X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Flfsck%2Flfsck_engine.c;h=74aa59bb48587292c95d097aa807f366fdbe2685;hb=d00c924149b5dd042e6fee09ef724013df1b4e14;hp=322c122805b5dadbbf47bd2679cf56f314ec961f;hpb=2e3875738171cfc3525d4f011b16fefffaa74a79;p=fs%2Flustre-release.git

diff --git a/lustre/lfsck/lfsck_engine.c b/lustre/lfsck/lfsck_engine.c
index 322c122..74aa59b 100644
--- a/lustre/lfsck/lfsck_engine.c
+++ b/lustre/lfsck/lfsck_engine.c
@@ -221,7 +221,7 @@ static int lfsck_needs_scan_dir(const struct lu_env *env,
 		fld_range_set_mdt(range);
 		rc = fld_local_lookup(env, ss->ss_server_fld,
 				      fid_seq(fid), range);
-		if (rc != 0 || range->lsr_index != idx) {
+		if (rc != 0 || range->lsr_index != idx)
 			/* Current FID should NOT be for the input parameter
 			 * @obj, because the lfsck_master_oit_engine() has
 			 * filtered out agent object. So current FID is for
@@ -229,10 +229,7 @@ static int lfsck_needs_scan_dir(const struct lu_env *env,
 			 * So the ancestor is a remote directory. The input
 			 * parameter @obj is local directory, and should be
 			 * scanned under such case. */
-			LASSERT(depth > 0);
-
 			return 1;
-		}
 
 		/* normal FID on this target (locally) must be for the
 		 * client-side visiable object. */
@@ -368,9 +365,8 @@ static void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck,
 	}
 }
 
-static void lfsck_close_dir(const struct lu_env *env,
-			    struct lfsck_instance *lfsck,
-			    int result)
+void lfsck_close_dir(const struct lu_env *env,
+		     struct lfsck_instance *lfsck, int result)
 {
 	struct lfsck_component *com;
 	ENTRY;
@@ -389,10 +385,12 @@ static void lfsck_close_dir(const struct lu_env *env,
 	}
 
 	if (lfsck->li_di_dir != NULL) {
-		const struct dt_it_ops	*dir_iops =
-				&lfsck->li_obj_dir->do_index_ops->dio_it;
+		const struct dt_it_ops	*dir_iops;
 		struct dt_it		*dir_di   = lfsck->li_di_dir;
 
+		LASSERT(lfsck->li_obj_dir != NULL);
+
+		dir_iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
 		lfsck_di_dir_put(env, lfsck);
 		dir_iops->fini(env, dir_di);
 	}
@@ -407,8 +405,8 @@ static void lfsck_close_dir(const struct lu_env *env,
 	EXIT;
 }
 
-static int lfsck_open_dir(const struct lu_env *env,
-			  struct lfsck_instance *lfsck, __u64 cookie)
+int lfsck_open_dir(const struct lu_env *env,
+		   struct lfsck_instance *lfsck, __u64 cookie)
 {
 	struct dt_object	*obj	= lfsck->li_obj_dir;
 	struct dt_it		*di	= lfsck->li_di_dir;
@@ -550,6 +548,9 @@ static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
 			pos->lp_dir_cookie = 0;
 
 		rc = lfsck_open_dir(env, lfsck, pos->lp_dir_cookie);
+		if (rc > 0)
+			/* The end of the directory. */
+			rc = 0;
 	}
 
 	GOTO(out, rc);
@@ -633,20 +634,46 @@ static int lfsck_exec_dir(const struct lu_env *env,
 	return 0;
 }
 
+static int lfsck_master_dir_engine(const struct lu_env *env,
+				   struct lfsck_instance *lfsck);
+
 static int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
 		      int result)
 {
 	struct lfsck_component *com;
 	struct lfsck_component *next;
-	int			rc  = 0;
-	int			rc1 = 0;
+	int			rc  = result;
 
 	lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
 	lfsck_close_dir(env, lfsck, result);
+
+	while (thread_is_running(&lfsck->li_thread) && rc > 0 &&
+	       !list_empty(&lfsck->li_list_lmv)) {
+		struct lfsck_lmv_unit *llu;
+
+		spin_lock(&lfsck->li_lock);
+		llu = list_entry(lfsck->li_list_lmv.next,
+				 struct lfsck_lmv_unit, llu_link);
+		list_del_init(&llu->llu_link);
+		spin_unlock(&lfsck->li_lock);
+
+		lfsck->li_lmv = &llu->llu_lmv;
+		lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
+		rc = lfsck_open_dir(env, lfsck, 0);
+		if (rc == 0) {
+			rc = lfsck_master_dir_engine(env, lfsck);
+			lfsck_close_dir(env, lfsck, result);
+		}
+	}
+
+	result = rc;
+
 	list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
 		rc = com->lc_ops->lfsck_post(env, com, result, false);
 		if (rc != 0)
-			rc1 = rc;
+			CDEBUG(D_LFSCK, "%s: lfsck_post at the component %u: "
+			       "rc = %d\n", lfsck_lfsck2name(lfsck),
+			       (__u32)com->lc_type, rc);
 	}
 
 	lfsck->li_time_last_checkpoint = cfs_time_current();
@@ -893,6 +920,26 @@ static int lfsck_master_oit_engine(const struct lu_env *env,
 			RETURN(0);
 
 		lfsck->li_current_oit_processed = 1;
+
+		if (!list_empty(&lfsck->li_list_lmv)) {
+			struct lfsck_lmv_unit *llu;
+
+			spin_lock(&lfsck->li_lock);
+			llu = list_entry(lfsck->li_list_lmv.next,
+					 struct lfsck_lmv_unit, llu_link);
+			list_del_init(&llu->llu_link);
+			spin_unlock(&lfsck->li_lock);
+
+			lfsck->li_lmv = &llu->llu_lmv;
+			lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
+			rc = lfsck_open_dir(env, lfsck, 0);
+			if (rc == 0)
+				rc = lfsck_master_dir_engine(env, lfsck);
+
+			if (rc <= 0)
+				RETURN(rc);
+		}
+
 		lfsck->li_new_scanned++;
 		lfsck->li_pos_current.lp_oit_cookie = iops->store(env, di);
 		rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
@@ -1242,7 +1289,7 @@ out:
 }
 
 /**
- * Notify the LFSCK event to the instatnces on remote servers.
+ * Notify the LFSCK event to the instances on remote servers.
  *
  * The LFSCK assistant thread notifies the LFSCK instances on other
  * servers (MDT/OST) about some events, such as start new scanning,
@@ -1288,8 +1335,7 @@ static int lfsck_assistant_notify_others(const struct lu_env *env,
 		if (com->lc_type != LFSCK_TYPE_LAYOUT)
 			goto next;
 
-		lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
-			       LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ;
+		lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN;
 		lr->lr_speed = bk->lb_speed_limit;
 		lr->lr_version = bk->lb_version;
 		lr->lr_param |= bk->lb_param;
@@ -1569,6 +1615,7 @@ int lfsck_assistant_engine(void *args)
 	struct l_wait_info		   lwi     = { 0 };
 	int				   rc      = 0;
 	int				   rc1	   = 0;
+	int				   rc2;
 	ENTRY;
 
 	CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread start\n",
@@ -1678,11 +1725,16 @@ int lfsck_assistant_engine(void *args)
 				com->lc_time_last_checkpoint +
 				cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 
+			CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before "
+			       "the second-stage scaning\n",
+			       lfsck_lfsck2name(lfsck));
+
 			/* Flush async updates before handling orphan. */
-			dt_sync(env, lfsck->li_next);
+			rc2 = dt_sync(env, lfsck->li_next);
 
 			CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 "
-			       "scan start\n", lfsck_lfsck2name(lfsck));
+			       "scan start, synced: rc = %d\n",
+			       lfsck_lfsck2name(lfsck), rc2);
 
 			if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
 				GOTO(cleanup2, rc = 0);
@@ -1799,8 +1851,14 @@ cleanup2:
 		rc = rc1;
 	}
 
+	CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before exit\n",
+	       lfsck_lfsck2name(lfsck));
+
 	/* Flush async updates before exit. */
-	dt_sync(env, lfsck->li_next);
+	rc2 = dt_sync(env, lfsck->li_next);
+
+	CDEBUG(D_LFSCK, "%s: LFSCK assistant synced before exit: rc = %d\n",
+	       lfsck_lfsck2name(lfsck), rc2);
 
 	/* Under force exit case, some requests may be just freed without
 	 * verification, those objects should be re-handled when next run.