Re: [PATCH v14 00/12] FUSE passthrough for file io

From: Amir Goldstein <amir73il@gmail.com>
To: Miklos Szeredi <miklos@szeredi.hu>
Cc: Bernd Schubert <bernd.schubert@fastmail.fm>,
	Daniel Rosenberg <drosen@google.com>,
	 Paul Lawrence <paullawrence@google.com>,
	Alessio Balsini <balsini@android.com>,
	 Christian Brauner <brauner@kernel.org>,
	fuse-devel@lists.sourceforge.net,  linux-fsdevel@vger.kernel.org,
	Dave Chinner <david@fromorbit.com>
Subject: Re: [PATCH v14 00/12] FUSE passthrough for file io
Date: Wed, 29 Nov 2023 17:52:05 +0200	[thread overview]
Message-ID: <CAOQ4uxgW6xpWW=jLQJuPKOCxN=i_oNeRwNnMEpxOhVD7RVwHHw@mail.gmail.com> (raw)
In-Reply-To: <CAJfpegtH1DP19cAuKgYAssZ8nkKhnyX42AYWtAT3h=nmi2j31A@mail.gmail.com>

On Wed, Nov 29, 2023 at 5:21 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
>
> On Wed, 29 Nov 2023 at 16:06, Amir Goldstein <amir73il@gmail.com> wrote:
> >
> > On Wed, Nov 29, 2023 at 4:14 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
> > >
> > > On Wed, 29 Nov 2023 at 08:25, Amir Goldstein <amir73il@gmail.com> wrote:
> > >
> > > > My proposed solution is to change the semantics with the init flag
> > > > FUSE_PASSTHROUGH to disallow mmap on FOPEN_DIRECT_IO
> > > > files.
> > >
> > > Why?  FOPEN_DIRECT_IO and FUSE_PASSTHROUGH should mix much more
> > > readily than FOPEN_DIRECT_IO with page cache.
> > >
> >
> > Am I misunderstanding how mmap works with FOPEN_DIRECT_IO file?
> > My understanding is that mmap of FOPEN_DIRECT_IO reads/writes
> > from page cache of fuse inode.
> >
> > To clarify, the plan is to never allow mixing open of passthrough and
> > cached files on the same inode.
>
> Yep.
>
> But passthrough mmap + direct I/O should work, no?
>

direct I/O read()/write() is never a problem.

The question is whether mmap() on a file opened with FOPEN_DIRECT_IO
when the inode is in passthrough mode, also uses fuse_passthrough_mmap()?
or denied, similar to how mmap with ff->open_flags & FOPEN_DIRECT_IO &&
vma->vm_flags & VM_MAYSHARE) && !fc->direct_io_relax
is denied?

> > It is allowed to open FOPEN_DIRECT_IO file for inode either in cached
> > or passthrough mode, but it is NOT allowed to mmap a FOPEN_DIRECT_IO
> > file for inode in passthrough mode.
> >
> > However, if inode only has file open in FOPEN_DIRECT_IO mode, then inode
> > mode is neutral. If we allow mmap in this state then a later open in passthourgh
> > mode and mmap in passthrough mode will collide with the direct mode mmap.
>
> We can keep track of when there are any page cache mmaps.  Does that
> not solve this?
>

A bit more challenging, because we will need to track unmounts, or at
least track
"was_cached_mmaped" state per file, but doable.

> >
> > Therefore, my proposal is than when filesystem is FUSE_PASSTHROUGH capable,
> > only passthrough file and cached file may be mmaped, but never allow to
> > mmap a FOPEN_DIRECT_IO file.
> >
> > Does that make sense?
>
> I'm not sure I understand how this is supposed to work.   Disallowing
> mmap will break applications.

How is this different from existing -ENODEV in fuse_file_mmap() for
!fc->direct_io_relax? What am I missing?
Is it because VM_MAYSHARE is far less common for applications?

Thanks,
Amir.