Re: [PATCH] proc: remove mark_inode_dirty() in proc_notify_change()

[Chao Yu]

Hi Andrew,

Sorry for the long delay. :(

On 2023/1/13 6:43, Andrew Morton wrote:
> On Thu, 12 Jan 2023 11:27:20 +0800 Chao Yu <chao@xxxxxxxxxx> wrote:
>
> > proc_notify_change() has updated i_uid, i_gid and i_mode into proc
> > dirent, we don't need to call mark_inode_dirty() for later writeback,
> > remove it.
> >
> > --- a/fs/proc/generic.c
> > +++ b/fs/proc/generic.c
> > @@ -127,7 +127,6 @@ static int proc_notify_change(struct user_namespace *mnt_userns,
> >   		return error;
> >   
> >   	setattr_copy(&init_user_ns, inode, iattr);
> > -	mark_inode_dirty(inode);
> >   
> >   	proc_set_user(de, inode->i_uid, inode->i_gid);
> >   	de->mode = inode->i_mode;
>
> procfs call mark_inode_dirty() in three places.

Correct.

> Does mark_inode_dirty() of a procfs file actually serve any purpose?

I don't see any particular reason that procfs inode needs to be set dirty,
as an in-memory filesystem, there is no backing device, so all attributes
should have been updated into procfs dirent directly in .setattr().

In fact, also procfs doesn't implement .dirty_inode, .write_inode or
.writepage{,s} interfaces which serves delayed inode update, pages writeback
after inode is set as dirty.

Thanks,