Re: [RFC PATCH bpf-next 2/4] bpf: Introduce process open coded iterator kfuncs

[Chuyi Zhou]

Hello, Alexei.

在 2023/9/6 04:09, Alexei Starovoitov 写道:
> On Sun, Aug 27, 2023 at 12:21 AM Chuyi Zhou <zhouchuyi@xxxxxxxxxxxxx> wrote:
> > This patch adds kfuncs bpf_iter_process_{new,next,destroy} which allow
> > creation and manipulation of struct bpf_iter_process in open-coded iterator
> > style. BPF programs can use these kfuncs or through bpf_for_each macro to
> > iterate all processes in the system.
> >
> > Signed-off-by: Chuyi Zhou <zhouchuyi@xxxxxxxxxxxxx>
> > ---
> >   include/uapi/linux/bpf.h       |  4 ++++
> >   kernel/bpf/helpers.c           |  3 +++
> >   kernel/bpf/task_iter.c         | 31 +++++++++++++++++++++++++++++++
> >   tools/include/uapi/linux/bpf.h |  4 ++++
> >   tools/lib/bpf/bpf_helpers.h    |  5 +++++
> >   5 files changed, 47 insertions(+)
> >
> > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > index 2a6e9b99564b..cfbd527e3733 100644
> > --- a/include/uapi/linux/bpf.h
> > +++ b/include/uapi/linux/bpf.h
> > @@ -7199,4 +7199,8 @@ struct bpf_iter_css_task {
> >          __u64 __opaque[1];
> >   } __attribute__((aligned(8)));
> >
> > +struct bpf_iter_process {
> > +       __u64 __opaque[1];
> > +} __attribute__((aligned(8)));
> > +
> >   #endif /* _UAPI__LINUX_BPF_H__ */
> > diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> > index cf113ad24837..81a2005edc26 100644
> > --- a/kernel/bpf/helpers.c
> > +++ b/kernel/bpf/helpers.c
> > @@ -2458,6 +2458,9 @@ BTF_ID_FLAGS(func, bpf_iter_num_destroy, KF_ITER_DESTROY)
> >   BTF_ID_FLAGS(func, bpf_iter_css_task_new, KF_ITER_NEW)
> >   BTF_ID_FLAGS(func, bpf_iter_css_task_next, KF_ITER_NEXT | KF_RET_NULL)
> >   BTF_ID_FLAGS(func, bpf_iter_css_task_destroy, KF_ITER_DESTROY)
> > +BTF_ID_FLAGS(func, bpf_iter_process_new, KF_ITER_NEW)
> > +BTF_ID_FLAGS(func, bpf_iter_process_next, KF_ITER_NEXT | KF_RET_NULL)
> > +BTF_ID_FLAGS(func, bpf_iter_process_destroy, KF_ITER_DESTROY)
> >   BTF_ID_FLAGS(func, bpf_dynptr_adjust)
> >   BTF_ID_FLAGS(func, bpf_dynptr_is_null)
> >   BTF_ID_FLAGS(func, bpf_dynptr_is_rdonly)
> > diff --git a/kernel/bpf/task_iter.c b/kernel/bpf/task_iter.c
> > index b1bdba40b684..a6717a76c1e0 100644
> > --- a/kernel/bpf/task_iter.c
> > +++ b/kernel/bpf/task_iter.c
> > @@ -862,6 +862,37 @@ __bpf_kfunc void bpf_iter_css_task_destroy(struct bpf_iter_css_task *it)
> >          kfree(kit->css_it);
> >   }
> >
> > +struct bpf_iter_process_kern {
> > +       struct task_struct *tsk;
> > +} __attribute__((aligned(8)));
> > +
> > +__bpf_kfunc int bpf_iter_process_new(struct bpf_iter_process *it)
> > +{
> > +       struct bpf_iter_process_kern *kit = (void *)it;
> > +
> > +       BUILD_BUG_ON(sizeof(struct bpf_iter_process_kern) != sizeof(struct bpf_iter_process));
> > +       BUILD_BUG_ON(__alignof__(struct bpf_iter_process_kern) !=
> > +                                       __alignof__(struct bpf_iter_process));
> > +
> > +       rcu_read_lock();
> > +       kit->tsk = &init_task;
> > +       return 0;
> > +}
> > +
> > +__bpf_kfunc struct task_struct *bpf_iter_process_next(struct bpf_iter_process *it)
> > +{
> > +       struct bpf_iter_process_kern *kit = (void *)it;
> > +
> > +       kit->tsk = next_task(kit->tsk);
> > +
> > +       return kit->tsk == &init_task ? NULL : kit->tsk;
> > +}
> > +
> > +__bpf_kfunc void bpf_iter_process_destroy(struct bpf_iter_process *it)
> > +{
> > +       rcu_read_unlock();
> > +}
>
> This iter can be used in all ctx-s which is nice, but let's
> make the verifier enforce rcu_read_lock/unlock done by bpf prog
> instead of doing in the ctor/dtor of iter, since
> in sleepable progs the verifier won't recognize that body is RCU CS.
> We'd need to teach the verifier to allow bpf_iter_process_new()
> inside in_rcu_cs() and make sure there is no rcu_read_unlock
> while BPF_ITER_STATE_ACTIVE.
> bpf_iter_process_destroy() would become a nop.

Thanks for your review!

I think bpf_iter_process_{new, next, destroy} should be protected by 
bpf_rcu_read_lock/unlock explicitly whether the prog is sleepable or 
not, right? I'm not very familiar with the BPF verifier, but I believe 
there is still a risk in directly calling these kfuns even if 
in_rcu_cs() is true.

Maby what we actually need here is to enforce BPF verifier to check 
env->cur_state->active_rcu_lock is true when we want to call these kfuncs.

Thanks.