Re: possible deadlock in pipe_lock (2)

[Miklos Szeredi <miklos@szeredi.hu>]

On Tue, Feb 12, 2019 at 2:39 PM Amir Goldstein <amir73il@gmail.com> wrote:
>
> > > My other thought is that perhaps sb_start_write() should invoke
> > > s_ops->start_write() so that overlay can do the freeze protection on
> > > the upper early.
> >
> > So my understanding of overlayfs is pretty basic so I'm sorry if I miss
> > something. If I'm right, we have three superblocks here: ovl, upper, lower.
> > Now 'lower' is read-only so for freezing purposes we can just forget about
> > it. 'upper' is where the real changes are going into and 'ovl' is a wrapper
> > virtual superblock that handles merging of 'lower' and 'upper'. Correct so
> > far?
>
> Yes.
>
> >
> > And the problem seems to be that when you acquire freeze protection for the
> > 'ovl' superblock, you in fact want to acquire freeze protection for the
> > 'upper' (as 'ovl' is just virtual and has no disk state to protect). So I
>
> There are use case for freezing ovl (i.e. ovl snapshots) but it is not
> implemented
> at the moment.
>
> Overlayfs already gets upper freeze protection internally before any
> modification
> to upper.
> The problem that locking order of upper freeze is currently under overlay
> inode mutex. And that brings a problem with the above pipe case.
>
> > agree that a callback to allow overlayfs to acquire freeze protection on
> > 'upper' right away would be one solution. Or we could make s_writers a
> > pointer and redirect ovl->s_writers to upper->s_writers. Then VFS should do
> > the right thing from the start unless overlayfs calls back into operations
> > on 'upper' that will try to acquire the freeze protection again. Thoughts?
>
> Overlayfs definitely calls into operations on upper and upper certainly
> acquires several levels of s_writers itself.

sb_start_write() calls cannot be nested (for the same reason two
shared locks may be part of a deadlock), so this would mean having to
make sure that none of the code that overlay calls does
sb_start_write() itself.  Which means quite a bit of vfs api churn and
the associated pain...

> The problem with the proposal to change locking order to
> ovl freeze -> upper freeze -> ovl inode -> upper inode
> is that for some non-write operations (e.g. lookup, readdir)
> overlay may end up updating xattrs on upper, so will need
> to take upper freeze after ovl inode lock without ovl freeze
> being called by vfs.
>
> I suggested that we may use upper freeze trylock in those
> cases and skip xattr update if trylock fails.
>
> Not sure if my assumption is correct that this would be ok
> w.r.t locking rules?

Yes, using trylock is always deadlock free.

Thanks,
Miklos