X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Flfsck%2Flfsck_engine.c;h=a1763a683953289c3edf6b6d91126152d8eadba0;hb=59842b15b028246d9d20fb9b8d276e16fffc908c;hp=c604bb9ed4f33b0d7b9592d85af22472c40249ac;hpb=d10200a80770f0029d1d665af954187b9ad883df;p=fs%2Flustre-release.git

diff --git a/lustre/lfsck/lfsck_engine.c b/lustre/lfsck/lfsck_engine.c
index c604bb9..a1763a6 100644
--- a/lustre/lfsck/lfsck_engine.c
+++ b/lustre/lfsck/lfsck_engine.c
@@ -136,6 +136,7 @@ static int lfsck_needs_scan_dir(const struct lu_env *env,
 	struct lfsck_thread_info *info    = lfsck_env_info(env);
 	struct lu_fid		 *fid     = &info->lti_fid;
 	struct lu_seq_range	 *range   = &info->lti_range;
+	struct lu_attr		 *la	  = &info->lti_la;
 	struct seq_server_site	 *ss	  = lfsck_dev_site(lfsck);
 	__u32			  idx	  = lfsck_dev_idx(lfsck);
 	int			  depth   = 0;
@@ -144,9 +145,21 @@ static int lfsck_needs_scan_dir(const struct lu_env *env,
 	if (list_empty(&lfsck->li_list_dir) || !S_ISDIR(lfsck_object_type(obj)))
 		return 0;
 
+	*fid = *lfsck_dto2fid(obj);
+	rc = dt_attr_get(env, obj, la);
+	if (unlikely(rc || (la->la_valid & LA_FLAGS &&
+			    la->la_flags & LUSTRE_ORPHAN_FL))) {
+		/* Orphan directory is empty, does not need scan. */
+		CDEBUG(D_INFO,
+		       "%s: skip orphan dir "DFID", %llx/%x: rc = %d\n",
+		       lfsck_lfsck2name(lfsck), PFID(fid),
+		       la->la_valid, la->la_flags, rc);
+
+		return rc;
+	}
+
 	LASSERT(ss != NULL);
 
-	*fid = *lfsck_dto2fid(obj);
 	while (1) {
 		/* Global /ROOT is visible. */
 		if (unlikely(lu_fid_eq(fid, &lfsck->li_global_root_fid)))
@@ -177,6 +190,7 @@ static int lfsck_needs_scan_dir(const struct lu_env *env,
 		if (fid_is_norm(fid))
 			return 1;
 
+		/* Only true after "obj = NULL" set below */
 		if (obj == NULL) {
 			obj = lfsck_object_find_bottom(env, lfsck, fid);
 			if (IS_ERR(obj))
@@ -187,7 +201,7 @@ static int lfsck_needs_scan_dir(const struct lu_env *env,
 				GOTO(out, rc = 0);
 		}
 
-		dt_read_lock(env, obj, MOR_TGT_CHILD);
+		dt_read_lock(env, obj, DT_TGT_CHILD);
 		if (unlikely(lfsck_is_dead_obj(obj))) {
 			dt_read_unlock(env, obj);
 
@@ -234,7 +248,7 @@ static int lfsck_load_stripe_lmv(const struct lu_env *env,
 	LASSERT(lfsck->li_obj_dir == NULL);
 	LASSERT(lfsck->li_lmv == NULL);
 
-	rc = lfsck_read_stripe_lmv(env, obj, lmv);
+	rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv);
 	if (rc == -ENODATA) {
 		lfsck->li_obj_dir = lfsck_object_get(obj);
 
@@ -362,7 +376,9 @@ int lfsck_open_dir(const struct lu_env *env,
 		GOTO(out, rc = PTR_ERR(di));
 
 	rc = iops->load(env, di, cookie);
-	if (rc == 0 || (rc > 0 && cookie > 0))
+	if (rc == -ENODATA)
+		rc = 1;
+	else if (rc == 0 || (rc > 0 && cookie > 0))
 		rc = iops->next(env, di);
 	else if (rc > 0)
 		rc = 0;
@@ -393,8 +409,7 @@ static int lfsck_checkpoint(const struct lu_env *env,
 	int			rc  = 0;
 	int			rc1 = 0;
 
-	if (likely(cfs_time_beforeq(cfs_time_current(),
-				    lfsck->li_time_next_checkpoint)))
+	if (likely(ktime_get_seconds() <= lfsck->li_time_next_checkpoint))
 		return 0;
 
 	lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
@@ -404,9 +419,9 @@ static int lfsck_checkpoint(const struct lu_env *env,
 			rc1 = rc;
 	}
 
-	lfsck->li_time_last_checkpoint = cfs_time_current();
+	lfsck->li_time_last_checkpoint = ktime_get_seconds();
 	lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
-				cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
+					 LFSCK_CHECKPOINT_INTERVAL;
 	return rc1 != 0 ? rc1 : rc;
 }
 
@@ -441,7 +456,7 @@ static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
 	/* Init otable-based iterator. */
 	if (pos == NULL) {
 		rc = iops->load(env, lfsck->li_di_oit, 0);
-		if (rc > 0) {
+		if (rc > 0 || unlikely(rc == -ENODATA)) {
 			lfsck->li_oit_over = 1;
 			rc = 0;
 		}
@@ -450,10 +465,10 @@ static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
 	}
 
 	rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie);
-	if (rc < 0)
-		GOTO(out, rc);
-	else if (rc > 0)
+	if (rc > 0 || unlikely(rc == -ENODATA))
 		lfsck->li_oit_over = 1;
+	else if (rc < 0)
+		GOTO(out, rc);
 
 	if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent))
 		GOTO(out, rc = 0);
@@ -507,9 +522,9 @@ out:
 			break;
 	}
 
-	lfsck->li_time_last_checkpoint = cfs_time_current();
+	lfsck->li_time_last_checkpoint = ktime_get_seconds();
 	lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
-				cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
+					 LFSCK_CHECKPOINT_INTERVAL;
 	return rc;
 }
 
@@ -606,9 +621,9 @@ static int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
 			       (__u32)com->lc_type, rc);
 	}
 
-	lfsck->li_time_last_checkpoint = cfs_time_current();
+	lfsck->li_time_last_checkpoint = ktime_get_seconds();
 	lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
-				cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
+					 LFSCK_CHECKPOINT_INTERVAL;
 
 	/* Ignore some component post failure to make other can go ahead. */
 	return result;
@@ -1015,6 +1030,10 @@ int lfsck_master_engine(void *args)
 	int			  rc;
 	ENTRY;
 
+	spin_lock(&lfsck->li_lock);
+	lfsck->li_task = current;
+	spin_unlock(&lfsck->li_lock);
+
 	/* There will be some objects verification during the LFSCK start,
 	 * such as the subsequent lfsck_verify_lpf(). Trigger low layer OI
 	 * OI scrub before that to handle the potential inconsistence. */
@@ -1058,6 +1077,11 @@ int lfsck_master_engine(void *args)
 	       current_pid());
 
 	spin_lock(&lfsck->li_lock);
+	if (unlikely(!thread_is_starting(thread))) {
+		spin_unlock(&lfsck->li_lock);
+		GOTO(fini_oit, rc = 0);
+	}
+
 	thread_set_flags(thread, SVC_RUNNING);
 	spin_unlock(&lfsck->li_lock);
 	wake_up_all(&thread->t_ctl_waitq);
@@ -1106,6 +1130,7 @@ fini_oit:
 fini_args:
 	spin_lock(&lfsck->li_lock);
 	thread_set_flags(thread, SVC_STOPPED);
+	lfsck->li_task = NULL;
 	spin_unlock(&lfsck->li_lock);
 	wake_up_all(&thread->t_ctl_waitq);
 	lfsck_thread_args_fini(lta);
@@ -1224,7 +1249,7 @@ again:
 	}
 	spin_unlock(&ltds->ltd_lock);
 
-	rc = ptlrpc_set_wait(set);
+	rc = ptlrpc_set_wait(env, set);
 	if (rc < 0) {
 		ptlrpc_set_destroy(set);
 		RETURN(rc);
@@ -1319,7 +1344,7 @@ static int lfsck_assistant_notify_others(const struct lu_env *env,
 		up_read(&ltds->ltd_rw_sem);
 
 		/* Sync up */
-		rc = ptlrpc_set_wait(set);
+		rc = ptlrpc_set_wait(env, set);
 		if (rc < 0) {
 			ptlrpc_set_destroy(set);
 			RETURN(rc);
@@ -1451,7 +1476,7 @@ again:
 		}
 		spin_unlock(&ltds->ltd_lock);
 
-		rc = ptlrpc_set_wait(set);
+		rc = ptlrpc_set_wait(env, set);
 		if (rc < 0) {
 			ptlrpc_set_destroy(set);
 			RETURN(rc);
@@ -1525,7 +1550,7 @@ again:
 		break;
 	}
 
-	rc1 = ptlrpc_set_wait(set);
+	rc1 = ptlrpc_set_wait(env, set);
 	ptlrpc_set_destroy(set);
 
 	RETURN(rc != 0 ? rc : rc1);
@@ -1595,7 +1620,7 @@ int lfsck_assistant_engine(void *args)
 		while (!list_empty(&lad->lad_req_list)) {
 			bool wakeup = false;
 
-			if (unlikely(lad->lad_exit ||
+			if (unlikely(test_bit(LAD_EXIT, &lad->lad_flags) ||
 				     !thread_is_running(mthread)))
 				GOTO(cleanup, rc = lad->lad_post_result);
 
@@ -1627,25 +1652,25 @@ int lfsck_assistant_engine(void *args)
 
 		l_wait_event(athread->t_ctl_waitq,
 			     !lfsck_assistant_req_empty(lad) ||
-			     lad->lad_exit ||
-			     lad->lad_to_post ||
-			     lad->lad_to_double_scan,
+			     test_bit(LAD_EXIT, &lad->lad_flags) ||
+			     test_bit(LAD_TO_POST, &lad->lad_flags) ||
+			     test_bit(LAD_TO_DOUBLE_SCAN, &lad->lad_flags),
 			     &lwi);
 
-		if (unlikely(lad->lad_exit))
+		if (unlikely(test_bit(LAD_EXIT, &lad->lad_flags)))
 			GOTO(cleanup, rc = lad->lad_post_result);
 
 		if (!list_empty(&lad->lad_req_list))
 			continue;
 
-		if (lad->lad_to_post) {
+		if (test_bit(LAD_TO_POST, &lad->lad_flags)) {
 			CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread post\n",
 			       lfsck_lfsck2name(lfsck), lad->lad_name);
 
-			if (unlikely(lad->lad_exit))
+			if (unlikely(test_bit(LAD_EXIT, &lad->lad_flags)))
 				GOTO(cleanup, rc = lad->lad_post_result);
 
-			lad->lad_to_post = 0;
+			clear_bit(LAD_TO_POST, &lad->lad_flags);
 			LASSERT(lad->lad_post_result > 0);
 
 			/* Wakeup the master engine to go ahead. */
@@ -1662,18 +1687,18 @@ int lfsck_assistant_engine(void *args)
 			       lad->lad_name, rc);
 		}
 
-		if (lad->lad_to_double_scan) {
-			lad->lad_to_double_scan = 0;
+		if (test_bit(LAD_TO_DOUBLE_SCAN, &lad->lad_flags)) {
+			clear_bit(LAD_TO_DOUBLE_SCAN, &lad->lad_flags);
 			atomic_inc(&lfsck->li_double_scan_count);
-			lad->lad_in_double_scan = 1;
+			set_bit(LAD_IN_DOUBLE_SCAN, &lad->lad_flags);
 			wake_up_all(&mthread->t_ctl_waitq);
 
 			com->lc_new_checked = 0;
 			com->lc_new_scanned = 0;
-			com->lc_time_last_checkpoint = cfs_time_current();
+			com->lc_time_last_checkpoint = ktime_get_seconds();
 			com->lc_time_next_checkpoint =
 				com->lc_time_last_checkpoint +
-				cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
+				LFSCK_CHECKPOINT_INTERVAL;
 
 			CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before "
 			       "the second-stage scaning\n",
@@ -1689,7 +1714,7 @@ int lfsck_assistant_engine(void *args)
 			if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
 				GOTO(cleanup, rc = 0);
 
-			while (lad->lad_in_double_scan) {
+			while (test_bit(LAD_IN_DOUBLE_SCAN, &lad->lad_flags)) {
 				rc = lfsck_assistant_query_others(env, com);
 				if (lfsck_phase2_next_ready(lad))
 					goto p2_next;
@@ -1704,12 +1729,13 @@ int lfsck_assistant_engine(void *args)
 							   NULL, NULL);
 				rc = l_wait_event(athread->t_ctl_waitq,
 					lfsck_phase2_next_ready(lad) ||
-					lad->lad_exit ||
+					test_bit(LAD_EXIT, &lad->lad_flags) ||
 					!thread_is_running(mthread),
 					&lwi);
 
-				if (unlikely(lad->lad_exit ||
-					     !thread_is_running(mthread)))
+				if (unlikely(
+					test_bit(LAD_EXIT, &lad->lad_flags) ||
+					!thread_is_running(mthread)))
 					GOTO(cleanup, rc = 0);
 
 				if (rc == -ETIMEDOUT)
@@ -1723,8 +1749,9 @@ p2_next:
 				if (rc != 0)
 					GOTO(cleanup, rc);
 
-				if (unlikely(lad->lad_exit ||
-					     !thread_is_running(mthread)))
+				if (unlikely(
+					test_bit(LAD_EXIT, &lad->lad_flags) ||
+					!thread_is_running(mthread)))
 					GOTO(cleanup, rc = 0);
 			}
 		}
@@ -1736,7 +1763,7 @@ cleanup:
 	if (rc < 0)
 		lad->lad_assistant_status = rc;
 
-	if (lad->lad_exit && lad->lad_post_result <= 0)
+	if (test_bit(LAD_EXIT, &lad->lad_flags) && lad->lad_post_result <= 0)
 		lao->la_fill_pos(env, com, &lfsck->li_pos_checkpoint);
 
 	thread_set_flags(athread, SVC_STOPPING);
@@ -1752,9 +1779,6 @@ cleanup:
 	}
 	spin_unlock(&lad->lad_lock);
 
-	LASSERTF(lad->lad_prefetched == 0, "unmatched prefeteched objs %d\n",
-		 lad->lad_prefetched);
-
 	memset(lr, 0, sizeof(*lr));
 	if (rc > 0) {
 		lr->lr_event = LE_PHASE2_DONE;
@@ -1813,8 +1837,8 @@ cleanup:
 	/* Under force exit case, some requests may be just freed without
 	 * verification, those objects should be re-handled when next run.
 	 * So not update the on-disk trace file under such case. */
-	if (lad->lad_in_double_scan) {
-		if (!lad->lad_exit)
+	if (test_bit(LAD_IN_DOUBLE_SCAN, &lad->lad_flags)) {
+		if (!test_bit(LAD_EXIT, &lad->lad_flags))
 			rc1 = lao->la_double_scan_result(env, com, rc);
 
 		CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 scan "
@@ -1823,13 +1847,12 @@ cleanup:
 	}
 
 fini:
-	if (lad->lad_in_double_scan)
+	if (test_bit(LAD_IN_DOUBLE_SCAN, &lad->lad_flags))
 		atomic_dec(&lfsck->li_double_scan_count);
 
 	spin_lock(&lad->lad_lock);
 	lad->lad_assistant_status = (rc1 != 0 ? rc1 : rc);
 	thread_set_flags(athread, SVC_STOPPED);
-	wake_up_all(&mthread->t_ctl_waitq);
 	lad->lad_task = NULL;
 	spin_unlock(&lad->lad_lock);
 
@@ -1838,6 +1861,7 @@ fini:
 	       lad->lad_assistant_status);
 
 	lfsck_thread_args_fini(lta);
+	wake_up_all(&mthread->t_ctl_waitq);
 
 	return rc;
 }