On Thu, Apr 6, 2017 at 11:13 AM, Amir Goldstein <amir73il@xxxxxxxxx> wrote: > ftruncate an overlayfs inode was wrongly freeze protecting the > overlay file system instead of the backing file system. > > Use file_start_write() instead of sb_start_write(), which > does the right thing and will also freeze protect both overlay and > backing fs when the time comes. > > Signed-off-by: Amir Goldstein <amir73il@xxxxxxxxx> > --- > fs/open.c | 4 ++-- > 1 file changed, 2 insertions(+), 2 deletions(-) > > diff --git a/fs/open.c b/fs/open.c > index 949cef2..53b1b33 100644 > --- a/fs/open.c > +++ b/fs/open.c > @@ -196,13 +196,13 @@ static long do_sys_ftruncate(unsigned int fd, loff_t length, int small) > if (IS_APPEND(inode)) > goto out_putf; > > - sb_start_write(inode->i_sb); > + file_start_write(f.file); Nah! this is wrong. Should be mnt_want_write_file(f.file) At the time of this patch mnt_want_write_file(f.file) translated to sb_start_write(inode->i_sb) but when next patch is going to change inode = file_inode(f.file) sb_start_write(inode->i_sb) will no longer be correct and mnt_want_write_file(f.file) is nicer than sb_start_write(locks_inode(f.file) > error = locks_verify_truncate(inode, f.file, length); > if (!error) > error = security_path_truncate(&f.file->f_path); > if (!error) > error = do_truncate(dentry, length, ATTR_MTIME|ATTR_CTIME, f.file); > - sb_end_write(inode->i_sb); > + file_end_write(f.file); > out_putf: > fdput(f); > out: > -- > 2.7.4 >
