[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAGWr4cTKgcrWKcNchdo6OKbKFTNaNc+jfPoKf+_qb+9W-nJT3g@mail.gmail.com>
Date: Thu, 2 Oct 2025 09:57:10 -0700
From: Octavian Purdila <tavip@...gle.com>
To: Maciej Fijalkowski <maciej.fijalkowski@...el.com>
Cc: Jakub Kicinski <kuba@...nel.org>, davem@...emloft.net, edumazet@...gle.com,
pabeni@...hat.com, horms@...nel.org, ast@...nel.org, daniel@...earbox.net,
hawk@...nel.org, john.fastabend@...il.com, sdf@...ichev.me, kuniyu@...gle.com,
aleksander.lobakin@...el.com, toke@...hat.com, lorenzo@...nel.org,
netdev@...r.kernel.org, bpf@...r.kernel.org,
syzbot+ff145014d6b0ce64a173@...kaller.appspotmail.com
Subject: Re: [PATCH net v2] xdp: update mem type when page pool is used for
generic XDP
On Thu, Oct 2, 2025 at 5:54 AM Maciej Fijalkowski
<maciej.fijalkowski@...el.com> wrote:
>
> On Wed, Oct 01, 2025 at 06:15:25PM -0700, Octavian Purdila wrote:
> > On Wed, Oct 1, 2025 at 12:06 PM Maciej Fijalkowski
> > <maciej.fijalkowski@...el.com> wrote:
> > >
> > > On Wed, Oct 01, 2025 at 08:27:37AM -0700, Jakub Kicinski wrote:
> > > > On Wed, 1 Oct 2025 16:42:29 +0200 Maciej Fijalkowski wrote:
> > > > > Here we piggy back on sk_buff::pp_recycle setting as it implies underlying
> > > > > memory is backed by page pool.
> > > >
> > > > skb->pp_recycle means that if the pages of the skb came from a pp then
> > > > the skb is holding a pp reference not a full page reference on those
> > > > pages. It does not mean that all pages of an skb came from pp.
> > > > In practice it may be equivalent, especially here. But I'm slightly
> > > > worried that checking pp_recycle will lead to confusion..
> > >
> > > Mmm ok - maybe that's safer and straight-forward?
> > >
> > > diff --git a/net/core/dev.c b/net/core/dev.c
> > > index 93a25d87b86b..7707a95ca8ed 100644
> > > --- a/net/core/dev.c
> > > +++ b/net/core/dev.c
> > > @@ -5269,6 +5269,9 @@ u32 bpf_prog_run_generic_xdp(struct sk_buff *skb, struct xdp_buff *xdp,
> > > orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
> > > orig_eth_type = eth->h_proto;
> > >
> > > + xdp->rxq->mem.type = page_pool_page_is_pp(virt_to_page(xdp->data)) ?
> > > + MEM_TYPE_PAGE_POOL : MEM_TYPE_PAGE_SHARED;
> > > +
> > > act = bpf_prog_run_xdp(xdp_prog, xdp);
> > >
> > > /* check if bpf_xdp_adjust_head was used */
> > >
> > > As you know we do not have that kind of granularity within xdp_buff where
> > > we could distinguish the memory provider per linear part and each frag...
> >
> > LGTM, based on my limited understanding. I can also confirm the syz
> > repro no longer crashes with this patch.
>
> Would you be OK with me submitting this fix and giving you the proper
> credit?
Yes, of course, thank you!
Powered by blists - more mailing lists