7 +1165,7 @@ static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
 
 	LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
 	LASSERT(&dcb->dcb_func != NULL);
-	cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
+	list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
 
 	return 0;
 }
@@ -1193,8 +1191,8 @@ static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
         osd_index_fini(obj);
         if (inode != NULL) {
 		struct qsd_instance	*qsd = osd_obj2dev(obj)->od_quota_slave;
-		qid_t			 uid = inode->i_uid;
-		qid_t			 gid = inode->i_gid;
+		qid_t			 uid = i_uid_read(inode);
+		qid_t			 gid = i_gid_read(inode);
 
                 iput(inode);
                 obj->oo_inode = NULL;
@@ -1242,6 +1240,8 @@ static int osd_object_print(const struct lu_env *env, void *cookie,
                     d ? d->id_ops->id_name : "plain");
 }
 
+#define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
+
 /*
  * Concurrency: shouldn't matter.
  */
@@ -1266,24 +1266,28 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d,
         }
 
 	spin_lock(&osd->od_osfs_lock);
-	/* cache 1 second */
-	if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
-		result = sb->s_op->statfs(sb->s_root, ksfs);
-		if (likely(result == 0)) { /* N.B. statfs can't really fail */
-			osd->od_osfs_age = cfs_time_current_64();
-			statfs_pack(&osd->od_statfs, ksfs);
-			if (sb->s_flags & MS_RDONLY)
-				sfs->os_state = OS_STATE_READONLY;
-		}
+	result = sb->s_op->statfs(sb->s_root, ksfs);
+	if (likely(result == 0)) { /* N.B. statfs can't really fail */
+		statfs_pack(sfs, ksfs);
+		if (sb->s_flags & MS_RDONLY)
+			sfs->os_state = OS_STATE_READONLY;
 	}
 
-	if (likely(result == 0))
-		*sfs = osd->od_statfs;
 	spin_unlock(&osd->od_osfs_lock);
 
-        if (unlikely(env == NULL))
+	if (unlikely(env == NULL))
                 OBD_FREE_PTR(ksfs);
 
+	/* Reserve a small amount of space for local objects like last_rcvd,
+	 * llog, quota files, ... */
+	if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
+		sfs->os_bavail = 0;
+	} else {
+		sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
+		/** Take out metadata overhead for indirect blocks */
+		sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
+	}
+
         return result;
 }
 
@@ -1306,6 +1310,7 @@ static void osd_conf_get(const struct lu_env *env,
                          struct dt_device_param *param)
 {
         struct super_block *sb = osd_sb(osd_dt_dev(dev));
+	int		   ea_overhead;
 
         /*
          * XXX should be taken from not-yet-existing fs abstraction layer.
@@ -1329,13 +1334,19 @@ static void osd_conf_get(const struct lu_env *env,
         if (test_opt(sb, POSIX_ACL))
                 param->ddp_mntopts |= MNTOPT_ACL;
 
+	/* LOD might calculate the max stripe count based on max_ea_size,
+	 * so we need take account in the overhead as well,
+	 * xattr_header + magic + xattr_entry_head */
+	ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
+		      LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
+
 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
-        if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
-                param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
-        else
+	if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
+		param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
+								ea_overhead;
+	else
 #endif
-                param->ddp_max_ea_size = sb->s_blocksize;
-
+		param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
 }
 
 /*
@@ -1424,55 +1435,57 @@ static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
  * If we mount with --data_journal we may need more.
  */
 const int osd_dto_credits_noquota[DTO_NR] = {
-        /**
-         * Insert/Delete.
-         * INDEX_EXTRA_TRANS_BLOCKS(8) +
-         * SINGLEDATA_TRANS_BLOCKS(8)
-         * XXX Note: maybe iam need more, since iam have more level than
-         *           EXT3 htree.
-         */
-        [DTO_INDEX_INSERT]  = 16,
-        [DTO_INDEX_DELETE]  = 16,
-        /**
+	/**
+	 * Insert.
+	 * INDEX_EXTRA_TRANS_BLOCKS(8) +
+	 * SINGLEDATA_TRANS_BLOCKS(8)
+	 * XXX Note: maybe iam need more, since iam have more level than
+	 *           EXT3 htree.
+	 */
+	[DTO_INDEX_INSERT]  = 16,
+	/**
+	 * Delete
+	 * just modify a single entry, probably merge few within a block
+	 */
+	[DTO_INDEX_DELETE]  = 1,
+	/**
 	 * Used for OI scrub
-         */
-        [DTO_INDEX_UPDATE]  = 16,
-        /**
-         * Create a object. The same as create object in EXT3.
-         * DATA_TRANS_BLOCKS(14) +
-         * INDEX_EXTRA_BLOCKS(8) +
-         * 3(inode bits, groups, GDT)
-         */
-        [DTO_OBJECT_CREATE] = 25,
-        /**
-         * XXX: real credits to be fixed
-         */
-        [DTO_OBJECT_DELETE] = 25,
-        /**
-         * Attr set credits (inode)
-         */
-        [DTO_ATTR_SET_BASE] = 1,
-        /**
-         * Xattr set. The same as xattr of EXT3.
-         * DATA_TRANS_BLOCKS(14)
-         * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
-         * are also counted in. Do not know why?
-         */
-        [DTO_XATTR_SET]     = 14,
-        [DTO_LOG_REC]       = 14,
-        /**
-         * credits for inode change during write.
-         */
-        [DTO_WRITE_BASE]    = 3,
-        /**
-         * credits for single block write.
-         */
-        [DTO_WRITE_BLOCK]   = 14,
-        /**
-         * Attr set credits for chown.
-         * This is extra credits for setattr, and it is null without quota
-         */
-        [DTO_ATTR_SET_CHOWN]= 0
+	 */
+	[DTO_INDEX_UPDATE]  = 16,
+	/**
+	 * 4(inode, inode bits, groups, GDT)
+	 *   notice: OI updates are counted separately with DTO_INDEX_INSERT
+	 */
+	[DTO_OBJECT_CREATE] = 4,
+	/**
+	 * 4(inode, inode bits, groups, GDT)
+	 *   notice: OI updates are counted separately with DTO_INDEX_DELETE
+	 */
+	[DTO_OBJECT_DELETE] = 4,
+	/**
+	 * Attr set credits (inode)
+	 */
+	[DTO_ATTR_SET_BASE] = 1,
+	/**
+	 * Xattr set. The same as xattr of EXT3.
+	 * DATA_TRANS_BLOCKS(14)
+	 * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
+	 * are also counted in. Do not know why?
+	 */
+	[DTO_XATTR_SET]     = 14,
+	/**
+	 * credits for inode change during write.
+	 */
+	[DTO_WRITE_BASE]    = 3,
+	/**
+	 * credits for single block write.
+	 */
+	[DTO_WRITE_BLOCK]   = 14,
+	/**
+	 * Attr set credits for chown.
+	 * This is extra credits for setattr, and it is null without quota
+	 */
+	[DTO_ATTR_SET_CHOWN] = 0
 };
 
 static const struct dt_device_operations osd_dt_ops = {
@@ -1674,26 +1687,26 @@ static struct timespec *osd_inode_time(const struct lu_env *env,
 
 
 static void osd_inode_getattr(const struct lu_env *env,
-                              struct inode *inode, struct lu_attr *attr)
-{
-        attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
-                               LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
-                               LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
-			       LA_TYPE;
-
-        attr->la_atime      = LTIME_S(inode->i_atime);
-        attr->la_mtime      = LTIME_S(inode->i_mtime);
-        attr->la_ctime      = LTIME_S(inode->i_ctime);
-        attr->la_mode       = inode->i_mode;
-        attr->la_size       = i_size_read(inode);
-        attr->la_blocks     = inode->i_blocks;
-        attr->la_uid        = inode->i_uid;
-        attr->la_gid        = inode->i_gid;
-        attr->la_flags      = LDISKFS_I(inode)->i_flags;
-        attr->la_nlink      = inode->i_nlink;
-        attr->la_rdev       = inode->i_rdev;
-	attr->la_blksize    = 1 << inode->i_blkbits;
-	attr->la_blkbits    = inode->i_blkbits;
+			      struct inode *inode, struct lu_attr *attr)
+{
+	attr->la_valid	|= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
+			   LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
+			   LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
+			   LA_TYPE;
+
+	attr->la_atime	 = LTIME_S(inode->i_atime);
+	attr->la_mtime	 = LTIME_S(inode->i_mtime);
+	attr->la_ctime	 = LTIME_S(inode->i_ctime);
+	attr->la_mode	 = inode->i_mode;
+	attr->la_size	 = i_size_read(inode);
+	attr->la_blocks	 = inode->i_blocks;
+	attr->la_uid	 = i_uid_read(inode);
+	attr->la_gid	 = i_gid_read(inode);
+	attr->la_flags	 = LDISKFS_I(inode)->i_flags;
+	attr->la_nlink	 = inode->i_nlink;
+	attr->la_rdev	 = inode->i_rdev;
+	attr->la_blksize = 1 << inode->i_blkbits;
+	attr->la_blkbits = inode->i_blkbits;
 }
 
 static int osd_attr_get(const struct lu_env *env,
@@ -1724,6 +1737,8 @@ static int osd_declare_attr_set(const struct lu_env *env,
 	struct osd_object      *obj;
 	struct osd_thread_info *info = osd_oti_get(env);
 	struct lquota_id_info  *qi = &info->oti_qi;
+	qid_t			uid;
+	qid_t			gid;
 	long long               bspace;
 	int			rc = 0;
 	bool			allocated;
@@ -1755,7 +1770,7 @@ static int osd_declare_attr_set(const struct lu_env *env,
 	 * credits for updating quota accounting files and to trigger quota
 	 * space adjustment once the operation is completed.*/
 	if ((attr->la_valid & LA_UID) != 0 &&
-	     attr->la_uid != obj->oo_inode->i_uid) {
+	     attr->la_uid != (uid = i_uid_read(obj->oo_inode))) {
 		qi->lqi_type = USRQUOTA;
 
 		/* inode accounting */
@@ -1772,7 +1787,7 @@ static int osd_declare_attr_set(const struct lu_env *env,
 			RETURN(rc);
 
 		/* and one less inode for the current uid */
-		qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
+		qi->lqi_id.qid_uid = uid;
 		qi->lqi_space      = -1;
 		rc = osd_declare_qid(env, oh, qi, true, NULL);
 		if (rc == -EDQUOT || rc == -EINPROGRESS)
@@ -1794,7 +1809,7 @@ static int osd_declare_attr_set(const struct lu_env *env,
 			RETURN(rc);
 
 		/* and finally less blocks for the current owner */
-		qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
+		qi->lqi_id.qid_uid = uid;
 		qi->lqi_space      = -bspace;
 		rc = osd_declare_qid(env, oh, qi, true, NULL);
 		if (rc == -EDQUOT || rc == -EINPROGRESS)
@@ -1804,7 +1819,7 @@ static int osd_declare_attr_set(const struct lu_env *env,
 	}
 
 	if (attr->la_valid & LA_GID &&
-	    attr->la_gid != obj->oo_inode->i_gid) {
+	    attr->la_gid != (gid = i_gid_read(obj->oo_inode))) {
 		qi->lqi_type = GRPQUOTA;
 
 		/* inode accounting */
@@ -1821,7 +1836,7 @@ static int osd_declare_attr_set(const struct lu_env *env,
 			RETURN(rc);
 
 		/* and one less inode for the current gid */
-		qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
+		qi->lqi_id.qid_gid = gid;
 		qi->lqi_space      = -1;
 		rc = osd_declare_qid(env, oh, qi, true, NULL);
 		if (rc == -EDQUOT || rc == -EINPROGRESS)
@@ -1843,7 +1858,7 @@ static int osd_declare_attr_set(const struct lu_env *env,
 			RETURN(rc);
 
 		/* and finally less blocks for the current owner */
-		qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
+		qi->lqi_id.qid_gid = gid;
 		qi->lqi_space      = -bspace;
 		rc = osd_declare_qid(env, oh, qi, true, NULL);
 		if (rc == -EDQUOT || rc == -EINPROGRESS)
@@ -1856,11 +1871,16 @@ static int osd_declare_attr_set(const struct lu_env *env,
 }
 
 static int osd_inode_setattr(const struct lu_env *env,
-                             struct inode *inode, const struct lu_attr *attr)
+			     struct inode *inode, const struct lu_attr *attr)
 {
-        __u64 bits;
+	__u64 bits = attr->la_valid;
+
+	/* Only allow set size for regular file */
+	if (!S_ISREG(inode->i_mode))
+		bits &= ~(LA_SIZE | LA_BLOCKS);
 
-        bits = attr->la_valid;
+	if (bits == 0)
+		return 0;
 
         if (bits & LA_ATIME)
                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
@@ -1879,17 +1899,17 @@ static int osd_inode_setattr(const struct lu_env *env,
         if (bits & LA_BLOCKS)
                 inode->i_blocks = attr->la_blocks;
 #endif
-        if (bits & LA_MODE)
-                inode->i_mode   = (inode->i_mode & S_IFMT) |
-                        (attr->la_mode & ~S_IFMT);
-        if (bits & LA_UID)
-                inode->i_uid    = attr->la_uid;
-        if (bits & LA_GID)
-                inode->i_gid    = attr->la_gid;
-        if (bits & LA_NLINK)
+	if (bits & LA_MODE)
+		inode->i_mode = (inode->i_mode & S_IFMT) |
+				(attr->la_mode & ~S_IFMT);
+	if (bits & LA_UID)
+		i_uid_write(inode, attr->la_uid);
+	if (bits & LA_GID)
+		i_gid_write(inode, attr->la_gid);
+	if (bits & LA_NLINK)
 		set_nlink(inode, attr->la_nlink);
-        if (bits & LA_RDEV)
-                inode->i_rdev   = attr->la_rdev;
+	if (bits & LA_RDEV)
+		inode->i_rdev = attr->la_rdev;
 
         if (bits & LA_FLAGS) {
                 /* always keep S_NOCMTIME */
@@ -1901,8 +1921,8 @@ static int osd_inode_setattr(const struct lu_env *env,
 
 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
 {
-	if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
-	    (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
+	if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
+	    (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
 		struct iattr	iattr;
 		int		rc;
 
@@ -1911,8 +1931,8 @@ static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
 			iattr.ia_valid |= ATTR_UID;
 		if (attr->la_valid & LA_GID)
 			iattr.ia_valid |= ATTR_GID;
-		iattr.ia_uid = attr->la_uid;
-		iattr.ia_gid = attr->la_gid;
+		iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
+		iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
 
 		rc = ll_vfs_dq_transfer(inode, &iattr);
 		if (rc) {
@@ -2195,7 +2215,6 @@ static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
 {
         LASSERT(ah);
 
-        memset(ah, 0, sizeof(*ah));
         ah->dah_parent = parent;
         ah->dah_mode = child_mode;
 }
@@ -2341,12 +2360,10 @@ static int osd_declare_object_create(const struct lu_env *env,
 
 	osd_trans_declare_op(env, oh, OSD_OT_CREATE,
 			     osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
-	if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
-			   lu_object_fid(&dt->do_lu), OI_CHECK_FLD))
-		/* Reuse idle OI block may cause additional one OI block
-		 * to be changed. */
-		osd_trans_declare_op(env, oh, OSD_OT_INSERT,
-				osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
+	/* Reuse idle OI block may cause additional one OI block
+	 * to be changed. */
+	osd_trans_declare_op(env, oh, OSD_OT_INSERT,
+			     osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
 
 	/* If this is directory, then we expect . and .. to be inserted as
 	 * well. The one directory block always needs to be created for the
@@ -2433,13 +2450,13 @@ static int osd_declare_object_destroy(const struct lu_env *env,
 	osd_trans_declare_op(env, oh, OSD_OT_DELETE,
 			     osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
 	/* one less inode */
-	rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
-				   false, true, NULL, false);
+	rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
+				   -1, oh, false, true, NULL, false);
 	if (rc)
 		RETURN(rc);
 	/* data to be truncated */
-	rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
-				   true, true, NULL, false);
+	rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
+				   0, oh, true, true, NULL, false);
 	RETURN(rc);
 }
 
@@ -2777,6 +2794,8 @@ static int osd_object_ref_add(const struct lu_env *env,
 
 	osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
 
+	CDEBUG(D_INODE, DFID" increase nlink %d\n",
+	       PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
 	/*
 	 * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
 	 * (65000) subdirectories by storing "1" in i_nlink if the link count
@@ -2788,9 +2807,17 @@ static int osd_object_ref_add(const struct lu_env *env,
 	 * This also has to properly handle the case of inodes with nlink == 0
 	 * in case they are being linked into the PENDING directory
 	 */
+#ifdef I_LINKABLE
+	/* This is necessary to increment from i_nlink == 0 */
+	spin_lock(&inode->i_lock);
+	inode->i_state |= I_LINKABLE;
+	spin_unlock(&inode->i_lock);
+#endif
+
 	spin_lock(&obj->oo_guard);
 	ldiskfs_inc_count(oh->ot_handle, inode);
-	LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
+	if (!S_ISDIR(inode->i_mode))
+		LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
 	spin_unlock(&obj->oo_guard);
 
 	ll_dirty_inode(inode, I_DIRTY_DATASYNC);
@@ -2851,6 +2878,9 @@ static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
 		return 0;
 	}
 
+	CDEBUG(D_INODE, DFID" decrease nlink %d\n",
+	       PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
+
 	ldiskfs_dec_count(oh->ot_handle, inode);
 	spin_unlock(&obj->oo_guard);
 
@@ -2911,16 +2941,32 @@ static int osd_declare_xattr_set(const struct lu_env *env,
                                  int fl, struct thandle *handle)
 {
 	struct osd_thandle *oh;
+	int credits;
 
 	LASSERT(handle != NULL);
 
 	oh = container_of0(handle, struct osd_thandle, ot_super);
 	LASSERT(oh->ot_handle == NULL);
 
-	osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
-			     strcmp(name, XATTR_NAME_VERSION) == 0 ?
-			     osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
-			     osd_dto_credits_noquota[DTO_XATTR_SET]);
+	/* optimistic optimization: LMA is set first and usually fit inode */
+	if (strcmp(name, XATTR_NAME_LMA) == 0) {
+		if (dt_object_exists(dt))
+			credits = 0;
+		else
+			credits = 1;
+	} else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
+		credits = 1;
+	} else {
+		struct osd_device  *osd = osd_dev(dt->do_lu.lo_dev);
+		struct super_block *sb = osd_sb(osd);
+		credits = osd_dto_credits_noquota[DTO_XATTR_SET];
+		if (buf && buf->lb_len > sb->s_blocksize) {
+			credits *= (buf->lb_len + sb->s_blocksize - 1) >>
+					sb->s_blocksize_bits;
+		}
+	}
+
+	osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
 
 	return 0;
 }
@@ -2970,6 +3016,9 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
                 return -EACCES;
 
+	CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
+	       PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
+
 	osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
 	if (fl & LU_XATTR_REPLACE)
 		fs_flags |= XATTR_REPLACE;
@@ -2977,6 +3026,22 @@ static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
 	if (fl & LU_XATTR_CREATE)
 		fs_flags |= XATTR_CREATE;
 
+	if (strcmp(name, XATTR_NAME_LMV) == 0) {
+		struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
+		int			 rc;
+
+		rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
+		if (rc != 0)
+			RETURN(rc);
+
+		lma->lma_incompat |= LMAI_STRIPED;
+		lustre_lma_swab(lma);
+		rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
+				     sizeof(*lma), XATTR_REPLACE);
+		if (rc != 0)
+			RETURN(rc);
+	}
+
 	return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
 			       fs_flags);
 }
@@ -2994,7 +3059,6 @@ static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
 
 	LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
-        LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
 
         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
                 return -EACCES;
@@ -3084,16 +3148,16 @@ static struct obd_capa *osd_capa_get(const struct lu_env *env,
 	case LC_ID_NONE:
 		RETURN(NULL);
 	case LC_ID_PLAIN:
-		capa->lc_uid = obj->oo_inode->i_uid;
-		capa->lc_gid = obj->oo_inode->i_gid;
+		capa->lc_uid = i_uid_read(obj->oo_inode);
+		capa->lc_gid = i_gid_read(obj->oo_inode);
 		capa->lc_flags = LC_ID_PLAIN;
 		break;
 	case LC_ID_CONVERT: {
 		__u32 d[4], s[4];
 
-		s[0] = obj->oo_inode->i_uid;
+		s[0] = i_uid_read(obj->oo_inode);
 		cfs_get_random_bytes(&(s[1]), sizeof(__u32));
-		s[2] = obj->oo_inode->i_gid;
+		s[2] = i_uid_read(obj->oo_inode);
 		cfs_get_random_bytes(&(s[3]), sizeof(__u32));
 		rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
 		if (unlikely(rc))
@@ -3137,7 +3201,8 @@ static struct obd_capa *osd_capa_get(const struct lu_env *env,
 	RETURN(oc);
 }
 
-static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
+static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
+			   __u64 start, __u64 end)
 {
 	struct osd_object	*obj    = osd_dt_obj(dt);
 	struct inode		*inode  = obj->oo_inode;
@@ -3154,13 +3219,19 @@ static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
 	file->f_mapping = inode->i_mapping;
 	file->f_op = inode->i_fop;
 	set_file_inode(file, inode);
-#ifndef HAVE_FILE_FSYNC_4ARGS
+
+#ifdef HAVE_FILE_FSYNC_4ARGS
+	rc = file->f_op->fsync(file, start, end, 0);
+#elif defined(HAVE_FILE_FSYNC_2ARGS)
 	mutex_lock(&inode->i_mutex);
-#endif
-	rc = do_fsync(file, 0);
-#ifndef HAVE_FILE_FSYNC_4ARGS
+	rc = file->f_op->fsync(file, 0);
+	mutex_unlock(&inode->i_mutex);
+#else
+	mutex_lock(&inode->i_mutex);
+	rc = file->f_op->fsync(file, dentry, 0);
 	mutex_unlock(&inode->i_mutex);
 #endif
+
 	RETURN(rc);
 }
 
@@ -3282,7 +3353,8 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
 			 * recheck under lock.
 			 */
 			if (!osd_has_index(obj))
-				result = osd_iam_container_init(env, obj, dir);
+				result = osd_iam_container_init(env, obj,
+								obj->oo_dir);
 			else
 				result = 0;
 			up_write(&obj->oo_ext_idx_sem);
@@ -3478,8 +3550,8 @@ static int osd_index_declare_ea_delete(const struct lu_env *env,
 	inode = osd_dt_obj(dt)->oo_inode;
 	LASSERT(inode);
 
-	rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
-				   true, true, NULL, false);
+	rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
+				   0, oh, true, true, NULL, false);
 	RETURN(rc);
 }
 
@@ -3499,12 +3571,20 @@ static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 			  struct lu_fid *fid)
 {
+	struct seq_server_site	*ss = osd_seq_site(osd);
 	ENTRY;
 
 	/* FID seqs not in FLDB, must be local seq */
 	if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
 		RETURN(0);
 
+	/* If FLD is not being initialized yet, it only happens during the
+	 * initialization, likely during mgs initialization, and we assume
+	 * this is local FID. */
+	if (ss == NULL || ss->ss_server_fld == NULL)
+		RETURN(0);
+
+	/* Only check the local FLDB here */
 	if (osd_seq_exists(env, osd, fid_seq(fid)))
 		RETURN(0);
 
@@ -4167,15 +4247,17 @@ struct osd_object *osd_object_find(const struct lu_env *env,
 	 * in the cache, otherwise lu_object_alloc() crashes
 	 * -bzzz
 	 */
-	luch = lu_object_find_at(env, ludev, fid, NULL);
-        if (!IS_ERR(luch)) {
-                if (lu_object_exists(luch)) {
-                        lo = lu_object_locate(luch->lo_header, ludev->ld_type);
-                        if (lo != NULL)
-                                child = osd_obj(lo);
-                        else
-                                LU_OBJECT_DEBUG(D_ERROR, env, luch,
-                                                "lu_object can't be located"
+	luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
+				 ludev : ludev->ld_site->ls_top_dev,
+				 fid, NULL);
+	if (!IS_ERR(luch)) {
+		if (lu_object_exists(luch)) {
+			lo = lu_object_locate(luch->lo_header, ludev->ld_type);
+			if (lo != NULL)
+				child = osd_obj(lo);
+			else
+				LU_OBJECT_DEBUG(D_ERROR, env, luch,
+						"lu_object can't be located"
 						DFID"\n", PFID(fid));
 
                         if (child == NULL) {
@@ -4220,7 +4302,7 @@ static int osd_index_declare_ea_insert(const struct lu_env *env,
 	int			rc;
 	ENTRY;
 
-	LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
+	LASSERT(!dt_object_remote(dt));
 	LASSERT(handle != NULL);
 
 	oh = container_of0(handle, struct osd_thandle, ot_super);
@@ -4229,21 +4311,14 @@ static int osd_index_declare_ea_insert(const struct lu_env *env,
 	osd_trans_declare_op(env, oh, OSD_OT_INSERT,
 			     osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 
-	if (osd_dt_obj(dt)->oo_inode == NULL) {
-		const char *name  = (const char *)key;
-		/* Object is not being created yet. Only happens when
-		 *     1. declare directory create
-		 *     2. declare insert .
-		 *     3. declare insert ..
-		 */
-		LASSERT(strcmp(name, dotdot) == 0 || strcmp(name, dot) == 0);
-	} else {
+	if (osd_dt_obj(dt)->oo_inode != NULL) {
 		struct inode *inode = osd_dt_obj(dt)->oo_inode;
 
 		/* We ignore block quota on meta pool (MDTs), so needn't
 		 * calculate how many blocks will be consumed by this index
 		 * insert */
-		rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0,
+		rc = osd_declare_inode_qid(env, i_uid_read(inode),
+					   i_gid_read(inode), 0,
 					   oh, true, true, NULL, false);
 	}
 
@@ -4351,6 +4426,9 @@ static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
 
 	rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
 
+	CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
+	       obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
+
 	iput(child_inode);
 	if (child != NULL)
 		osd_object_put(env, child);
@@ -4731,6 +4809,12 @@ 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
  * iterator's in-memory data structure with required
@@ -4742,11 +4826,11 @@ static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
  * \retval 0 on success
  * \retval 1 on buffer full
  */
-static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
+static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
                                loff_t offset, __u64 ino,
                                unsigned d_type)
 {
-        struct osd_it_ea        *it   = (struct osd_it_ea *)buf;
+	struct osd_it_ea	*it   = ((struct osd_filldir_cbs *)buf)->it;
 	struct osd_object	*obj  = it->oie_obj;
         struct osd_it_ea_dirent *ent  = it->oie_dirent;
         struct lu_fid           *fid  = &ent->oied_fid;
@@ -4801,6 +4885,7 @@ static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
  *
  * \retval   0 on success
  * \retval -ve on error
+ * \retval +1 reach the end of entry
  */
 static int osd_ldiskfs_it_fill(const struct lu_env *env,
                                const struct dt_it *di)
@@ -4809,7 +4894,14 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env,
         struct osd_object  *obj   = it->oie_obj;
         struct inode       *inode = obj->oo_inode;
         struct htree_lock  *hlock = NULL;
-        int                 result = 0;
+	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
+	};
 
         ENTRY;
         it->oie_dirent = it->oie_buf;
@@ -4823,22 +4915,31 @@ static int osd_ldiskfs_it_fill(const struct lu_env *env,
 		down_read(&obj->oo_ext_idx_sem);
         }
 
-        result = inode->i_fop->readdir(&it->oie_file, it,
-                                       (filldir_t) osd_ldiskfs_filldir);
+#ifdef HAVE_DIR_CONTEXT
+	buf.ctx.pos = filp->f_pos;
+	rc = inode->i_fop->iterate(filp, &buf.ctx);
+	filp->f_pos = buf.ctx.pos;
+#else
+	rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
+#endif
 
         if (hlock != NULL)
                 ldiskfs_htree_unlock(hlock);
         else
 		up_read(&obj->oo_ext_idx_sem);
 
-        if (it->oie_rd_dirent == 0) {
-                result = -EIO;
-        } else {
-                it->oie_dirent = it->oie_buf;
-                it->oie_it_dirent = 1;
-        }
+	if (it->oie_rd_dirent == 0) {
+		/*If it does not get any dirent, it means it has been reached
+		 *to the end of the dir */
+		it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
+		if (rc == 0)
+			rc = 1;
+	} else {
+		it->oie_dirent = it->oie_buf;
+		it->oie_it_dirent = 1;
+	}
 
-        RETURN(result);
+	RETURN(rc);
 }
 
 /**
@@ -4918,22 +5019,12 @@ osd_dirent_update(handle_t *jh, struct super_block *sb,
 	LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
 
 	rc = ldiskfs_journal_get_write_access(jh, bh);
-	if (rc != 0) {
-		CERROR("%.16s: fail to write access for update dirent: "
-		       "name = %.*s, rc = %d\n",
-		       LDISKFS_SB(sb)->s_es->s_volume_name,
-		       ent->oied_namelen, ent->oied_name, rc);
+	if (rc != 0)
 		RETURN(rc);
-	}
 
 	rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
 	fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
 	rc = ldiskfs_journal_dirty_metadata(jh, bh);
-	if (rc != 0)
-		CERROR("%.16s: fail to dirty metadata for update dirent: "
-		       "name = %.*s, rc = %d\n",
-		       LDISKFS_SB(sb)->s_es->s_volume_name,
-		       ent->oied_namelen, ent->oied_name, rc);
 
 	RETURN(rc);
 }
@@ -4982,13 +5073,8 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
 	if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
 				 dir->i_sb->s_blocksize)) {
 		rc = ldiskfs_journal_get_write_access(jh, bh);
-		if (rc != 0) {
-			CERROR("%.16s: fail to write access for reinsert "
-			       "dirent: name = %.*s, rc = %d\n",
-			       LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
-			       ent->oied_namelen, ent->oied_name, rc);
+		if (rc != 0)
 			RETURN(rc);
-		}
 
 		de->name[de->name_len] = 0;
 		rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
@@ -4997,23 +5083,13 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
 		de->file_type |= LDISKFS_DIRENT_LUFID;
 
 		rc = ldiskfs_journal_dirty_metadata(jh, bh);
-		if (rc != 0)
-			CERROR("%.16s: fail to dirty metadata for reinsert "
-			       "dirent: name = %.*s, rc = %d\n",
-			       LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
-			       ent->oied_namelen, ent->oied_name, rc);
 
 		RETURN(rc);
 	}
 
 	rc = ldiskfs_delete_entry(jh, dir, de, bh);
-	if (rc != 0) {
-		CERROR("%.16s: fail to delete entry for reinsert dirent: "
-		       "name = %.*s, rc = %d\n",
-		       LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
-		       ent->oied_namelen, ent->oied_name, rc);
+	if (rc != 0)
 		RETURN(rc);
-	}
 
 	dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
 					   ent->oied_namelen);
@@ -5025,10 +5101,11 @@ osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
 	/* It is too bad, we cannot reinsert the name entry back.
 	 * That means we lose it! */
 	if (rc != 0)
-		CERROR("%.16s: fail to insert entry for reinsert dirent: "
-		       "name = %.*s, rc = %d\n",
+		CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
+		       "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
 		       LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
-		       ent->oied_namelen, ent->oied_name, rc);
+		       dir->i_ino, dir->i_generation,
+		       ent->oied_namelen, ent->oied_name, PFID(fid), rc);
 
 	RETURN(rc);
 }
@@ -5089,10 +5166,11 @@ again:
 		jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
 		if (IS_ERR(jh)) {
 			rc = PTR_ERR(jh);
-			CERROR("%.16s: fail to start trans for dirent "
-			       "check_repair: credits %d, name %.*s, rc %d\n",
-			       devname, credits, ent->oied_namelen,
-			       ent->oied_name, rc);
+			CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
+			       "check_repair, dir = %lu/%u, credits = %d, "
+			       "name = %.*s: rc = %d\n",
+			       devname, dir->i_ino, dir->i_generation, credits,
+			       ent->oied_namelen, ent->oied_name, rc);
 			RETURN(rc);
 		}
 
@@ -5136,6 +5214,12 @@ again:
 		if (rc == -ENOENT || rc == -ESTALE) {
 			*attr |= LUDA_IGNORE;
 			rc = 0;
+		} else {
+			CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
+			       "check_repair, dir = %lu/%u, name = %.*s: "
+			       "rc = %d\n",
+			       devname, dir->i_ino, dir->i_generation,
+			       ent->oied_namelen, ent->oied_name, rc);
 		}
 
 		GOTO(out_journal, rc);
@@ -5177,6 +5261,13 @@ again:
 			rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
 			if (rc == 0)
 				*attr |= LUDA_REPAIR;
+			else
+				CDEBUG(D_LFSCK, "%.16s: fail to update FID "
+				       "in the dirent, dir = %lu/%u, "
+				       "name = %.*s, "DFID": rc = %d\n",
+				       devname, dir->i_ino, dir->i_generation,
+				       ent->oied_namelen, ent->oied_name,
+				       PFID(fid), rc);
 		} else {
 			/* Do not repair under dryrun mode. */
 			if (*attr & LUDA_VERIFY_DRYRUN) {
@@ -5203,6 +5294,13 @@ again:
 						 fid, bh, de, hlock);
 			if (rc == 0)
 				*attr |= LUDA_REPAIR;
+			else
+				CDEBUG(D_LFSCK, "%.16s: fail to append FID "
+				       "after the dirent, dir = %lu/%u, "
+				       "name = %.*s, "DFID": rc = %d\n",
+				       devname, dir->i_ino, dir->i_generation,
+				       ent->oied_namelen, ent->oied_name,
+				       PFID(fid), rc);
 		}
 	} else if (rc == -ENODATA) {
 		/* Do not repair under dryrun mode. */
@@ -5235,6 +5333,13 @@ again:
 			rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 			if (rc == 0)
 				*attr |= LUDA_REPAIR;
+			else
+				CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
+				       "update dirent, dir = %lu/%u, "
+				       "name = %.*s, "DFID": rc = %d\n",
+				       devname, dir->i_ino, dir->i_generation,
+				       ent->oied_namelen, ent->oied_name,
+				       PFID(fid), rc);
 		} else {
 			lu_igif_build(fid, inode->i_ino, inode->i_generation);
 			/* It is probably IGIF object. Only aappend the
@@ -5243,6 +5348,13 @@ again:
 						 fid, bh, de, hlock);
 			if (rc == 0)
 				*attr |= LUDA_UPGRADE;
+			else
+				CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
+				       "after the dirent, dir = %lu/%u, "
+				       "name = %.*s, "DFID": rc = %d\n",
+				       devname, dir->i_ino, dir->i_generation,
+				       ent->oied_namelen, ent->oied_name,
+				       PFID(fid), rc);
 		}
 	}
 
@@ -5308,7 +5420,10 @@ static inline int osd_it_ea_rec(const struct lu_env *env,
 	} else {
 		attr &= ~LU_DIRENT_ATTRS_MASK;
 		if (!fid_is_sane(fid)) {
-			if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
+			if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP) &&
+			    likely(it->oie_dirent->oied_namelen != 2 ||
+				   it->oie_dirent->oied_name[0] != '.' ||
+				   it->oie_dirent->oied_name[1] != '.'))
 				RETURN(-ENOENT);
 
 			rc = osd_ea_fid_get(env, obj, ino, fid, id);
@@ -5343,6 +5458,27 @@ static inline int osd_it_ea_rec(const struct lu_env *env,
 }
 
 /**
+ * Returns the record size size at current position.
+ *
+ * This function will return record(lu_dirent) size in bytes.
+ *
+ * \param[in] env	execution environment
+ * \param[in] di	iterator's in memory structure
+ * \param[in] attr	attribute of the entry, only requires LUDA_TYPE to
+ *                      calculate the lu_dirent size.
+ *
+ * \retval	record size(in bytes & in memory) of the current lu_dirent
+ *              entry.
+ */
+static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
+			      __u32 attr)
+{
+	struct osd_it_ea *it = (struct osd_it_ea *)di;
+
+	return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
+}
+
+/**
  * Returns a cookie for current position of the iterator head, so that
  * user can use this cookie to load/start the iterator next time.
  *
@@ -5377,6 +5513,9 @@ static int osd_it_ea_load(const struct lu_env *env,
         it->oie_file.f_pos = hash;
 
         rc =  osd_ldiskfs_it_fill(env, di);
+	if (rc > 0)
+		rc = -ENODATA;
+
         if (rc == 0)
                 rc = +1;
 
@@ -5417,23 +5556,24 @@ static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
  */
 static const struct dt_index_operations osd_index_ea_ops = {
-        .dio_lookup         = osd_index_ea_lookup,
-        .dio_declare_insert = osd_index_declare_ea_insert,
-        .dio_insert         = osd_index_ea_insert,
-        .dio_declare_delete = osd_index_declare_ea_delete,
-        .dio_delete         = osd_index_ea_delete,
-        .dio_it     = {
-                .init     = osd_it_ea_init,
-                .fini     = osd_it_ea_fini,
-                .get      = osd_it_ea_get,
-                .put      = osd_it_ea_put,
-                .next     = osd_it_ea_next,
-                .key      = osd_it_ea_key,
-                .key_size = osd_it_ea_key_size,
-                .rec      = osd_it_ea_rec,
-                .store    = osd_it_ea_store,
-                .load     = osd_it_ea_load
-        }
+	.dio_lookup         = osd_index_ea_lookup,
+	.dio_declare_insert = osd_index_declare_ea_insert,
+	.dio_insert         = osd_index_ea_insert,
+	.dio_declare_delete = osd_index_declare_ea_delete,
+	.dio_delete         = osd_index_ea_delete,
+	.dio_it     = {
+		.init     = osd_it_ea_init,
+		.fini     = osd_it_ea_fini,
+		.get      = osd_it_ea_get,
+		.put      = osd_it_ea_put,
+		.next     = osd_it_ea_next,
+		.key      = osd_it_ea_key,
+		.key_size = osd_it_ea_key_size,
+		.rec      = osd_it_ea_rec,
+		.rec_size = osd_it_ea_rec_size,
+		.store    = osd_it_ea_store,
+		.load     = osd_it_ea_load
+	}
 };
 
 static void *osd_key_init(const struct lu_context *ctx,
@@ -5509,6 +5649,43 @@ static int osd_device_init(const struct lu_env *env, struct lu_device *d,
 	return osd_procfs_init(osd, name);
 }
 
+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;
+	ENTRY;
+
+	if (osd->od_is_ost || osd->od_cl_seq != NULL)
+		RETURN(0);
+
+	if (unlikely(ss == NULL))
+		RETURN(-ENODEV);
+
+	OBD_ALLOC_PTR(osd->od_cl_seq);
+	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);
+}
+
+static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
+{
+	if (osd->od_cl_seq == NULL)
+		return;
+
+	seq_client_fini(osd->od_cl_seq);
+	OBD_FREE_PTR(osd->od_cl_seq);
+	osd->od_cl_seq = NULL;
+}
+
 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
 {
 	ENTRY;
@@ -5519,6 +5696,8 @@ static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
 		o->od_quota_slave = NULL;
 	}
 
+	osd_fid_fini(env, o);
+
 	RETURN(0);
 }
 
@@ -5618,6 +5797,11 @@ static int osd_mount(const struct lu_env *env,
 	if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
 					 LDISKFS_FEATURE_INCOMPAT_DIRDATA))
 		LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
+	else if (!o->od_is_ost)
+		CWARN("%s: device %s was upgraded from Lustre-1.x without "
+		      "enabling the dirdata feature. If you do not want to "
+		      "downgrade to Lustre-1.x again, you can enable it via "
+		      "'tune2fs -O dirdata device'\n", name, dev);
 #endif
 	inode = osd_sb(o)->s_root->d_inode;
 	lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
@@ -5679,7 +5863,6 @@ static int osd_device_init0(const struct lu_env *env,
 
 	spin_lock_init(&o->od_osfs_lock);
 	mutex_init(&o->od_otable_mutex);
-	o->od_osfs_age = cfs_time_shift_64(-1000);
 
 	o->od_capa_hash = init_capa_hash();
 	if (o->od_capa_hash == NULL)
@@ -5689,20 +5872,20 @@ static int osd_device_init0(const struct lu_env *env,
 	o->od_writethrough_cache = 1;
 	o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
 
-	rc = osd_mount(env, o, cfg);
-	if (rc)
-		GOTO(out_capa, rc);
-
 	cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
 			sizeof(o->od_svname));
 	if (cplen >= sizeof(o->od_svname)) {
 		rc = -E2BIG;
-		GOTO(out_mnt, rc);
+		GOTO(out_capa, rc);
 	}
 
 	if (server_name_is_ost(o->od_svname))
 		o->od_is_ost = 1;
 
+	rc = osd_mount(env, o, cfg);
+	if (rc != 0)
+		GOTO(out_capa, rc);
+
 	rc = osd_obj_map_init(env, o);
 	if (rc != 0)
 		GOTO(out_mnt, rc);
@@ -5719,7 +5902,7 @@ static int osd_device_init0(const struct lu_env *env,
 	/* self-repair LMA by default */
 	o->od_lma_self_repair = 1;
 
-	CFS_INIT_LIST_HEAD(&o->od_ios_list);
+	INIT_LIST_HEAD(&o->od_ios_list);
 	/* setup scrub, including OI files initialization */
 	rc = osd_scrub_setup(env, o);
 	if (rc < 0)
@@ -5825,12 +6008,13 @@ static int osd_process_config(const struct lu_env *env,
 		break;
 	case LCFG_PARAM:
 		LASSERT(&o->od_dt_dev);
-		rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
-					      cfg, &o->od_dt_dev);
+		rc = class_process_proc_seq_param(PARAM_OSD,
+						  lprocfs_osd_obd_vars,
+						  cfg, &o->od_dt_dev);
 		if (rc > 0 || rc == -ENOSYS)
-			rc = class_process_proc_param(PARAM_OST,
-						      lprocfs_osd_obd_vars,
-						      cfg, &o->od_dt_dev);
+			rc = class_process_proc_seq_param(PARAM_OST,
+							  lprocfs_osd_obd_vars,
+							  cfg, &o->od_dt_dev);
 		break;
 	default:
 		rc = -ENOSYS;
@@ -5914,13 +6098,26 @@ static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
 	int		   result = 0;
 	ENTRY;
 
-	if (osd->od_quota_slave != NULL)
+	if (osd->od_quota_slave != NULL) {
 		/* set up quota slave objects */
 		result = qsd_prepare(env, osd->od_quota_slave);
+		if (result != 0)
+			RETURN(result);
+	}
+
+	result = osd_fid_init(env, osd);
 
 	RETURN(result);
 }
 
+int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
+		  struct lu_fid *fid, struct md_op_data *op_data)
+{
+	struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
+
+	return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
+}
+
 static const struct lu_object_operations osd_lu_obj_ops = {
         .loo_object_init      = osd_object_init,
         .loo_object_delete    = osd_object_delete,
@@ -5964,7 +6161,8 @@ struct lu_device_type osd_device_type = {
 static struct obd_ops osd_obd_device_ops = {
 	.o_owner = THIS_MODULE,
 	.o_connect	= osd_obd_connect,
-	.o_disconnect	= osd_obd_disconnect
+	.o_disconnect	= osd_obd_disconnect,
+	.o_fid_alloc	= osd_fid_alloc,
 };
 
 static int __init osd_mod_init(void)
@@ -5977,11 +6175,12 @@ static int __init osd_mod_init(void)
 	if (rc)
 		return rc;
 
-	rc = class_register_type(&osd_obd_device_ops, NULL, NULL,
+	rc = class_register_type(&osd_obd_device_ops, NULL, true,
+				 lprocfs_osd_module_vars,
 #ifndef HAVE_ONLY_PROCFS_SEQ
-				lprocfs_osd_module_vars,
+				 NULL,
 #endif
-				LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
+				 LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
 	if (rc)
 		lu_kmem_fini(ldiskfs_caches);
 	return rc;