Re: [syzbot] [kernfs?] possible deadlock in kernfs_fop_llseek

[Amir Goldstein]

On Sat, Apr 6, 2024 at 7:09 AM Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
>
> On Thu, Apr 04, 2024 at 12:33:40PM +0300, Amir Goldstein wrote:
>
> > We do not (anymore) lock ovl inode in ovl_llseek(), see:
> > b1f9d3858f72 ovl: use ovl_inode_lock in ovl_llseek()
> > but ovl inode is held in operations (e.g. ovl_rename)
> > which trigger copy up and call vfs_llseek() on the lower file.
>
> OK, but why do we bother with ovl_inode_lock() there?
> Note that serialization on struct file level is provided
> on syscall level - see call of fdget_pos() in there.
> IOW, which object are you protecting?  If it's struct file
> passed your way, you should already have the serialization.
> If it's underlying file on disk, that's up to vfs_llseek().

You're right.

> Exclusion with copyup by a different operation?

Nah, don't see how this is relevant to file->f_pos.

>
> I'm not saying it's wrong - it's just that the thing is
> tricky enough, so some clarification might be a good idea.

I think I just used inode_lock() in
9e46b840c705 ("ovl: support stacked SEEK_HOLE/SEEK_DATA")
as a common coding pattern in overlayfs when protecting the
"master" copy of overlay inode attributes, but it was not needed
for file->f_pos.

Miklos, please ack that I am not missing anything and that
ovl_inode_lock() is indeed redundant in ovl_llseek().

Anyway, this lock is not part of the lockdep issue that started this thread.

Thanks,
Amir.