Patch applied to 2.12.0 on Fir MDS

2.12.0.pl4-LU-12037.diff

commit 565011c390b653fddfd26364e4df0097f862569e
Author: Lai Siyao <lai.siyao@whamcloud.com>
Date:   Mon Mar 4 23:56:16 2019 +0800

    LU-12037 mdt: add option for cross-MDT rename
    
    Add option mdt.mdt_remote_rename, if it's not set (it's set by
    default), do cross-MDT rename as cp, this is used for debug or
    user want to move inode in rename.
    
    Add sanity test_24z.
    
    Signed-off-by: Lai Siyao <lai.siyao@whamcloud.com>
    Change-Id: Ia0d122f1716f17078b375f770a193347a6e50708
    
    Conflicts:
    	lustre/mdt/mdt_internal.h
    	lustre/mdt/mdt_lproc.c

diff --git a/lustre/fld/fld_handler.c b/lustre/fld/fld_handler.c
index 9b84b62..42f00da 100644
--- a/lustre/fld/fld_handler.c
+++ b/lustre/fld/fld_handler.c
@@ -398,6 +398,7 @@ int fid_is_local(const struct lu_env *env,
 	}
 	return result;
 }
+EXPORT_SYMBOL(fid_is_local);
 
 static void fld_server_debugfs_fini(struct lu_server_fld *fld)
 {
diff --git a/lustre/mdt/mdt_handler.c b/lustre/mdt/mdt_handler.c
index 4caa22a..b537dd4 100644
--- a/lustre/mdt/mdt_handler.c
+++ b/lustre/mdt/mdt_handler.c
@@ -5077,6 +5077,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
 	m->mdt_enable_striped_dir = 1;
 	m->mdt_enable_dir_migration = 1;
 	m->mdt_enable_remote_dir_gid = 0;
+	m->mdt_enable_remote_rename = 1;
 
 	atomic_set(&m->mdt_mds_mds_conns, 0);
 	atomic_set(&m->mdt_async_commit_count, 0);
diff --git a/lustre/mdt/mdt_internal.h b/lustre/mdt/mdt_internal.h
index 5bf811f..788cb2a 100644
--- a/lustre/mdt/mdt_internal.h
+++ b/lustre/mdt/mdt_internal.h
@@ -251,6 +251,7 @@ struct mdt_device {
 				   mdt_enable_remote_dir:1,
 				   mdt_enable_striped_dir:1,
 				   mdt_enable_dir_migration:1,
+				   mdt_enable_remote_rename:1,
 				   mdt_skip_lfsck:1;
 
 				   /* user with gid can create remote/striped
diff --git a/lustre/mdt/mdt_lproc.c b/lustre/mdt/mdt_lproc.c
index da3c529..dc02fb6 100644
--- a/lustre/mdt/mdt_lproc.c
+++ b/lustre/mdt/mdt_lproc.c
@@ -979,6 +979,34 @@ mdt_migrate_hsm_allowed_seq_write(struct file *file, const char __user *buffer,
 }
 LPROC_SEQ_FOPS(mdt_migrate_hsm_allowed);
 
+static int mdt_enable_remote_rename_seq_show(struct seq_file *m, void *data)
+{
+	struct obd_device *obd = m->private;
+	struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
+
+	seq_printf(m, "%u\n",  mdt->mdt_enable_remote_rename);
+	return 0;
+}
+
+static ssize_t
+mdt_enable_remote_rename_seq_write(struct file *file, const char __user *buffer,
+				   size_t count, loff_t *off)
+{
+	struct seq_file *m = file->private_data;
+	struct obd_device *obd = m->private;
+	struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
+	bool val;
+	int rc;
+
+	rc = kstrtobool_from_user(buffer, count, &val);
+	if (rc)
+		return rc;
+
+	mdt->mdt_enable_remote_rename = val;
+	return count;
+}
+LPROC_SEQ_FOPS(mdt_enable_remote_rename);
+
 LPROC_SEQ_FOPS_RO_TYPE(mdt, recovery_status);
 LPROC_SEQ_FOPS_RO_TYPE(mdt, num_exports);
 LPROC_SEQ_FOPS_RO_TYPE(mdt, target_instance);
@@ -1048,6 +1076,8 @@ static struct lprocfs_vars lprocfs_mdt_obd_vars[] = {
 	  .fops =	&mdt_enable_striped_dir_fops		},
 	{ .name =	"enable_dir_migration",
 	  .fops =	&mdt_enable_dir_migration_fops		},
+	{ .name =	"enable_remote_rename",
+	  .fops =	&mdt_enable_remote_rename_fops		},
 	{ .name =	"hsm_control",
 	  .fops =	&mdt_hsm_cdt_control_fops		},
 	{ .name =	"recovery_time_hard",
diff --git a/lustre/mdt/mdt_reint.c b/lustre/mdt/mdt_reint.c
index fada9fc..e4c0fda 100644
--- a/lustre/mdt/mdt_reint.c
+++ b/lustre/mdt/mdt_reint.c
@@ -2257,20 +2257,10 @@ static int mdt_rename_determine_lock_order(struct mdt_thread_info *info,
  * 2 - srcdir child; 3 - tgtdir child.
  * Update on disk version of srcdir child.
  */
-/**
- * For DNE phase I, only these renames are allowed
- *	mv src_p/src_c tgt_p/tgt_c
- * 1. src_p/src_c/tgt_p/tgt_c are in the same MDT.
- * 2. src_p and tgt_p are same directory, and tgt_c does not
- *    exists. In this case, all of modification will happen
- *    in the MDT where ithesource parent is, only one remote
- *    update is needed, i.e. set c_time/m_time on the child.
- *    And tgt_c will be still in the same MDT as the original
- *    src_c.
- */
 static int mdt_reint_rename(struct mdt_thread_info *info,
 			    struct mdt_lock_handle *unused)
 {
+	struct mdt_device *mdt = info->mti_mdt;
 	struct mdt_reint_record *rr = &info->mti_rr;
 	struct md_attr *ma = &info->mti_attr;
 	struct ptlrpc_request *req = mdt_info_req(info);
@@ -2302,25 +2292,10 @@ static int mdt_reint_rename(struct mdt_thread_info *info,
 	    !fid_is_md_operative(rr->rr_fid2))
 		RETURN(-EPERM);
 
-	/*
-	 * Note: do not enqueue rename lock for replay request, because
-	 * if other MDT holds rename lock, but being blocked to wait for
-	 * this MDT to finish its recovery, and the failover MDT can not
-	 * get rename lock, which will cause deadlock.
-	 */
-	if (!req_is_replay(req)) {
-		rc = mdt_rename_lock(info, &rename_lh);
-		if (rc != 0) {
-			CERROR("%s: can't lock FS for rename: rc = %d\n",
-			       mdt_obd_name(info->mti_mdt), rc);
-			RETURN(rc);
-		}
-	}
-
 	/* find both parents. */
 	msrcdir = mdt_parent_find_check(info, rr->rr_fid1, 0);
 	if (IS_ERR(msrcdir))
-		GOTO(out_unlock_rename, rc = PTR_ERR(msrcdir));
+		RETURN(PTR_ERR(msrcdir));
 
 	OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME3, 5);
 
@@ -2333,9 +2308,36 @@ static int mdt_reint_rename(struct mdt_thread_info *info,
 			GOTO(out_put_srcdir, rc = PTR_ERR(mtgtdir));
 	}
 
+	/*
+	 * Note: do not enqueue rename lock for replay request, because
+	 * if other MDT holds rename lock, but being blocked to wait for
+	 * this MDT to finish its recovery, and the failover MDT can not
+	 * get rename lock, which will cause deadlock.
+	 */
+	if (!req_is_replay(req)) {
+		/*
+		 * Normally rename RPC is handled on the MDT with the target
+		 * directory (if target exists, it's on the MDT with the
+		 * target), if the source directory is remote, it's a hint that
+		 * source is remote too (this may not be true, but it won't
+		 * cause any issue), return -EXDEV early to avoid taking
+		 * rename_lock.
+		 */
+		if (!mdt->mdt_enable_remote_rename &&
+		    mdt_object_remote(msrcdir))
+			GOTO(out_put_tgtdir, rc = -EXDEV);
+
+		rc = mdt_rename_lock(info, &rename_lh);
+		if (rc != 0) {
+			CERROR("%s: can't lock FS for rename: rc = %d\n",
+			       mdt_obd_name(mdt), rc);
+			GOTO(out_put_tgtdir, rc);
+		}
+	}
+
 	rc = mdt_rename_determine_lock_order(info, msrcdir, mtgtdir);
 	if (rc < 0)
-		GOTO(out_put_tgtdir, rc);
+		GOTO(out_unlock_rename, rc);
 
 	reverse = rc;
 
@@ -2360,7 +2362,7 @@ relock:
 		rc = mdt_object_lock_save(info, mtgtdir, lh_tgtdirp, 1,
 					  cos_incompat);
 		if (rc)
-			GOTO(out_put_tgtdir, rc);
+			GOTO(out_unlock_rename, rc);
 
 		OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME, 5);
 
@@ -2368,13 +2370,13 @@ relock:
 					  cos_incompat);
 		if (rc != 0) {
 			mdt_object_unlock(info, mtgtdir, lh_tgtdirp, rc);
-			GOTO(out_put_tgtdir, rc);
+			GOTO(out_unlock_rename, rc);
 		}
 	} else {
 		rc = mdt_object_lock_save(info, msrcdir, lh_srcdirp, 0,
 					  cos_incompat);
 		if (rc)
-			GOTO(out_put_tgtdir, rc);
+			GOTO(out_unlock_rename, rc);
 
 		OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME, 5);
 
@@ -2390,7 +2392,7 @@ relock:
 		}
 		if (rc != 0) {
 			mdt_object_unlock(info, msrcdir, lh_srcdirp, rc);
-			GOTO(out_put_tgtdir, rc);
+			GOTO(out_unlock_rename, rc);
 		}
 	}
 
@@ -2420,6 +2422,9 @@ relock:
 		GOTO(out_put_old, rc = -ENOENT);
 	}
 
+	if (mdt_object_remote(mold) && !mdt->mdt_enable_remote_rename)
+		GOTO(out_put_old, rc = -EXDEV);
+
 	/* Check if @mtgtdir is subdir of @mold, before locking child
 	 * to avoid reverse locking. */
 	if (mtgtdir != msrcdir) {
@@ -2612,13 +2617,13 @@ out_put_old:
 out_unlock_parents:
 	mdt_object_unlock(info, mtgtdir, lh_tgtdirp, rc);
 	mdt_object_unlock(info, msrcdir, lh_srcdirp, rc);
+out_unlock_rename:
+	if (lustre_handle_is_used(&rename_lh))
+		mdt_rename_unlock(&rename_lh);
 out_put_tgtdir:
 	mdt_object_put(info->mti_env, mtgtdir);
 out_put_srcdir:
 	mdt_object_put(info->mti_env, msrcdir);
-out_unlock_rename:
-	if (lustre_handle_is_used(&rename_lh))
-		mdt_rename_unlock(&rename_lh);
 
 	/* If 'discard' is set then new_fid must exits.
 	 * DOM data discard need neither object nor lock,
diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh
index 8b1bc51..a2f7735 100755
--- a/lustre/tests/sanity.sh
+++ b/lustre/tests/sanity.sh
@@ -1287,6 +1287,36 @@ test_24y() {
 }
 run_test 24y "rename/link on the same dir should succeed"
 
+test_24z() {
+	[[ $MDSCOUNT -lt 2 ]] && skip_env "needs >= 2 MDTs"
+	[[ $MDS1_VERSION -lt $(version_code 2.12.51) ]] &&
+		skip "Need MDS version at least 2.12.51"
+
+	local index
+
+	for index in 0 1; do
+		$LFS mkdir -i $index $DIR/$tdir.$index || error "mkdir failed"
+		touch $DIR/$tdir.0/$tfile.$index || error "touch failed"
+	done
+
+	mv $DIR/$tdir.0/$tfile.0 $DIR/$tdir.1 || error "mv $tfile.0 failed"
+
+	index=$($LFS getstripe -m $DIR/$tdir.1/$tfile.0)
+	[ $index -eq 0 ] || error "$tfile.0 is on MDT$index"
+
+	local mdts=$(comma_list $(mdts_nodes))
+
+	do_nodes $mdts $LCTL set_param mdt.*.enable_remote_rename=0
+	stack_trap "do_nodes $mdts $LCTL \
+		set_param mdt.*.enable_remote_rename=1" EXIT
+
+	mv $DIR/$tdir.0/$tfile.1 $DIR/$tdir.1 || error "mv $tfile.1 failed"
+
+	index=$($LFS getstripe -m $DIR/$tdir.1/$tfile.1)
+	[ $index -eq 1 ] || error "$tfile.1 is on MDT$index"
+}
+run_test 24z "cross-MDT rename is done as cp"
+
 test_24A() { # LU-3182
 	local NFILES=5000