Date:   Wed, 26 Aug 2020 20:17:11 +0100
From:   "Dr. David Alan Gilbert" <dgilbert@...hat.com>
To:     Stefan Hajnoczi <stefanha@...hat.com>
Cc:     Vivek Goyal <vgoyal@...hat.com>,
        Miklos Szeredi <miklos@...redi.hu>,
        linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org,
        linux-nvdimm <linux-nvdimm@...ts.01.org>,
        virtio-fs-list <virtio-fs@...hat.com>,
        Dan Williams <dan.j.williams@...el.com>
Subject: Re: [PATCH v3 11/18] fuse: implement FUSE_INIT map_alignment field

* Stefan Hajnoczi (stefanha@...hat.com) wrote:
> On Wed, Aug 26, 2020 at 11:51:42AM -0400, Vivek Goyal wrote:
> > On Wed, Aug 26, 2020 at 04:06:35PM +0200, Miklos Szeredi wrote:
> > > On Thu, Aug 20, 2020 at 12:21 AM Vivek Goyal <vgoyal@...hat.com> wrote:
> > > >
> > > > The device communicates FUSE_SETUPMAPPING/FUSE_REMOVMAPPING alignment
> > > > constraints via the FUST_INIT map_alignment field.  Parse this field and
> > > > ensure our DAX mappings meet the alignment constraints.
> > > >
> > > > We don't actually align anything differently since our mappings are
> > > > already 2MB aligned.  Just check the value when the connection is
> > > > established.  If it becomes necessary to honor arbitrary alignments in
> > > > the future we'll have to adjust how mappings are sized.
> > > >
> > > > The upshot of this commit is that we can be confident that mappings will
> > > > work even when emulating x86 on Power and similar combinations where the
> > > > host page sizes are different.
> > > >
> > > > Signed-off-by: Stefan Hajnoczi <stefanha@...hat.com>
> > > > Signed-off-by: Vivek Goyal <vgoyal@...hat.com>
> > > > ---
> > > >  fs/fuse/fuse_i.h          |  5 ++++-
> > > >  fs/fuse/inode.c           | 18 ++++++++++++++++--
> > > >  include/uapi/linux/fuse.h |  4 +++-
> > > >  3 files changed, 23 insertions(+), 4 deletions(-)
> > > >
> > > > diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
> > > > index 478c940b05b4..4a46e35222c7 100644
> > > > --- a/fs/fuse/fuse_i.h
> > > > +++ b/fs/fuse/fuse_i.h
> > > > @@ -47,7 +47,10 @@
> > > >  /** Number of dentries for each connection in the control filesystem */
> > > >  #define FUSE_CTL_NUM_DENTRIES 5
> > > >
> > > > -/* Default memory range size, 2MB */
> > > > +/*
> > > > + * Default memory range size.  A power of 2 so it agrees with common FUSE_INIT
> > > > + * map_alignment values 4KB and 64KB.
> > > > + */
> > > >  #define FUSE_DAX_SZ    (2*1024*1024)
> > > >  #define FUSE_DAX_SHIFT (21)
> > > >  #define FUSE_DAX_PAGES (FUSE_DAX_SZ/PAGE_SIZE)
> > > > diff --git a/fs/fuse/inode.c b/fs/fuse/inode.c
> > > > index b82eb61d63cc..947abdd776ca 100644
> > > > --- a/fs/fuse/inode.c
> > > > +++ b/fs/fuse/inode.c
> > > > @@ -980,9 +980,10 @@ static void process_init_reply(struct fuse_conn *fc, struct fuse_args *args,
> > > >  {
> > > >         struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
> > > >         struct fuse_init_out *arg = &ia->out;
> > > > +       bool ok = true;
> > > >
> > > >         if (error || arg->major != FUSE_KERNEL_VERSION)
> > > > -               fc->conn_error = 1;
> > > > +               ok = false;
> > > >         else {
> > > >                 unsigned long ra_pages;
> > > >
> > > > @@ -1045,6 +1046,13 @@ static void process_init_reply(struct fuse_conn *fc, struct fuse_args *args,
> > > >                                         min_t(unsigned int, FUSE_MAX_MAX_PAGES,
> > > >                                         max_t(unsigned int, arg->max_pages, 1));
> > > >                         }
> > > > +                       if ((arg->flags & FUSE_MAP_ALIGNMENT) &&
> > > > +                           (FUSE_DAX_SZ % (1ul << arg->map_alignment))) {
> > > 
> > > This just obfuscates "arg->map_alignment != FUSE_DAX_SHIFT".
> > > 
> > > So the intention was that userspace can ask the kernel for a
> > > particular alignment, right?
> > 
> > My understanding is that device will specify alignment for
> > the foffset/moffset fields in fuse_setupmapping_in/fuse_removemapping_one.
> > And DAX mapping can be any size meeting that alignment contraint.
> > 
> > > 
> > > In that case kernel can definitely succeed if the requested alignment
> > > is smaller than the kernel provided one, no? 
> > 
> > Yes. So if map_alignemnt is 64K and DAX mapping size is 2MB, that's just
> > fine because it meets 4K alignment contraint. Just that we can't use
> > 4K size DAX mapping in that case.
> > 
> > > It would also make
> > > sense to make this a two way negotiation.  I.e. send the largest
> > > alignment (FUSE_DAX_SHIFT in this implementation) that the kernel can
> > > provide in fuse_init_in.   In that case the only error would be if
> > > userspace ignored the given constraints.
> > 
> > We could make it two way negotiation if it helps. So if we support
> > multiple mapping sizes in future, say 4K, 64K, 2MB, 1GB. So idea is
> > to send alignment of largest mapping size to device/user_space (1GB)
> > in this case? And that will allow device to choose an alignment
> > which best fits its needs?
> > 
> > But problem here is that sending (log2(1GB)) does not mean we support
> > all the alignments in that range. For example, if device selects say
> > 256MB as minimum alignment, kernel might not support it.
> > 
> > So there seem to be two ways to handle this.
> > 
> > A.Let device be conservative and always specify the minimum aligment
> >   it can work with and let guest kernel automatically choose a mapping
> >   size which meets that min_alignment contraint.
> > 
> > B.Send all the mapping sizes supported by kernel to device and then
> >   device chooses an alignment as it sees fit. We could probably send
> >   a 64bit field and set a bit for every size we support as dax mapping.
> >   If we were to go down this path, I think in that case client should
> >   respond back with exact mapping size we should use (and not with
> >   minimum alignment).
> > 
> > I thought intent behind this patch was to implement A.
> > 
> > Stefan/David, this patch came from you folks. What do you think?
> 
> Yes, I agree with Vivek.
> 
> The FUSE server is telling the client the minimum alignment for
> foffset/moffset. The client can map any size it likes as long as
> foffset/moffset meet the alignment constraint. I can't think of a reason
> to do two-way negotiation.

Agreed, because there's not much that the server can do about it if the
client would like a smaller granularity - the servers granularity might
be dictated by it's mmap/pagesize/filesystem.  If the client wants a
larger granularity that's it's choice when it sends the setupmapping
calls.

Dave

> Stefan


-- 
Dr. David Alan Gilbert / dgilbert@...hat.com / Manchester, UK

Powered by blists - more mailing lists