X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Flfsck%2Flfsck_lib.c;h=db73616915f0a9bd98d70b61660888c503422ac1;hb=d0ba82e44038b2658bbd1a06d52641fda9fc1fe7;hp=c2f2b50eef74532ea55d1f95dfe150a86878a64a;hpb=e1a87d94e89c175ac8a432be5d6eec138e8915fc;p=fs%2Flustre-release.git

diff --git a/lustre/lfsck/lfsck_lib.c b/lustre/lfsck/lfsck_lib.c
index c2f2b50..db73616 100644
--- a/lustre/lfsck/lfsck_lib.c
+++ b/lustre/lfsck/lfsck_lib.c
@@ -92,6 +92,9 @@ const char *lfsck_param_names[] = {
 	"failout",
 	"dryrun",
 	"all_targets",
+	"broadcast",
+	"orphan",
+	"create_ostobj",
 	NULL
 };
 
@@ -287,7 +290,7 @@ __lfsck_component_find(struct lfsck_instance *lfsck, __u16 type, cfs_list_t *lis
 	return NULL;
 }
 
-static struct lfsck_component *
+struct lfsck_component *
 lfsck_component_find(struct lfsck_instance *lfsck, __u16 type)
 {
 	struct lfsck_component *com;
@@ -322,6 +325,470 @@ void lfsck_component_cleanup(const struct lu_env *env,
 	lfsck_component_put(env, com);
 }
 
+int lfsck_fid_alloc(const struct lu_env *env, struct lfsck_instance *lfsck,
+		    struct lu_fid *fid, bool locked)
+{
+	struct lfsck_bookmark	*bk = &lfsck->li_bookmark_ram;
+	int			 rc = 0;
+	ENTRY;
+
+	if (!locked)
+		mutex_lock(&lfsck->li_mutex);
+
+	rc = seq_client_alloc_fid(env, lfsck->li_seq, fid);
+	if (rc >= 0) {
+		bk->lb_last_fid = *fid;
+		/* We do not care about whether the subsequent sub-operations
+		 * failed or not. The worst case is that one FID is lost that
+		 * is not a big issue for the LFSCK since it is relative rare
+		 * for LFSCK create. */
+		rc = lfsck_bookmark_store(env, lfsck);
+	}
+
+	if (!locked)
+		mutex_unlock(&lfsck->li_mutex);
+
+	RETURN(rc);
+}
+
+static const char dot[] = ".";
+static const char dotdot[] = "..";
+
+static int lfsck_create_lpf_local(const struct lu_env *env,
+				  struct lfsck_instance *lfsck,
+				  struct dt_object *parent,
+				  struct dt_object *child,
+				  struct lu_attr *la,
+				  struct dt_object_format *dof,
+				  const char *name)
+{
+	struct dt_device	*dev	= lfsck->li_bottom;
+	struct lfsck_bookmark	*bk	= &lfsck->li_bookmark_ram;
+	struct dt_object	*bk_obj = lfsck->li_bookmark_obj;
+	const struct lu_fid	*cfid	= lu_object_fid(&child->do_lu);
+	struct thandle		*th	= NULL;
+	loff_t			 pos	= 0;
+	int			 len	= sizeof(struct lfsck_bookmark);
+	int			 rc	= 0;
+	ENTRY;
+
+	th = dt_trans_create(env, dev);
+	if (IS_ERR(th))
+		RETURN(PTR_ERR(th));
+
+	/* 1a. create child */
+	rc = dt_declare_create(env, child, la, NULL, dof, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 2a. increase child nlink */
+	rc = dt_declare_ref_add(env, child, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 3a. insert name into parent dir */
+	rc = dt_declare_insert(env, parent, (const struct dt_rec *)cfid,
+			       (const struct dt_key *)name, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 4a. increase parent nlink */
+	rc = dt_declare_ref_add(env, parent, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 5a. update bookmark */
+	rc = dt_declare_record_write(env, bk_obj,
+				     lfsck_buf_get(env, bk, len), 0, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	rc = dt_trans_start_local(env, dev, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	dt_write_lock(env, child, 0);
+	/* 1b.1 create child */
+	rc = dt_create(env, child, la, NULL, dof, th);
+	if (rc != 0)
+		GOTO(unlock, rc);
+
+	if (unlikely(!dt_try_as_dir(env, child)))
+		GOTO(unlock, rc = -ENOTDIR);
+
+	/* 1b.2 insert dot into child dir */
+	rc = dt_insert(env, child, (const struct dt_rec *)cfid,
+		       (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
+	if (rc != 0)
+		GOTO(unlock, rc);
+
+	/* 1b.3 insert dotdot into child dir */
+	rc = dt_insert(env, child, (const struct dt_rec *)&LU_LPF_FID,
+		       (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
+	if (rc != 0)
+		GOTO(unlock, rc);
+
+	/* 2b. increase child nlink */
+	rc = dt_ref_add(env, child, th);
+	dt_write_unlock(env, child);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 3b. insert name into parent dir */
+	rc = dt_insert(env, parent, (const struct dt_rec *)cfid,
+		       (const struct dt_key *)name, th, BYPASS_CAPA, 1);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	dt_write_lock(env, parent, 0);
+	/* 4b. increase parent nlink */
+	rc = dt_ref_add(env, parent, th);
+	dt_write_unlock(env, parent);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	bk->lb_lpf_fid = *cfid;
+	lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
+
+	/* 5b. update bookmark */
+	rc = dt_record_write(env, bk_obj,
+			     lfsck_buf_get(env, bk, len), &pos, th);
+
+	GOTO(stop, rc);
+
+unlock:
+	dt_write_unlock(env, child);
+
+stop:
+	dt_trans_stop(env, dev, th);
+
+	return rc;
+}
+
+static int lfsck_create_lpf_remote(const struct lu_env *env,
+				   struct lfsck_instance *lfsck,
+				   struct dt_object *parent,
+				   struct dt_object *child,
+				   struct lu_attr *la,
+				   struct dt_object_format *dof,
+				   const char *name)
+{
+	struct lfsck_bookmark	*bk	= &lfsck->li_bookmark_ram;
+	struct dt_object	*bk_obj = lfsck->li_bookmark_obj;
+	const struct lu_fid	*cfid	= lu_object_fid(&child->do_lu);
+	struct thandle		*th	= NULL;
+	struct dt_device	*dev;
+	loff_t			 pos	= 0;
+	int			 len	= sizeof(struct lfsck_bookmark);
+	int			 rc	= 0;
+	ENTRY;
+
+	/* Create .lustre/lost+found/MDTxxxx. */
+
+	/* XXX: Currently, cross-MDT create operation needs to create the child
+	 *	object firstly, then insert name into the parent directory. For
+	 *	this case, the child object resides on current MDT (local), but
+	 *	the parent ".lustre/lost+found" may be on remote MDT. It is not
+	 *	easy to contain all the sub-modifications orderly within single
+	 *	transaction.
+	 *
+	 *	To avoid more inconsistency, we split the create operation into
+	 *	two transactions:
+	 *
+	 *	1) create the child locally.
+	 *	2) insert the name "MDTXXXX" in the parent ".lustre/lost+found"
+	 *	   remotely and update the lfsck_bookmark::lb_lpf_fid locally.
+	 *
+	 *	If 1) done but 2) failed, then the worst case is that we lose
+	 *	one object locally, which is not a big issue. (can be repaird
+	 *	by LFSCK phase III) */
+
+	/* Transaction I: */
+
+	dev = lfsck->li_bottom;
+	th = dt_trans_create(env, dev);
+	if (IS_ERR(th))
+		RETURN(PTR_ERR(th));
+
+	/* 1a. create child locally. */
+	rc = dt_declare_create(env, child, la, NULL, dof, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 2a. increase child nlink locally. */
+	rc = dt_declare_ref_add(env, child, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	rc = dt_trans_start_local(env, dev, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	dt_write_lock(env, child, 0);
+	/* 1b. create child locally. */
+	rc = dt_create(env, child, la, NULL, dof, th);
+	if (rc != 0)
+		GOTO(unlock, rc);
+
+	if (unlikely(!dt_try_as_dir(env, child)))
+		GOTO(unlock, rc = -ENOTDIR);
+
+	/* 2b.1 insert dot into child dir locally. */
+	rc = dt_insert(env, child, (const struct dt_rec *)cfid,
+		       (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
+	if (rc != 0)
+		GOTO(unlock, rc);
+
+	/* 2b.2 insert dotdot into child dir locally. */
+	rc = dt_insert(env, child, (const struct dt_rec *)&LU_LPF_FID,
+		       (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
+	if (rc != 0)
+		GOTO(unlock, rc);
+
+	/* 2b.3 increase child nlink locally. */
+	rc = dt_ref_add(env, child, th);
+	dt_write_unlock(env, child);
+	dt_trans_stop(env, dev, th);
+	if (rc != 0)
+		RETURN(rc);
+
+	/* Transaction II: */
+
+	dev = lfsck->li_next;
+	th = dt_trans_create(env, dev);
+	if (IS_ERR(th))
+		RETURN(PTR_ERR(th));
+
+	/* 3a. insert name into parent dir remotely. */
+	rc = dt_declare_insert(env, parent, (const struct dt_rec *)cfid,
+			       (const struct dt_key *)name, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 4a. increase parent nlink remotely. */
+	rc = dt_declare_ref_add(env, parent, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 5a. decrease child nlink for dotdot locally if former remote
+	 *     update failed. */
+	rc = dt_declare_ref_del(env, child, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 6a. decrease child nlink for dot locally if former remote
+	 *     update failed. */
+	rc = dt_declare_ref_del(env, child, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 7a. destroy child locally if former remote update failed. */
+	rc = dt_declare_destroy(env, child, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 8a. update bookmark locally. */
+	rc = dt_declare_record_write(env, bk_obj,
+				     lfsck_buf_get(env, bk, len), 0, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	rc = dt_trans_start(env, dev, th);
+	if (rc != 0)
+		GOTO(stop, rc);
+
+	/* 3b. insert name into parent dir remotely. */
+	rc = dt_insert(env, parent, (const struct dt_rec *)cfid,
+		       (const struct dt_key *)name, th, BYPASS_CAPA, 1);
+	if (rc == 0) {
+		dt_write_lock(env, parent, 0);
+		/* 4b. increase parent nlink remotely. */
+		rc = dt_ref_add(env, parent, th);
+		dt_write_unlock(env, parent);
+	}
+	if (rc != 0) {
+		/* 5b. decrease child nlink for dotdot locally. */
+		dt_ref_del(env, child, th);
+		/* 6b. decrease child nlink for dot locally. */
+		dt_ref_del(env, child, th);
+		/* 7b. destroy child locally. */
+		dt_destroy(env, child, th);
+		GOTO(stop, rc);
+	}
+
+	bk->lb_lpf_fid = *cfid;
+	lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
+
+	/* 8b. update bookmark locally. */
+	rc = dt_record_write(env, bk_obj,
+			     lfsck_buf_get(env, bk, len), &pos, th);
+
+	GOTO(stop, rc);
+
+unlock:
+	dt_write_unlock(env, child);
+stop:
+	dt_trans_stop(env, dev, th);
+
+	return rc;
+}
+
+/* Do NOT create .lustre/lost+found/MDTxxxx when register the lfsck instance,
+ * because the MDT0 maybe not reaady for sequence allocation yet. We do that
+ * only when it is required, such as orphan OST-objects repairing. */
+int lfsck_create_lpf(const struct lu_env *env, struct lfsck_instance *lfsck)
+{
+	struct lfsck_bookmark	 *bk	= &lfsck->li_bookmark_ram;
+	struct lfsck_thread_info *info	= lfsck_env_info(env);
+	struct lu_fid		 *cfid	= &info->lti_fid2;
+	struct lu_attr		 *la	= &info->lti_la;
+	struct dt_object_format  *dof	= &info->lti_dof;
+	struct dt_object	 *parent = NULL;
+	struct dt_object	 *child	= NULL;
+	char			  name[8];
+	int			  node	= lfsck_dev_idx(lfsck->li_bottom);
+	int			  rc	= 0;
+	ENTRY;
+
+	LASSERT(lfsck->li_master);
+
+	sprintf(name, "MDT%04x", node);
+	if (node == 0) {
+		parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
+						  &LU_LPF_FID);
+	} else {
+		struct lfsck_tgt_desc *ltd;
+
+		ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, 0);
+		if (unlikely(ltd == NULL))
+			RETURN(-ENXIO);
+
+		parent = lfsck_object_find_by_dev(env, ltd->ltd_tgt,
+						  &LU_LPF_FID);
+		lfsck_tgt_put(ltd);
+	}
+	if (IS_ERR(parent))
+		RETURN(PTR_ERR(parent));
+
+	if (unlikely(!dt_try_as_dir(env, parent)))
+		GOTO(out, rc = -ENOTDIR);
+
+	mutex_lock(&lfsck->li_mutex);
+	if (lfsck->li_lpf_obj != NULL)
+		GOTO(unlock, rc = 0);
+
+	if (fid_is_zero(&bk->lb_lpf_fid)) {
+		/* There is corner case that: in former LFSCK scanning we have
+		 * created the .lustre/lost+found/MDTxxxx but failed to update
+		 * the lfsck_bookmark::lb_lpf_fid successfully. So need lookup
+		 * it from MDT0 firstly. */
+		rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
+			       (const struct dt_key *)name, BYPASS_CAPA);
+		if (rc != 0 && rc != -ENOENT)
+			GOTO(unlock, rc);
+
+		if (rc == 0) {
+			bk->lb_lpf_fid = *cfid;
+			rc = lfsck_bookmark_store(env, lfsck);
+		} else {
+			rc = lfsck_fid_alloc(env, lfsck, cfid, true);
+		}
+		if (rc != 0)
+			GOTO(unlock, rc);
+	} else {
+		*cfid = bk->lb_lpf_fid;
+	}
+
+	child = lfsck_object_find_by_dev(env, lfsck->li_bottom, cfid);
+	if (IS_ERR(child))
+		GOTO(unlock, rc = PTR_ERR(child));
+
+	if (dt_object_exists(child) != 0) {
+		if (unlikely(!dt_try_as_dir(env, child)))
+			GOTO(unlock, rc = -ENOTDIR);
+
+		lfsck->li_lpf_obj = child;
+		GOTO(unlock, rc = 0);
+	}
+
+	memset(la, 0, sizeof(*la));
+	la->la_atime = la->la_mtime = la->la_ctime = cfs_time_current_sec();
+	la->la_mode = S_IFDIR | S_IRWXU;
+	la->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
+		       LA_UID | LA_GID;
+	memset(dof, 0, sizeof(*dof));
+	dof->dof_type = dt_mode_to_dft(S_IFDIR);
+
+	if (node == 0)
+		rc = lfsck_create_lpf_local(env, lfsck, parent, child, la,
+					    dof, name);
+	else
+		rc = lfsck_create_lpf_remote(env, lfsck, parent, child, la,
+					     dof, name);
+	if (rc == 0)
+		lfsck->li_lpf_obj = child;
+
+	GOTO(unlock, rc);
+
+unlock:
+	mutex_unlock(&lfsck->li_mutex);
+	if (rc != 0 && child != NULL && !IS_ERR(child))
+		lu_object_put(env, &child->do_lu);
+out:
+	if (parent != NULL && !IS_ERR(parent))
+		lu_object_put(env, &parent->do_lu);
+
+	return rc;
+}
+
+static int lfsck_fid_init(struct lfsck_instance *lfsck)
+{
+	struct lfsck_bookmark	*bk	= &lfsck->li_bookmark_ram;
+	struct seq_server_site	*ss;
+	char			*prefix;
+	int			 rc	= 0;
+	ENTRY;
+
+	ss = lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
+	if (unlikely(ss == NULL))
+		RETURN(-ENXIO);
+
+	OBD_ALLOC_PTR(lfsck->li_seq);
+	if (lfsck->li_seq == NULL)
+		RETURN(-ENOMEM);
+
+	OBD_ALLOC(prefix, MAX_OBD_NAME + 7);
+	if (prefix == NULL)
+		GOTO(out, rc = -ENOMEM);
+
+	snprintf(prefix, MAX_OBD_NAME + 7, "lfsck-%s", lfsck_lfsck2name(lfsck));
+	rc = seq_client_init(lfsck->li_seq, NULL, LUSTRE_SEQ_METADATA, prefix,
+			     ss->ss_server_seq);
+	OBD_FREE(prefix, MAX_OBD_NAME + 7);
+	if (rc != 0)
+		GOTO(out, rc);
+
+	if (fid_is_sane(&bk->lb_last_fid))
+		lfsck->li_seq->lcs_fid = bk->lb_last_fid;
+
+	RETURN(0);
+
+out:
+	OBD_FREE_PTR(lfsck->li_seq);
+	lfsck->li_seq = NULL;
+
+	return rc;
+}
+
+static void lfsck_fid_fini(struct lfsck_instance *lfsck)
+{
+	if (lfsck->li_seq != NULL) {
+		seq_client_fini(lfsck->li_seq);
+		OBD_FREE_PTR(lfsck->li_seq);
+		lfsck->li_seq = NULL;
+	}
+}
+
 void lfsck_instance_cleanup(const struct lu_env *env,
 			    struct lfsck_instance *lfsck)
 {
@@ -370,11 +837,18 @@ void lfsck_instance_cleanup(const struct lu_env *env,
 		lfsck->li_bookmark_obj = NULL;
 	}
 
+	if (lfsck->li_lpf_obj != NULL) {
+		lu_object_put(env, &lfsck->li_lpf_obj->do_lu);
+		lfsck->li_lpf_obj = NULL;
+	}
+
 	if (lfsck->li_los != NULL) {
 		local_oid_storage_fini(env, lfsck->li_los);
 		lfsck->li_los = NULL;
 	}
 
+	lfsck_fid_fini(lfsck);
+
 	OBD_FREE_PTR(lfsck);
 }
 
@@ -397,8 +871,8 @@ __lfsck_instance_find(struct dt_device *key, bool ref, bool unlink)
 	return NULL;
 }
 
-static inline struct lfsck_instance *lfsck_instance_find(struct dt_device *key,
-							 bool ref, bool unlink)
+struct lfsck_instance *lfsck_instance_find(struct dt_device *key, bool ref,
+					   bool unlink)
 {
 	struct lfsck_instance *lfsck;
 
@@ -433,8 +907,9 @@ int lfsck_bits_dump(char **buf, int *len, int bits, const char *names[],
 	int flag;
 	int rc;
 	int i;
+	bool newline = (bits != 0 ? false : true);
 
-	rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
+	rc = snprintf(*buf, *len, "%s:%c", prefix, newline ? '\n' : ' ');
 	if (rc <= 0)
 		return -ENOSPC;
 
@@ -444,8 +919,11 @@ int lfsck_bits_dump(char **buf, int *len, int bits, const char *names[],
 		if (flag & bits) {
 			bits &= ~flag;
 			if (names[i] != NULL) {
+				if (bits == 0)
+					newline = true;
+
 				rc = snprintf(*buf, *len, "%s%c", names[i],
-					      bits != 0 ? ',' : '\n');
+					      newline ? '\n' : ',');
 				if (rc <= 0)
 					return -ENOSPC;
 
@@ -454,6 +932,16 @@ int lfsck_bits_dump(char **buf, int *len, int bits, const char *names[],
 			}
 		}
 	}
+
+	if (!newline) {
+		rc = snprintf(*buf, *len, "\n");
+		if (rc <= 0)
+			return -ENOSPC;
+
+		*buf += rc;
+		*len -= rc;
+	}
+
 	return save - *len;
 }
 
@@ -533,9 +1021,10 @@ void lfsck_pos_fill(const struct lu_env *env, struct lfsck_instance *lfsck,
 	}
 }
 
-static void __lfsck_set_speed(struct lfsck_instance *lfsck, __u32 limit)
+bool __lfsck_set_speed(struct lfsck_instance *lfsck, __u32 limit)
 {
-	lfsck->li_bookmark_ram.lb_speed_limit = limit;
+	bool dirty = false;
+
 	if (limit != LFSCK_SPEED_NO_LIMIT) {
 		if (limit > HZ) {
 			lfsck->li_sleep_rate = limit / HZ;
@@ -548,6 +1037,13 @@ static void __lfsck_set_speed(struct lfsck_instance *lfsck, __u32 limit)
 		lfsck->li_sleep_jif = 0;
 		lfsck->li_sleep_rate = 0;
 	}
+
+	if (lfsck->li_bookmark_ram.lb_speed_limit != limit) {
+		lfsck->li_bookmark_ram.lb_speed_limit = limit;
+		dirty = true;
+	}
+
+	return dirty;
 }
 
 void lfsck_control_speed(struct lfsck_instance *lfsck)
@@ -620,8 +1116,8 @@ static int lfsck_needs_scan_dir(const struct lu_env *env,
 			return 1;
 		}
 
-		/* .lustre doesn't contain "real" user objects, no need lfsck */
-		if (fid_is_dot_lustre(lfsck_dto2fid(obj))) {
+		/* No need to check .lustre and its children. */
+		if (fid_seq_is_dot_lustre(fid_seq(lfsck_dto2fid(obj)))) {
 			if (depth > 0)
 				lfsck_object_put(env, obj);
 			return 0;
@@ -671,10 +1167,16 @@ static int lfsck_needs_scan_dir(const struct lu_env *env,
 			return 0;
 		}
 
-		/* Currently, only client visible directory can be remote. */
 		if (dt_object_remote(obj)) {
+			/* .lustre/lost+found/MDTxxx can be remote directory. */
+			if (fid_seq_is_dot_lustre(fid_seq(lfsck_dto2fid(obj))))
+				rc = 0;
+			else
+				/* Other remote directory should be client
+				 * visible and need to be checked. */
+				rc = 1;
 			lfsck_object_put(env, obj);
-			return 1;
+			return rc;
 		}
 
 		depth++;
@@ -973,6 +1475,7 @@ static void lfsck_interpret(const struct lu_env *env,
 	struct lfsck_async_interpret_args *laia = args;
 	struct lfsck_component		  *com;
 
+	LASSERT(laia->laia_com == NULL);
 	LASSERT(laia->laia_shared);
 
 	spin_lock(&lfsck->li_lock);
@@ -1000,8 +1503,7 @@ int lfsck_double_scan(const struct lu_env *env, struct lfsck_instance *lfsck)
 	int			rc  = 0;
 	int			rc1 = 0;
 
-	cfs_list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
-				     lc_link) {
+	list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
 		if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
 			com->lc_journal = 0;
 
@@ -1014,6 +1516,17 @@ int lfsck_double_scan(const struct lu_env *env, struct lfsck_instance *lfsck)
 		     atomic_read(&lfsck->li_double_scan_count) == 0,
 		     &lwi);
 
+	if (lfsck->li_status != LS_PAUSED &&
+	    lfsck->li_status != LS_CO_PAUSED) {
+		list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
+					 lc_link) {
+			spin_lock(&lfsck->li_lock);
+			list_del_init(&com->lc_link);
+			list_add_tail(&com->lc_link, &lfsck->li_list_idle);
+			spin_unlock(&lfsck->li_lock);
+		}
+	}
+
 	return rc1 != 0 ? rc1 : rc;
 }
 
@@ -1068,12 +1581,23 @@ void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck)
 				 lc_link) {
 		if (com->lc_ops->lfsck_quit != NULL)
 			com->lc_ops->lfsck_quit(env, com);
+
+		spin_lock(&lfsck->li_lock);
+		list_del_init(&com->lc_link);
+		list_del_init(&com->lc_link_dir);
+		list_add_tail(&com->lc_link, &lfsck->li_list_idle);
+		spin_unlock(&lfsck->li_lock);
 	}
 
 	list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
 				 lc_link) {
 		if (com->lc_ops->lfsck_quit != NULL)
 			com->lc_ops->lfsck_quit(env, com);
+
+		spin_lock(&lfsck->li_lock);
+		list_del_init(&com->lc_link);
+		list_add_tail(&com->lc_link, &lfsck->li_list_idle);
+		spin_unlock(&lfsck->li_lock);
 	}
 }
 
@@ -1106,9 +1630,6 @@ int lfsck_async_request(const struct lu_env *env, struct obd_export *exp,
 	struct req_format		  *format;
 	int				   rc;
 
-	if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
-		return -EOPNOTSUPP;
-
 	switch (request) {
 	case LFSCK_NOTIFY:
 		format = &RQF_LFSCK_NOTIFY;
@@ -1166,7 +1687,7 @@ int lfsck_get_speed(struct dt_device *key, void *buf, int len)
 			      lfsck->li_bookmark_ram.lb_speed_limit);
 		lfsck_instance_put(&env, lfsck);
 	} else {
-		rc = -ENODEV;
+		rc = -ENXIO;
 	}
 
 	lu_env_fini(&env);
@@ -1189,12 +1710,12 @@ int lfsck_set_speed(struct dt_device *key, int val)
 	lfsck = lfsck_instance_find(key, true, false);
 	if (likely(lfsck != NULL)) {
 		mutex_lock(&lfsck->li_mutex);
-		__lfsck_set_speed(lfsck, val);
-		rc = lfsck_bookmark_store(&env, lfsck);
+		if (__lfsck_set_speed(lfsck, val))
+			rc = lfsck_bookmark_store(&env, lfsck);
 		mutex_unlock(&lfsck->li_mutex);
 		lfsck_instance_put(&env, lfsck);
 	} else {
-		rc = -ENODEV;
+		rc = -ENXIO;
 	}
 
 	lu_env_fini(&env);
@@ -1220,7 +1741,7 @@ int lfsck_get_windows(struct dt_device *key, void *buf, int len)
 			      lfsck->li_bookmark_ram.lb_async_windows);
 		lfsck_instance_put(&env, lfsck);
 	} else {
-		rc = -ENODEV;
+		rc = -ENXIO;
 	}
 
 	lu_env_fini(&env);
@@ -1243,12 +1764,12 @@ int lfsck_set_windows(struct dt_device *key, int val)
 	lfsck = lfsck_instance_find(key, true, false);
 	if (likely(lfsck != NULL)) {
 		if (val > LFSCK_ASYNC_WIN_MAX) {
-			CERROR("%s: Too large async windows size, which "
-			       "may cause memory issues. The valid range "
-			       "is [0 - %u]. If you do not want to restrict "
-			       "the windows size for async requests pipeline, "
-			       "just set it as 0.\n",
-			       lfsck_lfsck2name(lfsck), LFSCK_ASYNC_WIN_MAX);
+			CWARN("%s: Too large async window size, which "
+			      "may cause memory issues. The valid range "
+			      "is [0 - %u]. If you do not want to restrict "
+			      "the window size for async requests pipeline, "
+			      "just set it as 0.\n",
+			      lfsck_lfsck2name(lfsck), LFSCK_ASYNC_WIN_MAX);
 			rc = -EINVAL;
 		} else if (lfsck->li_bookmark_ram.lb_async_windows != val) {
 			mutex_lock(&lfsck->li_mutex);
@@ -1258,7 +1779,7 @@ int lfsck_set_windows(struct dt_device *key, int val)
 		}
 		lfsck_instance_put(&env, lfsck);
 	} else {
-		rc = -ENODEV;
+		rc = -ENXIO;
 	}
 
 	lu_env_fini(&env);
@@ -1291,7 +1812,7 @@ int lfsck_dump(struct dt_device *key, void *buf, int len, enum lfsck_type type)
 
 		lfsck_instance_put(&env, lfsck);
 	} else {
-		rc = -ENODEV;
+		rc = -ENXIO;
 	}
 
 	lu_env_fini(&env);
@@ -1499,8 +2020,8 @@ int lfsck_start(const struct lu_env *env, struct dt_device *key,
 	struct lfsck_component		*com;
 	struct l_wait_info		 lwi    = { 0 };
 	struct lfsck_thread_args	*lta;
-	bool				 dirty  = false;
-	long				 rc     = 0;
+	struct task_struct		*task;
+	int				 rc     = 0;
 	__u16				 valid  = 0;
 	__u16				 flags  = 0;
 	__u16				 type   = 1;
@@ -1508,7 +2029,7 @@ int lfsck_start(const struct lu_env *env, struct dt_device *key,
 
 	lfsck = lfsck_instance_find(key, true, false);
 	if (unlikely(lfsck == NULL))
-		RETURN(-ENODEV);
+		RETURN(-ENXIO);
 
 	/* System is not ready, try again later. */
 	if (unlikely(lfsck->li_namespace == NULL))
@@ -1526,6 +2047,11 @@ int lfsck_start(const struct lu_env *env, struct dt_device *key,
 	spin_lock(&lfsck->li_lock);
 	if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
 		rc = -EALREADY;
+		if (unlikely(start == NULL)) {
+			spin_unlock(&lfsck->li_lock);
+			GOTO(out, rc);
+		}
+
 		while (start->ls_active != 0) {
 			if (!(type & start->ls_active)) {
 				type <<= 1;
@@ -1573,78 +2099,6 @@ int lfsck_start(const struct lu_env *env, struct dt_device *key,
 	}
 
 	start->ls_version = bk->lb_version;
-	if (start->ls_valid & LSV_SPEED_LIMIT) {
-		__lfsck_set_speed(lfsck, start->ls_speed_limit);
-		dirty = true;
-	}
-
-	if (start->ls_valid & LSV_ASYNC_WINDOWS &&
-	    bk->lb_async_windows != start->ls_async_windows) {
-		bk->lb_async_windows = start->ls_async_windows;
-		dirty = true;
-	}
-
-	if (start->ls_valid & LSV_ERROR_HANDLE) {
-		valid |= DOIV_ERROR_HANDLE;
-		if (start->ls_flags & LPF_FAILOUT)
-			flags |= DOIF_FAILOUT;
-
-		if ((start->ls_flags & LPF_FAILOUT) &&
-		    !(bk->lb_param & LPF_FAILOUT)) {
-			bk->lb_param |= LPF_FAILOUT;
-			dirty = true;
-		} else if (!(start->ls_flags & LPF_FAILOUT) &&
-			   (bk->lb_param & LPF_FAILOUT)) {
-			bk->lb_param &= ~LPF_FAILOUT;
-			dirty = true;
-		}
-	}
-
-	if (start->ls_valid & LSV_DRYRUN) {
-		valid |= DOIV_DRYRUN;
-		if (start->ls_flags & LPF_DRYRUN)
-			flags |= DOIF_DRYRUN;
-
-		if ((start->ls_flags & LPF_DRYRUN) &&
-		    !(bk->lb_param & LPF_DRYRUN)) {
-			bk->lb_param |= LPF_DRYRUN;
-			dirty = true;
-		} else if (!(start->ls_flags & LPF_DRYRUN) &&
-			   (bk->lb_param & LPF_DRYRUN)) {
-			bk->lb_param &= ~LPF_DRYRUN;
-			lfsck->li_drop_dryrun = 1;
-			dirty = true;
-		}
-	}
-
-	if (bk->lb_param & LPF_ALL_TGT &&
-	    !(start->ls_flags & LPF_ALL_TGT)) {
-		bk->lb_param &= ~LPF_ALL_TGT;
-		dirty = true;
-	} else if (!(bk->lb_param & LPF_ALL_TGT) &&
-		   start->ls_flags & LPF_ALL_TGT) {
-		bk->lb_param |= LPF_ALL_TGT;
-		dirty = true;
-	}
-
-	if (bk->lb_param & LPF_ORPHAN &&
-	    !(start->ls_flags & LPF_ORPHAN)) {
-		bk->lb_param &= ~LPF_ORPHAN;
-		dirty = true;
-	} else if (!(bk->lb_param & LPF_ORPHAN) &&
-		   start->ls_flags & LPF_ORPHAN) {
-		bk->lb_param |= LPF_ORPHAN;
-		dirty = true;
-	}
-
-	if (dirty) {
-		rc = lfsck_bookmark_store(env, lfsck);
-		if (rc != 0)
-			GOTO(out, rc);
-	}
-
-	if (start->ls_flags & LPF_RESET)
-		flags |= DOIF_RESET;
 
 	if (start->ls_active != 0) {
 		struct lfsck_component *next;
@@ -1657,8 +2111,8 @@ int lfsck_start(const struct lu_env *env, struct dt_device *key,
 			GOTO(out, rc = -ENOTSUPP);
 		}
 
-		cfs_list_for_each_entry_safe(com, next,
-					     &lfsck->li_list_scan, lc_link) {
+		list_for_each_entry_safe(com, next,
+					 &lfsck->li_list_scan, lc_link) {
 			if (!(com->lc_type & start->ls_active)) {
 				rc = com->lc_ops->lfsck_post(env, com, 0,
 							     false);
@@ -1675,9 +2129,9 @@ int lfsck_start(const struct lu_env *env, struct dt_device *key,
 					/* The component status will be updated
 					 * when its prep() is called later by
 					 * the LFSCK main engine. */
-					cfs_list_del_init(&com->lc_link);
-					cfs_list_add_tail(&com->lc_link,
-							  &lfsck->li_list_scan);
+					list_del_init(&com->lc_link);
+					list_add_tail(&com->lc_link,
+						      &lfsck->li_list_scan);
 				}
 				start->ls_active &= ~type;
 			}
@@ -1685,7 +2139,28 @@ int lfsck_start(const struct lu_env *env, struct dt_device *key,
 		}
 	}
 
-	cfs_list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
+	if (list_empty(&lfsck->li_list_scan)) {
+		/* The speed limit will be used to control both the LFSCK and
+		 * low layer scrub (if applied), need to be handled firstly. */
+		if (start->ls_valid & LSV_SPEED_LIMIT) {
+			if (__lfsck_set_speed(lfsck, start->ls_speed_limit)) {
+				rc = lfsck_bookmark_store(env, lfsck);
+				if (rc != 0)
+					GOTO(out, rc);
+			}
+		}
+
+		goto trigger;
+	}
+
+	if (start->ls_flags & LPF_RESET)
+		flags |= DOIF_RESET;
+
+	rc = lfsck_set_param(env, lfsck, start, !!(flags & DOIF_RESET));
+	if (rc != 0)
+		GOTO(out, rc);
+
+	list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
 		start->ls_active |= com->lc_type;
 		if (flags & DOIF_RESET) {
 			rc = com->lc_ops->lfsck_reset(env, com, false);
@@ -1696,18 +2171,22 @@ int lfsck_start(const struct lu_env *env, struct dt_device *key,
 
 trigger:
 	lfsck->li_args_dir = LUDA_64BITHASH | LUDA_VERIFY;
-	if (bk->lb_param & LPF_DRYRUN) {
+	if (bk->lb_param & LPF_DRYRUN)
 		lfsck->li_args_dir |= LUDA_VERIFY_DRYRUN;
-		valid |= DOIV_DRYRUN;
-		flags |= DOIF_DRYRUN;
-	}
 
-	if (bk->lb_param & LPF_FAILOUT) {
+	if (start != NULL && start->ls_valid & LSV_ERROR_HANDLE) {
 		valid |= DOIV_ERROR_HANDLE;
-		flags |= DOIF_FAILOUT;
+		if (start->ls_flags & LPF_FAILOUT)
+			flags |= DOIF_FAILOUT;
 	}
 
-	if (!cfs_list_empty(&lfsck->li_list_scan))
+	if (start != NULL && start->ls_valid & LSV_DRYRUN) {
+		valid |= DOIV_DRYRUN;
+		if (start->ls_flags & LPF_DRYRUN)
+			flags |= DOIF_DRYRUN;
+	}
+
+	if (!list_empty(&lfsck->li_list_scan))
 		flags |= DOIF_OUTUSED;
 
 	lfsck->li_args_oit = (flags << DT_OTABLE_IT_FLAGS_SHIFT) | valid;
@@ -1716,9 +2195,11 @@ trigger:
 	if (IS_ERR(lta))
 		GOTO(out, rc = PTR_ERR(lta));
 
-	rc = PTR_ERR(kthread_run(lfsck_master_engine, lta, "lfsck"));
-	if (IS_ERR_VALUE(rc)) {
-		CERROR("%s: cannot start LFSCK thread: rc = %ld\n",
+	__lfsck_set_speed(lfsck, bk->lb_speed_limit);
+	task = kthread_run(lfsck_master_engine, lta, "lfsck");
+	if (IS_ERR(task)) {
+		rc = PTR_ERR(task);
+		CERROR("%s: cannot start LFSCK thread: rc = %d\n",
 		       lfsck_lfsck2name(lfsck), rc);
 		lfsck_thread_args_fini(lta);
 
@@ -1784,7 +2265,7 @@ int lfsck_stop(const struct lu_env *env, struct dt_device *key,
 
 	lfsck = lfsck_instance_find(key, true, false);
 	if (unlikely(lfsck == NULL))
-		RETURN(-ENODEV);
+		RETURN(-ENXIO);
 
 	thread = &lfsck->li_thread;
 	/* release lfsck::li_mutex to avoid deadlock. */
@@ -1869,13 +2350,15 @@ int lfsck_in_notify(const struct lu_env *env, struct dt_device *key,
 	case LE_PHASE1_DONE:
 	case LE_PHASE2_DONE:
 	case LE_FID_ACCESSED:
-	case LE_PEER_EXIT: {
+	case LE_PEER_EXIT:
+	case LE_CONDITIONAL_DESTROY:
+	case LE_PAIRS_VERIFY: {
 		struct lfsck_instance  *lfsck;
 		struct lfsck_component *com;
 
 		lfsck = lfsck_instance_find(key, true, false);
 		if (unlikely(lfsck == NULL))
-			RETURN(-ENODEV);
+			RETURN(-ENXIO);
 
 		com = lfsck_component_find(lfsck, lr->lr_active);
 		if (likely(com != NULL)) {
@@ -1904,7 +2387,7 @@ int lfsck_query(const struct lu_env *env, struct dt_device *key,
 
 	lfsck = lfsck_instance_find(key, true, false);
 	if (unlikely(lfsck == NULL))
-		RETURN(-ENODEV);
+		RETURN(-ENXIO);
 
 	com = lfsck_component_find(lfsck, lr->lr_active);
 	if (likely(com != NULL)) {
@@ -1924,7 +2407,7 @@ int lfsck_register_namespace(const struct lu_env *env, struct dt_device *key,
 			     struct ldlm_namespace *ns)
 {
 	struct lfsck_instance  *lfsck;
-	int			rc	= -ENODEV;
+	int			rc	= -ENXIO;
 
 	lfsck = lfsck_instance_find(key, true, false);
 	if (likely(lfsck != NULL)) {
@@ -2031,6 +2514,10 @@ int lfsck_register(const struct lu_env *env, struct dt_device *key,
 		GOTO(out, rc);
 
 	if (master) {
+		rc = lfsck_fid_init(lfsck);
+		if (rc < 0)
+			GOTO(out, rc);
+
 		rc = lfsck_namespace_setup(env, lfsck);
 		if (rc < 0)
 			GOTO(out, rc);