[PATCH v3 0/5] overlayfs: support freeze/thaw

[PATCH v3 0/5] overlayfs: support freeze/thaw

[Amir Goldstein]

Miklos,

Patches 1-3 address your review comments and some
issues I found with fallocate() and copy_file_range()
helpers.
They are supposed to replace the patch already in overlayfs-next
97e147358bea ("vfs: wrap write f_ops with file_{start,end}_write()")

Let me know if you want me to rebase them on top of the already
committed change.

Patches 4-5 implement overlayfs freeze and are more of an RFC
at this point, but they are tested.

v3:
- Verify regular file before calling file_start_write()

v2:
- file_start_write() gets freeze protection on both upper and overlay
- convert file ops wraped with sb_start_write() to file_start_write()

v1:
- freeze upper on freeze of overlay

Amir Goldstein (5):
  vfs: deny fallocate() on directory
  vfs: deny copy_file_range() for non regular files
  vfs: wrap write f_ops with file_{start,end}_write()
  vfs: freeze protect overlay inode on file_start_write()
  ovl: support freeze/thaw super

 fs/open.c            | 14 ++++++--------
 fs/overlayfs/super.c | 12 ++++++++++++
 fs/read_write.c      |  9 +++++++--
 include/linux/fs.h   | 43 +++++++++++++++++++++++++++++--------------
 4 files changed, 54 insertions(+), 24 deletions(-)

-- 
2.7.4

[PATCH v3 1/5] vfs: deny fallocate() on directory

[Amir Goldstein]

There was an obscure use case of fallocate of directory inode
in the vfs helper with the comment:
"Let individual file system decide if it supports preallocation
 for directories or not."

But there is no in-tree file system that implements fallocate
for directory operations.

Deny an attempt to fallocate a directory with EISDIR error.

This change is needed prior to converting sb_start_write()
to  file_start_write(), so freeze protection is correctly
handled for cases of fallocate file and blockdev.

Cc: linux-api@vger.kernel.org
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
---
 fs/open.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/fs/open.c b/fs/open.c
index 9921f70..f9118f4 100644
--- a/fs/open.c
+++ b/fs/open.c
@@ -301,12 +301,10 @@ int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 	if (S_ISFIFO(inode->i_mode))
 		return -ESPIPE;
 
-	/*
-	 * Let individual file system decide if it supports preallocation
-	 * for directories or not.
-	 */
-	if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode) &&
-	    !S_ISBLK(inode->i_mode))
+	if (S_ISDIR(inode->i_mode))
+		return -EISDIR;
+
+	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 		return -ENODEV;
 
 	/* Check for wrap through zero too */
-- 
2.7.4

Re: [PATCH v3 1/5] vfs: deny fallocate() on directory

[Christoph Hellwig]

Looks fine,

Reviewed-by: Christoph Hellwig <hch@lst.de>

[PATCH v3 2/5] vfs: deny copy_file_range() for non regular files

[Amir Goldstein]

There is no in-tree file system that implements copy_file_range()
for non regular files.

Deny an attempt to copy_file_range() a directory with EISDIR
and any other non regualr file with EINVAL to conform with
behavior of vfs_{clone,dedup}_file_range().

This change is needed prior to converting sb_start_write()
to  file_start_write() in the vfs helper.

Cc: linux-api@vger.kernel.org
Cc: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
---
 fs/read_write.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/fs/read_write.c b/fs/read_write.c
index 5816d4c..511178d 100644
--- a/fs/read_write.c
+++ b/fs/read_write.c
@@ -1518,6 +1518,11 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
 	if (flags != 0)
 		return -EINVAL;
 
+	if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
+		return -EISDIR;
+	if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
+		return -EINVAL;
+
 	ret = rw_verify_area(READ, file_in, &pos_in, len);
 	if (unlikely(ret))
 		return ret;
-- 
2.7.4

Re: [PATCH v3 2/5] vfs: deny copy_file_range() for non regular files

[Christoph Hellwig]

Looks fine,

Reviewed-by: Christoph Hellwig <hch@lst.de>

[PATCH v3 3/5] vfs: wrap write f_ops with file_{start,end}_write()

[Amir Goldstein]

Before calling write f_ops, call file_start_write() instead
of sb_start_write().

Replace {sb,file}_start_write() for {copy,clone}_file_range() and
for fallocate().

Beyond correct semantics, this avoids freeze protection to sb when
operating on special inodes, such as fallocate() on a blockdev.

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
---
 fs/open.c          |  4 ++--
 fs/read_write.c    |  4 ++--
 include/linux/fs.h | 26 +++++++++++++-------------
 3 files changed, 17 insertions(+), 17 deletions(-)

diff --git a/fs/open.c b/fs/open.c
index f9118f4..949cef2 100644
--- a/fs/open.c
+++ b/fs/open.c
@@ -314,7 +314,7 @@ int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 	if (!file->f_op->fallocate)
 		return -EOPNOTSUPP;
 
-	sb_start_write(inode->i_sb);
+	file_start_write(file);
 	ret = file->f_op->fallocate(file, mode, offset, len);
 
 	/*
@@ -327,7 +327,7 @@ int vfs_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
 	if (ret == 0)
 		fsnotify_modify(file);
 
-	sb_end_write(inode->i_sb);
+	file_end_write(file);
 	return ret;
 }
 EXPORT_SYMBOL_GPL(vfs_fallocate);
diff --git a/fs/read_write.c b/fs/read_write.c
index 511178d..820a3f0 100644
--- a/fs/read_write.c
+++ b/fs/read_write.c
@@ -1543,7 +1543,7 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
 	if (len == 0)
 		return 0;
 
-	sb_start_write(inode_out->i_sb);
+	file_start_write(file_out);
 
 	/*
 	 * Try cloning first, this is supported by more file systems, and
@@ -1579,7 +1579,7 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,
 	inc_syscr(current);
 	inc_syscw(current);
 
-	sb_end_write(inode_out->i_sb);
+	file_end_write(file_out);
 
 	return ret;
 }
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 4a7f3cc..78c81e6 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -1741,19 +1741,6 @@ extern int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
 extern int vfs_dedupe_file_range(struct file *file,
 				 struct file_dedupe_range *same);
 
-static inline int do_clone_file_range(struct file *file_in, loff_t pos_in,
-				      struct file *file_out, loff_t pos_out,
-				      u64 len)
-{
-	int ret;
-
-	sb_start_write(file_inode(file_out)->i_sb);
-	ret = vfs_clone_file_range(file_in, pos_in, file_out, pos_out, len);
-	sb_end_write(file_inode(file_out)->i_sb);
-
-	return ret;
-}
-
 struct super_operations {
    	struct inode *(*alloc_inode)(struct super_block *sb);
 	void (*destroy_inode)(struct inode *);
@@ -2564,6 +2551,19 @@ static inline void file_end_write(struct file *file)
 	__sb_end_write(file_inode(file)->i_sb, SB_FREEZE_WRITE);
 }
 
+static inline int do_clone_file_range(struct file *file_in, loff_t pos_in,
+				      struct file *file_out, loff_t pos_out,
+				      u64 len)
+{
+	int ret;
+
+	file_start_write(file_out);
+	ret = vfs_clone_file_range(file_in, pos_in, file_out, pos_out, len);
+	file_end_write(file_out);
+
+	return ret;
+}
+
 /*
  * get_write_access() gets write permission for a file.
  * put_write_access() releases this write permission.
-- 
2.7.4

Re: [PATCH v3 3/5] vfs: wrap write f_ops with file_{start,end}_write()

[Christoph Hellwig]

Looks fine,

Reviewed-by: Christoph Hellwig <hch@lst.de>

[PATCH v3 4/5] vfs: freeze protect overlay inode on file_start_write()

[Amir Goldstein]

Before writing to overlay file, need to check if either overlay
or upper fs are frozen.
This allows freezing overlayfs mount independently of freezing
underlying fs.

Signed-off-by: Amir Goldstein <amir73il@gmail.com>
---
 include/linux/fs.h | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)

diff --git a/include/linux/fs.h b/include/linux/fs.h
index 78c81e6..8a4dbc6 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -2534,14 +2534,26 @@ static inline void file_start_write(struct file *file)
 {
 	if (!S_ISREG(file_inode(file)->i_mode))
 		return;
+	__sb_start_write(locks_inode(file)->i_sb, SB_FREEZE_WRITE, true);
+	if (likely(locks_inode(file) == file_inode(file)))
+		return;
 	__sb_start_write(file_inode(file)->i_sb, SB_FREEZE_WRITE, true);
+
 }
 
 static inline bool file_start_write_trylock(struct file *file)
 {
+	int ret;
+
 	if (!S_ISREG(file_inode(file)->i_mode))
 		return true;
-	return __sb_start_write(file_inode(file)->i_sb, SB_FREEZE_WRITE, false);
+	ret = __sb_start_write(locks_inode(file)->i_sb, SB_FREEZE_WRITE, false);
+	if (!ret || likely(locks_inode(file) == file_inode(file)))
+		return ret;
+	ret = __sb_start_write(file_inode(file)->i_sb, SB_FREEZE_WRITE, false);
+	if (!ret)
+		__sb_end_write(locks_inode(file)->i_sb, SB_FREEZE_WRITE);
+	return ret;
 }
 
 static inline void file_end_write(struct file *file)
@@ -2549,6 +2561,9 @@ static inline void file_end_write(struct file *file)
 	if (!S_ISREG(file_inode(file)->i_mode))
 		return;
 	__sb_end_write(file_inode(file)->i_sb, SB_FREEZE_WRITE);
+	if (likely(locks_inode(file) == file_inode(file)))
+		return;
+	__sb_end_write(locks_inode(file)->i_sb, SB_FREEZE_WRITE);
 }
 
 static inline int do_clone_file_range(struct file *file_in, loff_t pos_in,
-- 
2.7.4

[PATCH v3 5/5] ovl: support freeze/thaw super

[Amir Goldstein]

Signed-off-by: Amir Goldstein <amir73il@gmail.com>
---
 fs/overlayfs/super.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/fs/overlayfs/super.c b/fs/overlayfs/super.c
index 346f668..cb85eeb 100644
--- a/fs/overlayfs/super.c
+++ b/fs/overlayfs/super.c
@@ -179,6 +179,16 @@ static int ovl_sync_fs(struct super_block *sb, int wait)
 	return ret;
 }
 
+static int ovl_freeze(struct super_block *sb)
+{
+	return 0;
+}
+
+static int ovl_unfreeze(struct super_block *sb)
+{
+	return 0;
+}
+
 /**
  * ovl_statfs
  * @sb: The overlayfs super block
@@ -242,6 +252,8 @@ static int ovl_remount(struct super_block *sb, int *flags, char *data)
 static const struct super_operations ovl_super_operations = {
 	.put_super	= ovl_put_super,
 	.sync_fs	= ovl_sync_fs,
+	.freeze_fs	= ovl_freeze,
+	.unfreeze_fs	= ovl_unfreeze,
 	.statfs		= ovl_statfs,
 	.show_options	= ovl_show_options,
 	.remount_fs	= ovl_remount,
-- 
2.7.4