[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAF=yD-+si68K5YhHW1omgSEVUw7yFpwo8ku2e-+PXT5xJgxAsQ@mail.gmail.com>
Date: Tue, 22 Dec 2020 13:29:23 -0500
From: Willem de Bruijn <willemdebruijn.kernel@...il.com>
To: Jonathan Lemon <jonathan.lemon@...il.com>
Cc: Network Development <netdev@...r.kernel.org>,
Eric Dumazet <edumazet@...gle.com>,
Kernel Team <kernel-team@...com>
Subject: Re: [PATCH 09/12 v2 RFC] skbuff: add zc_flags to ubuf_info for ubuf setup
On Tue, Dec 22, 2020 at 1:17 PM Jonathan Lemon <jonathan.lemon@...il.com> wrote:
>
> On Tue, Dec 22, 2020 at 10:00:37AM -0500, Willem de Bruijn wrote:
> > On Mon, Dec 21, 2020 at 7:09 PM Jonathan Lemon <jonathan.lemon@...il.com> wrote:
> > >
> > > From: Jonathan Lemon <bsd@...com>
> > >
> > > Currently, an ubuf is attached to a new skb, the skb zc_flags
> > > is initialized to a fixed value. Instead of doing this, set
> > > the default zc_flags in the ubuf, and have new skb's inherit
> > > from this default.
> > >
> > > This is needed when setting up different zerocopy types.
> > >
> > > Signed-off-by: Jonathan Lemon <jonathan.lemon@...il.com>
> > > ---
> > > include/linux/skbuff.h | 3 ++-
> > > net/core/skbuff.c | 1 +
> > > 2 files changed, 3 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
> > > index da0c1dddd0da..b90be4b0b2de 100644
> > > --- a/include/linux/skbuff.h
> > > +++ b/include/linux/skbuff.h
> > > @@ -478,6 +478,7 @@ struct ubuf_info {
> > > };
> > > };
> > > refcount_t refcnt;
> > > + u8 zc_flags;
> > >
> > > struct mmpin {
> > > struct user_struct *user;
> >
> > When allocating ubuf_info for msg_zerocopy, we actually allocate the
> > notification skb, to be sure that notifications won't be dropped due
> > to memory pressure at notification time. We actually allocate the skb
> > and place ubuf_info in skb->cb[].
> >
> > The struct is exactly 48 bytes on 64-bit platforms, filling all of cb.
> > This new field fills a 4B hole, so it should still be fine.
>
> Yes, I was careful not to increase the size. I have future changes
> for this structure, moving 'struct mmpin' into a union. Making the
> flags 16 bits shouldn't be a problem either.
>
>
> > Just being very explicit, as this is a fragile bit of code. Come to
> > think of it, this probably deserves a BUILD_BUG_ON.
>
> You mean like the one which exists in sock_zerocopy_alloc()?
>
> BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
Ah good. Yes. I should have remembered that was there.
Powered by blists - more mailing lists