Re: [PATCH mptcp-next 1/4] bpf: Add mptcp_subflow bpf_iter

[Geliang Tang <geliang@kernel.org>]

Hi Andrii,

On Fri, 2024-09-06 at 14:29 -0700, Andrii Nakryiko wrote:
> On Thu, Sep 5, 2024 at 11:25 AM Martin KaFai Lau
> <martin.lau@linux.dev> wrote:
> > 
> > On 9/5/24 6:52 AM, Geliang Tang wrote:
> > > From: Geliang Tang <tanggeliang@kylinos.cn>
> > > 
> > > It's necessary to traverse all subflows on the conn_list of an
> > > MPTCP
> > > socket and then call kfunc to modify the fields of each subflow.
> > > In
> > > kernel space, mptcp_for_each_subflow() helper is used for this:
> > > 
> > >   mptcp_for_each_subflow(msk, subflow)
> > >           kfunc(subflow);
> > > 
> > > But in the MPTCP BPF program, this has not yet been implemented.
> > > As
> > > Martin suggested recently, this conn_list walking + modify-by-
> > > kfunc
> > > usage fits the bpf_iter use case.
> > > 
> > > This patch adds a new bpf_iter type named "mptcp_subflow" to do
> > > this.
> > > 
> > > Suggested-by: Martin KaFai Lau <martin.lau@kernel.org>
> > > Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
> > > ---
> > >   kernel/bpf/helpers.c |  3 +++
> > >   net/mptcp/bpf.c      | 57
> > > ++++++++++++++++++++++++++++++++++++++++++++
> > >   2 files changed, 60 insertions(+)
> > > 
> > > diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> > > index b5f0adae8293..2340ba967444 100644
> > > --- a/kernel/bpf/helpers.c
> > > +++ b/kernel/bpf/helpers.c
> > > @@ -3023,6 +3023,9 @@ BTF_ID_FLAGS(func, bpf_preempt_enable)
> > >   BTF_ID_FLAGS(func, bpf_iter_bits_new, KF_ITER_NEW)
> > >   BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT |
> > > KF_RET_NULL)
> > >   BTF_ID_FLAGS(func, bpf_iter_bits_destroy, KF_ITER_DESTROY)
> > > +BTF_ID_FLAGS(func, bpf_iter_mptcp_subflow_new)
> > > +BTF_ID_FLAGS(func, bpf_iter_mptcp_subflow_next)
> > > +BTF_ID_FLAGS(func, bpf_iter_mptcp_subflow_destroy)
> > >   BTF_KFUNCS_END(common_btf_ids)
> > > 
> > >   static const struct btf_kfunc_id_set common_kfunc_set = {
> > > diff --git a/net/mptcp/bpf.c b/net/mptcp/bpf.c
> > > index 9672a70c24b0..cda09bbfd617 100644
> > > --- a/net/mptcp/bpf.c
> > > +++ b/net/mptcp/bpf.c
> > > @@ -204,6 +204,63 @@ static const struct btf_kfunc_id_set
> > > bpf_mptcp_fmodret_set = {
> > >       .set   = &bpf_mptcp_fmodret_ids,
> > >   };
> > > 
> > > +struct bpf_iter__mptcp_subflow {
> > > +     __bpf_md_ptr(struct bpf_iter_meta *, meta);
> > > +     __bpf_md_ptr(struct mptcp_sock *, msk);
> > > +     __bpf_md_ptr(struct list_head *, pos);
> > > +};
> > > +
> > > +DEFINE_BPF_ITER_FUNC(mptcp_subflow, struct bpf_iter_meta *meta,
> > > +                  struct mptcp_sock *msk, struct list_head *pos)
> 
> this is defining BPF iterator *program type* (effectively), which is
> different from open-coded iterator. Do you need a BPF iterator
> program
> type for this? Or open-coded iterator called from other BPF program
> types would be sufficient?

Yes, no need to define DEFINE_BPF_ITER_FUNC here, will drop it in v2.

> 
> > > +
> > > +struct bpf_iter_mptcp_subflow {
> > > +     __u64 __opaque[3];
> > > +} __attribute__((aligned(8)));
> > > +
> > > +struct bpf_iter_mptcp_subflow_kern {
> > > +     struct mptcp_sock *msk;
> > > +     struct list_head *pos;
> > > +} __attribute__((aligned(8)));
> 
> opaque[3], but you are using two pointers here. Why the difference?

Should be 2, not 3. will update in v2.

Thanks,
-Geliang

> 
> > > +
> > > +__bpf_kfunc_start_defs();
> > > +
> > > +__bpf_kfunc int bpf_iter_mptcp_subflow_new(struct
> > > bpf_iter_mptcp_subflow *it,
> > > +                                        struct mptcp_sock *msk)
> > > +{
> > > +     struct bpf_iter_mptcp_subflow_kern *kit = (void *)it;
> > > +
> > > +     kit->msk = msk;
> > > +     kit->pos = &msk->conn_list;
> > > +     spin_lock_bh(&msk->pm.lock);
> > 
> > I don't think spin_lock here without unlock can be used. e.g. What
> > if
> > bpf_iter_mptcp_subflow_new() is called twice back-to-back.
> > 
> > I haven't looked at the mptcp details, some questions:
> > The list is protected by msk->pm.lock?
> > What happen to the sk_lock of the msk?
> > Can this be rcu-ify? or it needs some cares when walking the
> > established TCP
> > subflow?
> > 
> > 
> > [ Please cc the bpf list. Helping to review patches is a good way
> > to contribute
> > back to the mailing list. ]
> > 
> > > +
> > > +     return 0;
> > > +}
> > > +
> > > +__bpf_kfunc struct mptcp_subflow_context *
> > > +bpf_iter_mptcp_subflow_next(struct bpf_iter_mptcp_subflow *it)
> > > +{
> > > +     struct bpf_iter_mptcp_subflow_kern *kit = (void *)it;
> > > +     struct mptcp_subflow_context *subflow;
> > > +     struct mptcp_sock *msk = kit->msk;
> > > +
> > > +     subflow = list_entry((kit->pos)->next, struct
> > > mptcp_subflow_context, node);
> > > +     if (list_entry_is_head(subflow, &msk->conn_list, node))
> > > +             return NULL;
> > > +
> > > +     kit->pos = &subflow->node;
> > > +     return subflow;
> > > +}
> > > +
> > > +__bpf_kfunc void bpf_iter_mptcp_subflow_destroy(struct
> > > bpf_iter_mptcp_subflow *it)
> > > +{
> > > +     struct bpf_iter_mptcp_subflow_kern *kit = (void *)it;
> > > +     struct mptcp_sock *msk = kit->msk;
> > > +
> > > +     spin_unlock_bh(&msk->pm.lock);
> > > +}
> > > +
> > > +__bpf_kfunc_end_defs();
> > > +
> > >   __diag_push();
> > >   __diag_ignore_all("-Wmissing-prototypes",
> > >                 "kfuncs which will be used in BPF programs");
> > 
> >