Re: [PATCH] fuse: enable larger read buffers for readdir.

[Jaco Kroon]

Hi,

On 2023/07/26 15:53, Bernd Schubert wrote:
> On 7/26/23 12:59, Jaco Kroon wrote:
> > Signed-off-by: Jaco Kroon <jaco@xxxxxxxxx>
> > ---
> >   fs/fuse/Kconfig   | 16 ++++++++++++++++
> >   fs/fuse/readdir.c | 42 ++++++++++++++++++++++++------------------
> >   2 files changed, 40 insertions(+), 18 deletions(-)
> >
> > diff --git a/fs/fuse/Kconfig b/fs/fuse/Kconfig
> > index 038ed0b9aaa5..0783f9ee5cd3 100644
> > --- a/fs/fuse/Kconfig
> > +++ b/fs/fuse/Kconfig
> > @@ -18,6 +18,22 @@ config FUSE_FS
> >         If you want to develop a userspace FS, or if you want to use
> >         a filesystem based on FUSE, answer Y or M.
> >   +config FUSE_READDIR_ORDER
> > +    int
> > +    range 0 5
> > +    default 5
> > +    help
> > +        readdir performance varies greatly depending on the size of 
> > the read.
> > +        Larger buffers results in larger reads, thus fewer reads and 
> > higher
> > +        performance in return.
> > +
> > +        You may want to reduce this value on seriously constrained 
> > memory
> > +        systems where 128KiB (assuming 4KiB pages) cache pages is 
> > not ideal.
> > +
> > +        This value reprents the order of the number of pages to 
> > allocate (ie,
> > +        the shift value).  A value of 0 is thus 1 page (4KiB) where 
> > 5 is 32
> > +        pages (128KiB).
> > +
>
> I like the idea of a larger readdir size, but shouldn't that be a 
> server/daemon/library decision which size to use, instead of kernel 
> compile time? So should be part of FUSE_INIT negotiation?

Yes sure, but there still needs to be a default.  And one page at a time 
doesn't cut it.

-- snip --

> >   -    page = alloc_page(GFP_KERNEL);
> > +    page = alloc_pages(GFP_KERNEL, READDIR_PAGES_ORDER);
>
> I guess that should become folio alloc(), one way or the other. Now I 
> think order 0 was chosen before to avoid risk of allocation failure. I 
> guess it might work to try a large size and to fall back to 0 when 
> that failed. Or fail back to the slower vmalloc.

If this varies then a bunch of other code will become somewhat more 
complex, especially if one alloc succeeds, and then a follow-up succeeds.

I'm not familiar with the differences between the different mechanisms 
available for allocation.

-- snip --

> Thanks,
My pleasure,
Jaco