[PATCH v2 0/4] ubifs: Add overlayfs support

[PATCH v2 0/4] ubifs: Add overlayfs support

[Richard Weinberger]

To support overlayfs in a proper way a filesystem has to offer support
for DT_CHR, RENAME_WHITEOUT and RENAME_EXCHANGE.
While UBIFS supports DT_CHR it lacks support for RENAME_WHITEOUT and
RENAME_EXCHANGE. This patch series implement the missing bits.
RENAME_WHITEOUT itself depends on O_TMPFILE, therefore O_TMPFILE is now
also supported.

Changes since v1:
 - Build fix for CONFIG_LOCKDEP=y
 - Fix for wrong dentry name length calculation in ubifs_jnl_xrename()
 - Fix for wrong directory link count after RENAME_EXCHANGE

Richard Weinberger (4):
  ubifs: Implement O_TMPFILE
  ubifs: Implement RENAME_WHITEOUT
  ubifs: Implement RENAME_EXCHANGE
  ubifs: Use move variable in ubifs_rename()

 fs/ubifs/dir.c     | 242 +++++++++++++++++++++++++++++++++++++++++++++++++----
 fs/ubifs/journal.c | 188 +++++++++++++++++++++++++++++++++++++----
 fs/ubifs/ubifs.h   |   8 +-
 3 files changed, 404 insertions(+), 34 deletions(-)

-- 
2.7.3

[PATCH 1/4] ubifs: Implement O_TMPFILE

[Richard Weinberger]

This patchs adds O_TMPFILE support to UBIFS.
A temp file is a reference to an unlinked inode, a user
holding the reference can use it. As soon it is being closed
all data vanishes.

Signed-off-by: Richard Weinberger <richard@nod.at>
---
 fs/ubifs/dir.c | 71 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 71 insertions(+)

diff --git a/fs/ubifs/dir.c b/fs/ubifs/dir.c
index 4b86d3a..8460f48 100644
--- a/fs/ubifs/dir.c
+++ b/fs/ubifs/dir.c
@@ -301,6 +301,76 @@ out_budg:
 	return err;
 }
 
+static int ubifs_tmpfile(struct inode *dir, struct dentry *dentry,
+			 umode_t mode)
+{
+	struct inode *inode;
+	struct ubifs_info *c = dir->i_sb->s_fs_info;
+	struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1};
+	struct ubifs_budget_req ino_req = { .dirtied_ino = 1 };
+	struct ubifs_inode *ui, *dir_ui = ubifs_inode(dir);
+	int err, instantiated = 0;
+
+	/*
+	 * Budget request settings: new dirty inode, new direntry,
+	 * budget for dirtied inode will be released via writeback.
+	 */
+
+	dbg_gen("dent '%pd', mode %#hx in dir ino %lu",
+		dentry, mode, dir->i_ino);
+
+	err = ubifs_budget_space(c, &req);
+	if (err)
+		return err;
+
+	err = ubifs_budget_space(c, &ino_req);
+	if (err) {
+		ubifs_release_budget(c, &req);
+		return err;
+	}
+
+	inode = ubifs_new_inode(c, dir, mode);
+	if (IS_ERR(inode)) {
+		err = PTR_ERR(inode);
+		goto out_budg;
+	}
+	ui = ubifs_inode(inode);
+
+	err = ubifs_init_security(dir, inode, &dentry->d_name);
+	if (err)
+		goto out_inode;
+
+	mutex_lock(&ui->ui_mutex);
+	insert_inode_hash(inode);
+	d_tmpfile(dentry, inode);
+	ubifs_assert(ui->dirty);
+	instantiated = 1;
+	mutex_unlock(&ui->ui_mutex);
+
+	mutex_lock(&dir_ui->ui_mutex);
+	err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 1, 0);
+	if (err)
+		goto out_cancel;
+	mutex_unlock(&dir_ui->ui_mutex);
+
+	ubifs_release_budget(c, &req);
+
+	return 0;
+
+out_cancel:
+	mutex_unlock(&dir_ui->ui_mutex);
+out_inode:
+	make_bad_inode(inode);
+	if (!instantiated)
+		iput(inode);
+out_budg:
+	ubifs_release_budget(c, &req);
+	if (!instantiated)
+		ubifs_release_budget(c, &ino_req);
+	ubifs_err(c, "cannot create temporary file, error %d", err);
+	return err;
+}
+
 /**
  * vfs_dent_type - get VFS directory entry type.
  * @type: UBIFS directory entry type
@@ -1189,6 +1259,7 @@ const struct inode_operations ubifs_dir_inode_operations = {
 #ifdef CONFIG_UBIFS_ATIME_SUPPORT
 	.update_time = ubifs_update_time,
 #endif
+	.tmpfile     = ubifs_tmpfile,
 };
 
 const struct file_operations ubifs_dir_operations = {
-- 
2.7.3

[PATCH 2/4] ubifs: Implement RENAME_WHITEOUT

[Richard Weinberger]

Adds RENAME_WHITEOUT support to UBIFS, we implement
it in the same way as ext4 and xfs do.
For an overview of other ways to implement it please
refere to commit 7dcf5c3e4527 ("xfs: add RENAME_WHITEOUT support").

Signed-off-by: Richard Weinberger <richard@nod.at>
---
 fs/ubifs/dir.c     | 120 ++++++++++++++++++++++++++++++++++++++++++++---------
 fs/ubifs/journal.c |  43 +++++++++++++------
 fs/ubifs/ubifs.h   |   4 +-
 3 files changed, 134 insertions(+), 33 deletions(-)

diff --git a/fs/ubifs/dir.c b/fs/ubifs/dir.c
index 8460f48..e028253 100644
--- a/fs/ubifs/dir.c
+++ b/fs/ubifs/dir.c
@@ -301,8 +301,8 @@ out_budg:
 	return err;
 }
 
-static int ubifs_tmpfile(struct inode *dir, struct dentry *dentry,
-			 umode_t mode)
+static int do_tmpfile(struct inode *dir, struct dentry *dentry,
+		      umode_t mode, struct inode **whiteout)
 {
 	struct inode *inode;
 	struct ubifs_info *c = dir->i_sb->s_fs_info;
@@ -336,14 +336,27 @@ static int ubifs_tmpfile(struct inode *dir, struct dentry *dentry,
 	}
 	ui = ubifs_inode(inode);
 
+	if (whiteout) {
+		init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
+		ubifs_assert(inode->i_op == &ubifs_file_inode_operations);
+	}
+
 	err = ubifs_init_security(dir, inode, &dentry->d_name);
 	if (err)
 		goto out_inode;
 
 	mutex_lock(&ui->ui_mutex);
 	insert_inode_hash(inode);
-	d_tmpfile(dentry, inode);
+
+	if (whiteout) {
+		mark_inode_dirty(inode);
+		drop_nlink(inode);
+		*whiteout = inode;
+	} else {
+		d_tmpfile(dentry, inode);
+	}
 	ubifs_assert(ui->dirty);
+
 	instantiated = 1;
 	mutex_unlock(&ui->ui_mutex);
 
@@ -371,6 +384,12 @@ out_budg:
 	return err;
 }
 
+static int ubifs_tmpfile(struct inode *dir, struct dentry *dentry,
+			 umode_t mode)
+{
+	return do_tmpfile(dir, dentry, mode, NULL);
+}
+
 /**
  * vfs_dent_type - get VFS directory entry type.
  * @type: UBIFS directory entry type
@@ -997,37 +1016,43 @@ out_budg:
 }
 
 /**
- * lock_3_inodes - a wrapper for locking three UBIFS inodes.
+ * lock_4_inodes - a wrapper for locking three UBIFS inodes.
  * @inode1: first inode
  * @inode2: second inode
  * @inode3: third inode
+ * @inode4: fouth inode
  *
  * This function is used for 'ubifs_rename()' and @inode1 may be the same as
- * @inode2 whereas @inode3 may be %NULL.
+ * @inode2 whereas @inode3 and @inode4 may be %NULL.
  *
  * We do not implement any tricks to guarantee strict lock ordering, because
  * VFS has already done it for us on the @i_mutex. So this is just a simple
  * wrapper function.
  */
-static void lock_3_inodes(struct inode *inode1, struct inode *inode2,
-			  struct inode *inode3)
+static void lock_4_inodes(struct inode *inode1, struct inode *inode2,
+			  struct inode *inode3, struct inode *inode4)
 {
 	mutex_lock_nested(&ubifs_inode(inode1)->ui_mutex, WB_MUTEX_1);
 	if (inode2 != inode1)
 		mutex_lock_nested(&ubifs_inode(inode2)->ui_mutex, WB_MUTEX_2);
 	if (inode3)
 		mutex_lock_nested(&ubifs_inode(inode3)->ui_mutex, WB_MUTEX_3);
+	if (inode4)
+		mutex_lock_nested(&ubifs_inode(inode4)->ui_mutex, WB_MUTEX_4);
 }
 
 /**
- * unlock_3_inodes - a wrapper for unlocking three UBIFS inodes for rename.
+ * unlock_4_inodes - a wrapper for unlocking three UBIFS inodes for rename.
  * @inode1: first inode
  * @inode2: second inode
  * @inode3: third inode
+ * @inode4: fouth inode
  */
-static void unlock_3_inodes(struct inode *inode1, struct inode *inode2,
-			    struct inode *inode3)
+static void unlock_4_inodes(struct inode *inode1, struct inode *inode2,
+			    struct inode *inode3, struct inode *inode4)
 {
+	if (inode4)
+		mutex_unlock(&ubifs_inode(inode4)->ui_mutex);
 	if (inode3)
 		mutex_unlock(&ubifs_inode(inode3)->ui_mutex);
 	if (inode1 != inode2)
@@ -1036,12 +1061,15 @@ static void unlock_3_inodes(struct inode *inode1, struct inode *inode2,
 }
 
 static int ubifs_rename(struct inode *old_dir, struct dentry *old_dentry,
-			struct inode *new_dir, struct dentry *new_dentry)
+			struct inode *new_dir, struct dentry *new_dentry,
+			unsigned int flags)
 {
 	struct ubifs_info *c = old_dir->i_sb->s_fs_info;
 	struct inode *old_inode = d_inode(old_dentry);
 	struct inode *new_inode = d_inode(new_dentry);
+	struct inode *whiteout = NULL;
 	struct ubifs_inode *old_inode_ui = ubifs_inode(old_inode);
+	struct ubifs_inode *whiteout_ui = NULL;
 	int err, release, sync = 0, move = (new_dir != old_dir);
 	int is_dir = S_ISDIR(old_inode->i_mode);
 	int unlink = !!new_inode;
@@ -1063,15 +1091,18 @@ static int ubifs_rename(struct inode *old_dir, struct dentry *old_dentry,
 	 * separately.
 	 */
 
-	dbg_gen("dent '%pd' ino %lu in dir ino %lu to dent '%pd' in dir ino %lu",
+	dbg_gen("dent '%pd' ino %lu in dir ino %lu to dent '%pd' in dir ino %lu flags 0x%x",
 		old_dentry, old_inode->i_ino, old_dir->i_ino,
-		new_dentry, new_dir->i_ino);
+		new_dentry, new_dir->i_ino, flags);
+
+	if (flags & ~(RENAME_NOREPLACE | RENAME_WHITEOUT))
+		return -EINVAL;
+
 	ubifs_assert(inode_is_locked(old_dir));
 	ubifs_assert(inode_is_locked(new_dir));
 	if (unlink)
 		ubifs_assert(inode_is_locked(new_inode));
 
-
 	if (unlink && is_dir) {
 		err = check_dir_empty(c, new_inode);
 		if (err)
@@ -1087,7 +1118,32 @@ static int ubifs_rename(struct inode *old_dir, struct dentry *old_dentry,
 		return err;
 	}
 
-	lock_3_inodes(old_dir, new_dir, new_inode);
+	if (flags & RENAME_WHITEOUT) {
+		union ubifs_dev_desc *dev = NULL;
+
+		dev = kmalloc(sizeof(union ubifs_dev_desc), GFP_NOFS);
+		if (!dev) {
+			ubifs_release_budget(c, &req);
+			ubifs_release_budget(c, &ino_req);
+			return -ENOMEM;
+		}
+
+		err = do_tmpfile(old_dir, old_dentry, S_IFCHR | WHITEOUT_MODE, &whiteout);
+		if (err) {
+			ubifs_release_budget(c, &req);
+			ubifs_release_budget(c, &ino_req);
+			kfree(dev);
+			return err;
+		}
+
+		whiteout->i_state |= I_LINKABLE;
+		whiteout_ui = ubifs_inode(whiteout);
+		whiteout_ui->data = dev;
+		whiteout_ui->data_len = ubifs_encode_dev(dev, MKDEV(0, 0));
+		ubifs_assert(!whiteout_ui->dirty);
+	}
+
+	lock_4_inodes(old_dir, new_dir, new_inode, whiteout);
 
 	/*
 	 * Like most other Unix systems, set the @i_ctime for inodes on a
@@ -1157,12 +1213,34 @@ static int ubifs_rename(struct inode *old_dir, struct dentry *old_dentry,
 		if (unlink && IS_SYNC(new_inode))
 			sync = 1;
 	}
-	err = ubifs_jnl_rename(c, old_dir, old_dentry, new_dir, new_dentry,
+
+	if (whiteout) {
+		struct ubifs_budget_req wht_req = { .dirtied_ino = 1,
+				.dirtied_ino_d = \
+				ALIGN(ubifs_inode(whiteout)->data_len, 8) };
+
+		err = ubifs_budget_space(c, &wht_req);
+		if (err) {
+			ubifs_release_budget(c, &req);
+			ubifs_release_budget(c, &ino_req);
+			kfree(whiteout_ui->data);
+			whiteout_ui->data_len = 0;
+			iput(whiteout);
+			return err;
+		}
+
+		inc_nlink(whiteout);
+		mark_inode_dirty(whiteout);
+		whiteout->i_state &= ~I_LINKABLE;
+		iput(whiteout);
+	}
+
+	err = ubifs_jnl_rename(c, old_dir, old_dentry, new_dir, new_dentry, whiteout,
 			       sync);
 	if (err)
 		goto out_cancel;
 
-	unlock_3_inodes(old_dir, new_dir, new_inode);
+	unlock_4_inodes(old_dir, new_dir, new_inode, whiteout);
 	ubifs_release_budget(c, &req);
 
 	mutex_lock(&old_inode_ui->ui_mutex);
@@ -1195,7 +1273,11 @@ out_cancel:
 				inc_nlink(old_dir);
 		}
 	}
-	unlock_3_inodes(old_dir, new_dir, new_inode);
+	if (whiteout) {
+		drop_nlink(whiteout);
+		iput(whiteout);
+	}
+	unlock_4_inodes(old_dir, new_dir, new_inode, whiteout);
 	ubifs_release_budget(c, &ino_req);
 	ubifs_release_budget(c, &req);
 	return err;
@@ -1249,7 +1331,7 @@ const struct inode_operations ubifs_dir_inode_operations = {
 	.mkdir       = ubifs_mkdir,
 	.rmdir       = ubifs_rmdir,
 	.mknod       = ubifs_mknod,
-	.rename      = ubifs_rename,
+	.rename2     = ubifs_rename,
 	.setattr     = ubifs_setattr,
 	.getattr     = ubifs_getattr,
 	.setxattr    = generic_setxattr,
diff --git a/fs/ubifs/journal.c b/fs/ubifs/journal.c
index 0b9da5b..6b27ddd 100644
--- a/fs/ubifs/journal.c
+++ b/fs/ubifs/journal.c
@@ -917,14 +917,15 @@ int ubifs_jnl_delete_inode(struct ubifs_info *c, const struct inode *inode)
  * @sync: non-zero if the write-buffer has to be synchronized
  *
  * This function implements the re-name operation which may involve writing up
- * to 3 inodes and 2 directory entries. It marks the written inodes as clean
+ * to 4 inodes and 2 directory entries. It marks the written inodes as clean
  * and returns zero on success. In case of failure, a negative error code is
  * returned.
  */
 int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
 		     const struct dentry *old_dentry,
 		     const struct inode *new_dir,
-		     const struct dentry *new_dentry, int sync)
+		     const struct dentry *new_dentry,
+		     const struct inode *whiteout, int sync)
 {
 	void *p;
 	union ubifs_key key;
@@ -980,13 +981,19 @@ int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
 	zero_dent_node_unused(dent);
 	ubifs_prep_grp_node(c, dent, dlen1, 0);
 
-	/* Make deletion dent */
 	dent2 = (void *)dent + aligned_dlen1;
 	dent2->ch.node_type = UBIFS_DENT_NODE;
 	dent_key_init_flash(c, &dent2->key, old_dir->i_ino,
 			    &old_dentry->d_name);
-	dent2->inum = 0;
-	dent2->type = DT_UNKNOWN;
+
+	if (whiteout) {
+		dent2->inum = cpu_to_le64(whiteout->i_ino);
+		dent2->type = get_dent_type(whiteout->i_mode);
+	} else {
+		/* Make deletion dent */
+		dent2->inum = 0;
+		dent2->type = DT_UNKNOWN;
+	}
 	dent2->nlen = cpu_to_le16(old_dentry->d_name.len);
 	memcpy(dent2->name, old_dentry->d_name.name, old_dentry->d_name.len);
 	dent2->name[old_dentry->d_name.len] = '\0';
@@ -1035,16 +1042,26 @@ int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
 	if (err)
 		goto out_ro;
 
-	err = ubifs_add_dirt(c, lnum, dlen2);
-	if (err)
-		goto out_ro;
+	offs += aligned_dlen1;
+	if (whiteout) {
+		dent_key_init(c, &key, old_dir->i_ino, &old_dentry->d_name);
+		err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen2, &old_dentry->d_name);
+		if (err)
+			goto out_ro;
 
-	dent_key_init(c, &key, old_dir->i_ino, &old_dentry->d_name);
-	err = ubifs_tnc_remove_nm(c, &key, &old_dentry->d_name);
-	if (err)
-		goto out_ro;
+		ubifs_delete_orphan(c, whiteout->i_ino);
+	} else {
+		err = ubifs_add_dirt(c, lnum, dlen2);
+		if (err)
+			goto out_ro;
+
+		dent_key_init(c, &key, old_dir->i_ino, &old_dentry->d_name);
+		err = ubifs_tnc_remove_nm(c, &key, &old_dentry->d_name);
+		if (err)
+			goto out_ro;
+	}
 
-	offs += aligned_dlen1 + aligned_dlen2;
+	offs += aligned_dlen2;
 	if (new_inode) {
 		ino_key_init(c, &key, new_inode->i_ino);
 		err = ubifs_tnc_add(c, &key, lnum, offs, ilen);
diff --git a/fs/ubifs/ubifs.h b/fs/ubifs/ubifs.h
index 4617d45..ca86c93 100644
--- a/fs/ubifs/ubifs.h
+++ b/fs/ubifs/ubifs.h
@@ -157,6 +157,7 @@ enum {
 	WB_MUTEX_1 = 0,
 	WB_MUTEX_2 = 1,
 	WB_MUTEX_3 = 2,
+	WB_MUTEX_4 = 3,
 };
 
 /*
@@ -1523,7 +1524,8 @@ int ubifs_jnl_delete_inode(struct ubifs_info *c, const struct inode *inode);
 int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
 		     const struct dentry *old_dentry,
 		     const struct inode *new_dir,
-		     const struct dentry *new_dentry, int sync);
+		     const struct dentry *new_dentry,
+		     const struct inode *whiteout, int sync);
 int ubifs_jnl_truncate(struct ubifs_info *c, const struct inode *inode,
 		       loff_t old_size, loff_t new_size);
 int ubifs_jnl_delete_xattr(struct ubifs_info *c, const struct inode *host,
-- 
2.7.3

[PATCH 3/4] ubifs: Implement RENAME_EXCHANGE

[Richard Weinberger]

Adds RENAME_EXCHANGE to UBIFS, the operation itself
is completely disjunct from a regular rename() that's
why we dispatch very early in ubifs_reaname().

RENAME_EXCHANGE used by the renameat2() system call
allows the caller to exchange two paths atomically.
Both paths have to exist and have to be on the same
filesystem.

Signed-off-by: Richard Weinberger <richard@nod.at>
---
 fs/ubifs/dir.c     |  65 +++++++++++++++++++++---
 fs/ubifs/journal.c | 141 +++++++++++++++++++++++++++++++++++++++++++++++++++++
 fs/ubifs/ubifs.h   |   4 ++
 3 files changed, 204 insertions(+), 6 deletions(-)

diff --git a/fs/ubifs/dir.c b/fs/ubifs/dir.c
index e028253..ccd9128 100644
--- a/fs/ubifs/dir.c
+++ b/fs/ubifs/dir.c
@@ -1095,11 +1095,6 @@ static int ubifs_rename(struct inode *old_dir, struct dentry *old_dentry,
 		old_dentry, old_inode->i_ino, old_dir->i_ino,
 		new_dentry, new_dir->i_ino, flags);
 
-	if (flags & ~(RENAME_NOREPLACE | RENAME_WHITEOUT))
-		return -EINVAL;
-
-	ubifs_assert(inode_is_locked(old_dir));
-	ubifs_assert(inode_is_locked(new_dir));
 	if (unlink)
 		ubifs_assert(inode_is_locked(new_inode));
 
@@ -1283,6 +1278,64 @@ out_cancel:
 	return err;
 }
 
+static int ubifs_xrename(struct inode *old_dir, struct dentry *old_dentry,
+			struct inode *new_dir, struct dentry *new_dentry)
+{
+	struct ubifs_info *c = old_dir->i_sb->s_fs_info;
+	struct ubifs_budget_req req = { .new_dent = 1, .mod_dent = 1,
+				.dirtied_ino = 2 };
+	int sync = IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir);
+	struct inode *fst_inode = d_inode(old_dentry);
+	struct inode *snd_inode = d_inode(new_dentry);
+	struct timespec time;
+	int err;
+
+	ubifs_assert(fst_inode && snd_inode);
+
+	lock_4_inodes(old_dir, new_dir, NULL, NULL);
+
+	time = ubifs_current_time(old_dir);
+	fst_inode->i_ctime = time;
+	snd_inode->i_ctime = time;
+	old_dir->i_mtime = old_dir->i_ctime = time;
+	new_dir->i_mtime = new_dir->i_ctime = time;
+
+	if (old_dir != new_dir) {
+		if (S_ISDIR(fst_inode->i_mode) && !S_ISDIR(snd_inode->i_mode)) {
+			inc_nlink(new_dir);
+			drop_nlink(old_dir);
+		}
+		else if (!S_ISDIR(fst_inode->i_mode) && S_ISDIR(snd_inode->i_mode)) {
+			drop_nlink(new_dir);
+			inc_nlink(old_dir);
+		}
+	}
+
+	err = ubifs_jnl_xrename(c, old_dir, old_dentry, new_dir, new_dentry,
+				sync);
+
+	unlock_4_inodes(old_dir, new_dir, NULL, NULL);
+	ubifs_release_budget(c, &req);
+
+	return err;
+}
+
+static int ubifs_rename2(struct inode *old_dir, struct dentry *old_dentry,
+			struct inode *new_dir, struct dentry *new_dentry,
+			unsigned int flags)
+{
+	if (flags & ~(RENAME_NOREPLACE | RENAME_WHITEOUT | RENAME_EXCHANGE))
+		return -EINVAL;
+
+	ubifs_assert(inode_is_locked(old_dir));
+	ubifs_assert(inode_is_locked(new_dir));
+
+	if (flags & RENAME_EXCHANGE)
+		return ubifs_xrename(old_dir, old_dentry, new_dir, new_dentry);
+
+	return ubifs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
+}
+
 int ubifs_getattr(struct vfsmount *mnt, struct dentry *dentry,
 		  struct kstat *stat)
 {
@@ -1331,7 +1384,7 @@ const struct inode_operations ubifs_dir_inode_operations = {
 	.mkdir       = ubifs_mkdir,
 	.rmdir       = ubifs_rmdir,
 	.mknod       = ubifs_mknod,
-	.rename2     = ubifs_rename,
+	.rename2     = ubifs_rename2,
 	.setattr     = ubifs_setattr,
 	.getattr     = ubifs_getattr,
 	.setxattr    = generic_setxattr,
diff --git a/fs/ubifs/journal.c b/fs/ubifs/journal.c
index 6b27ddd..8e07f13 100644
--- a/fs/ubifs/journal.c
+++ b/fs/ubifs/journal.c
@@ -908,6 +908,147 @@ int ubifs_jnl_delete_inode(struct ubifs_info *c, const struct inode *inode)
 }
 
 /**
+ * ubifs_jnl_xrename - cross rename two directory entries.
+ * @c: UBIFS file-system description object
+ * @fst_dir: parent inode of 1st directory entry to exchange
+ * @fst_dentry: 1st directory entry to exchange
+ * @snd_dir: parent inode of 2nd directory entry to exchange
+ * @snd_dentry: 2nd directory entry to exchange
+ * @sync: non-zero if the write-buffer has to be synchronized
+ *
+ * This function implements the cross rename operation which may involve
+ * writing 2 inodes and 2 directory entries. It marks the written inodes as clean
+ * and returns zero on success. In case of failure, a negative error code is
+ * returned.
+ */
+int ubifs_jnl_xrename(struct ubifs_info *c, const struct inode *fst_dir,
+		      const struct dentry *fst_dentry,
+		      const struct inode *snd_dir,
+		      const struct dentry *snd_dentry, int sync)
+{
+	union ubifs_key key;
+	struct ubifs_dent_node *dent1, *dent2;
+	int err, dlen1, dlen2, lnum, offs, len, plen = UBIFS_INO_NODE_SZ;
+	int aligned_dlen1, aligned_dlen2;
+	int twoparents = (fst_dir != snd_dir);
+	const struct inode *fst_inode = d_inode(fst_dentry);
+	const struct inode *snd_inode = d_inode(snd_dentry);
+	void *p;
+
+	dbg_jnl("dent '%pd' in dir ino %lu between dent '%pd' in dir ino %lu",
+		fst_dentry, fst_dir->i_ino, snd_dentry, snd_dir->i_ino);
+
+	ubifs_assert(ubifs_inode(fst_dir)->data_len == 0);
+	ubifs_assert(ubifs_inode(snd_dir)->data_len == 0);
+	ubifs_assert(mutex_is_locked(&ubifs_inode(fst_dir)->ui_mutex));
+	ubifs_assert(mutex_is_locked(&ubifs_inode(snd_dir)->ui_mutex));
+
+	dlen1 = UBIFS_DENT_NODE_SZ + snd_dentry->d_name.len + 1;
+	dlen2 = UBIFS_DENT_NODE_SZ + fst_dentry->d_name.len + 1;
+	aligned_dlen1 = ALIGN(dlen1, 8);
+	aligned_dlen2 = ALIGN(dlen2, 8);
+
+	len = aligned_dlen1 + aligned_dlen2 + ALIGN(plen, 8);
+	if (twoparents)
+		len += plen;
+
+	dent1 = kmalloc(len, GFP_NOFS);
+	if (!dent1)
+		return -ENOMEM;
+
+	/* Make reservation before allocating sequence numbers */
+	err = make_reservation(c, BASEHD, len);
+	if (err)
+		goto out_free;
+
+	/* Make new dent for 1st entry */
+	dent1->ch.node_type = UBIFS_DENT_NODE;
+	dent_key_init_flash(c, &dent1->key, snd_dir->i_ino, &snd_dentry->d_name);
+	dent1->inum = cpu_to_le64(fst_inode->i_ino);
+	dent1->type = get_dent_type(fst_inode->i_mode);
+	dent1->nlen = cpu_to_le16(snd_dentry->d_name.len);
+	memcpy(dent1->name, snd_dentry->d_name.name, snd_dentry->d_name.len);
+	dent1->name[snd_dentry->d_name.len] = '\0';
+	zero_dent_node_unused(dent1);
+	ubifs_prep_grp_node(c, dent1, dlen1, 0);
+
+	/* Make new dent for 2nd entry */
+	dent2 = (void *)dent1 + aligned_dlen1;
+	dent2->ch.node_type = UBIFS_DENT_NODE;
+	dent_key_init_flash(c, &dent2->key, fst_dir->i_ino, &fst_dentry->d_name);
+	dent2->inum = cpu_to_le64(snd_inode->i_ino);
+	dent2->type = get_dent_type(snd_inode->i_mode);
+	dent2->nlen = cpu_to_le16(fst_dentry->d_name.len);
+	memcpy(dent2->name, fst_dentry->d_name.name, fst_dentry->d_name.len);
+	dent2->name[fst_dentry->d_name.len] = '\0';
+	zero_dent_node_unused(dent2);
+	ubifs_prep_grp_node(c, dent2, dlen2, 0);
+
+	p = (void *)dent2 + aligned_dlen2;
+	if (!twoparents)
+		pack_inode(c, p, fst_dir, 1);
+	else {
+		pack_inode(c, p, fst_dir, 0);
+		p += ALIGN(plen, 8);
+		pack_inode(c, p, snd_dir, 1);
+	}
+
+	err = write_head(c, BASEHD, dent1, len, &lnum, &offs, sync);
+	if (err)
+		goto out_release;
+	if (!sync) {
+		struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
+
+		ubifs_wbuf_add_ino_nolock(wbuf, fst_dir->i_ino);
+		ubifs_wbuf_add_ino_nolock(wbuf, snd_dir->i_ino);
+	}
+	release_head(c, BASEHD);
+
+	dent_key_init(c, &key, snd_dir->i_ino, &snd_dentry->d_name);
+	err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, &snd_dentry->d_name);
+	if (err)
+		goto out_ro;
+
+	offs += aligned_dlen1;
+	dent_key_init(c, &key, fst_dir->i_ino, &fst_dentry->d_name);
+	err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen2, &fst_dentry->d_name);
+	if (err)
+		goto out_ro;
+
+	offs += aligned_dlen2;
+
+	ino_key_init(c, &key, fst_dir->i_ino);
+	err = ubifs_tnc_add(c, &key, lnum, offs, plen);
+	if (err)
+		goto out_ro;
+
+	if (twoparents) {
+		offs += ALIGN(plen, 8);
+		ino_key_init(c, &key, snd_dir->i_ino);
+		err = ubifs_tnc_add(c, &key, lnum, offs, plen);
+		if (err)
+			goto out_ro;
+	}
+
+	finish_reservation(c);
+
+	mark_inode_clean(c, ubifs_inode(fst_dir));
+	if (twoparents)
+		mark_inode_clean(c, ubifs_inode(snd_dir));
+	kfree(dent1);
+	return 0;
+
+out_release:
+	release_head(c, BASEHD);
+out_ro:
+	ubifs_ro_mode(c, err);
+	finish_reservation(c);
+out_free:
+	kfree(dent1);
+	return err;
+}
+
+/**
  * ubifs_jnl_rename - rename a directory entry.
  * @c: UBIFS file-system description object
  * @old_dir: parent inode of directory entry to rename
diff --git a/fs/ubifs/ubifs.h b/fs/ubifs/ubifs.h
index ca86c93..096035e 100644
--- a/fs/ubifs/ubifs.h
+++ b/fs/ubifs/ubifs.h
@@ -1521,6 +1521,10 @@ int ubifs_jnl_write_data(struct ubifs_info *c, const struct inode *inode,
 			 const union ubifs_key *key, const void *buf, int len);
 int ubifs_jnl_write_inode(struct ubifs_info *c, const struct inode *inode);
 int ubifs_jnl_delete_inode(struct ubifs_info *c, const struct inode *inode);
+int ubifs_jnl_xrename(struct ubifs_info *c, const struct inode *fst_dir,
+		      const struct dentry *fst_dentry,
+		      const struct inode *snd_dir,
+		      const struct dentry *snd_dentry, int sync);
 int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
 		     const struct dentry *old_dentry,
 		     const struct inode *new_dir,
-- 
2.7.3

[PATCH 4/4] ubifs: Use move variable in ubifs_rename()

[Richard Weinberger]

...to make the code more consistent since we use
move already in other places.

Signed-off-by: Richard Weinberger <richard@nod.at>
---
 fs/ubifs/journal.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/fs/ubifs/journal.c b/fs/ubifs/journal.c
index 8e07f13..91bc76dc 100644
--- a/fs/ubifs/journal.c
+++ b/fs/ubifs/journal.c
@@ -1100,7 +1100,7 @@ int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
 	aligned_dlen1 = ALIGN(dlen1, 8);
 	aligned_dlen2 = ALIGN(dlen2, 8);
 	len = aligned_dlen1 + aligned_dlen2 + ALIGN(ilen, 8) + ALIGN(plen, 8);
-	if (old_dir != new_dir)
+	if (move)
 		len += plen;
 	dent = kmalloc(len, GFP_NOFS);
 	if (!dent)
@@ -1216,7 +1216,7 @@ int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
 	if (err)
 		goto out_ro;
 
-	if (old_dir != new_dir) {
+	if (move) {
 		offs += ALIGN(plen, 8);
 		ino_key_init(c, &key, new_dir->i_ino);
 		err = ubifs_tnc_add(c, &key, lnum, offs, plen);
-- 
2.7.3

Re: [PATCH v2 0/4] ubifs: Add overlayfs support

[Felix Fietkau]

On 2016-09-14 22:28, Richard Weinberger wrote:
> To support overlayfs in a proper way a filesystem has to offer support
> for DT_CHR, RENAME_WHITEOUT and RENAME_EXCHANGE.
> While UBIFS supports DT_CHR it lacks support for RENAME_WHITEOUT and
> RENAME_EXCHANGE. This patch series implement the missing bits.
> RENAME_WHITEOUT itself depends on O_TMPFILE, therefore O_TMPFILE is now
> also supported.
> 
> Changes since v1:
>  - Build fix for CONFIG_LOCKDEP=y
>  - Fix for wrong dentry name length calculation in ubifs_jnl_xrename()
>  - Fix for wrong directory link count after RENAME_EXCHANGE
> 
> Richard Weinberger (4):
>   ubifs: Implement O_TMPFILE
>   ubifs: Implement RENAME_WHITEOUT
>   ubifs: Implement RENAME_EXCHANGE
>   ubifs: Use move variable in ubifs_rename()
Tested-by: Felix Fietkau <nbd@nbd.name>