---- 在 星期五, 2021-11-26 17:10:07 Jan Kara [off-list ref] 撰写 ----
 > On Mon 22-11-21 11:00:33, Chengguang Xu wrote:
 > > From: Chengguang Xu [off-list ref]
 > > 
 > > We simply mark overlayfs inode dirty when it has upper,
 > > it's much simpler than mark dirtiness on modification.
 > > 
 > > Signed-off-by: Chengguang Xu [off-list ref]
 > > ---
 > >  fs/overlayfs/inode.c | 4 +++-
 > >  fs/overlayfs/util.c  | 1 +
 > >  2 files changed, 4 insertions(+), 1 deletion(-)
 > > 
 > > diff --git a/fs/overlayfs/inode.c b/fs/overlayfs/inode.c
 > > index 1f36158c7dbe..027ffc0a2539 100644
 > > --- a/fs/overlayfs/inode.c
 > > +++ b/fs/overlayfs/inode.c
 > > @@ -778,8 +778,10 @@ void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip,
 > >  {
 > >      struct inode *realinode;
 > >  
 > > -    if (oip->upperdentry)
 > > +    if (oip->upperdentry) {
 > >          OVL_I(inode)->__upperdentry = oip->upperdentry;
 > > +        mark_inode_dirty(inode);
 > > +    }
 > >      if (oip->lowerpath && oip->lowerpath->dentry)
 > >          OVL_I(inode)->lower = igrab(d_inode(oip->lowerpath->dentry));
 > >      if (oip->lowerdata)
 > 
 > Hum, does this get called only for inodes with upper inode existing? I
 > suppose we do not need to track inodes that were not copied up because they
 > cannot be dirty? I'm sorry, my knowledge of overlayfs is rather limited so
 > I may be missing something basic.
 > 

Well, as long as overly inode has upper it can be modified without copy-up,
so we need to track all overlay inodes which have upper inode.

Thanks,
Chengguang