Re: [PATCH bpf-next 1/3] bpf: add bpf_progenyof helper

[Javier Honduvilla Coto <javierhonduco@fb.com>]

On Sat, Mar 02, 2019 at 12:01:14AM +0000, Martin Lau wrote:
> On Fri, Mar 01, 2019 at 09:28:39AM -0800, Javier Honduvilla Coto wrote:
> > On Wed, Feb 27, 2019 at 06:26:41AM +0000, Martin Lau wrote:
> > > On Tue, Feb 26, 2019 at 02:36:49PM -0800, Javier Honduvilla Coto wrote:
> > > > This patch adds the bpf_progenyof helper which receives a PID and returns
> > > What is progenof?
> > >
> > > > 1 if the process currently being executed is in the process hierarchy
> > > > including itself or 0 if not.
> > > >
> > > > This is very useful in tracing programs when we want to filter by a
> > > > given PID and all the children it might spawn. The current workarounds
> > > > most people implement for this purpose have issues:
> > > >
> > > > - Attaching to process spawning syscalls and dynamically add those PIDs
> > > >   to some bpf map that would be used to filter is cumbersome and
> > > > potentially racy.
> > > > - Unrolling some loop to perform what this helper is doing consumes lots
> > > >   of instructions. That and the impossibility to jump backwards makes it
> > > > really hard to be correct in really large process chains.
> > > >
> > > > Signed-off-by: Javier Honduvilla Coto <javierhonduco@fb.com>
> > > > ---
> > > >  include/linux/bpf.h      |  1 +
> > > >  include/uapi/linux/bpf.h |  3 ++-
> > > >  kernel/bpf/core.c        |  1 +
> > > >  kernel/bpf/helpers.c     | 29 +++++++++++++++++++++++++++++
> > > >  kernel/trace/bpf_trace.c |  2 ++
> > > >  5 files changed, 35 insertions(+), 1 deletion(-)
> > > >
> > > > diff --git a/include/linux/bpf.h b/include/linux/bpf.h
> > > > index de18227b3d95..447395ba202b 100644
> > > > --- a/include/linux/bpf.h
> > > > +++ b/include/linux/bpf.h
> > > > @@ -921,6 +921,7 @@ extern const struct bpf_func_proto bpf_sk_redirect_map_proto;
> > > >  extern const struct bpf_func_proto bpf_spin_lock_proto;
> > > >  extern const struct bpf_func_proto bpf_spin_unlock_proto;
> > > >  extern const struct bpf_func_proto bpf_get_local_storage_proto;
> > > > +extern const struct bpf_func_proto bpf_progenyof_proto;
> > > It seems only used in bpf_trace.c.  Does it have to be here?
> > >
> > > >
> > > >  /* Shared helpers among cBPF and eBPF. */
> > > >  void bpf_user_rnd_init_once(void);
> > > > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > > > index bcdd2474eee7..804e4218eb28 100644
> > > > --- a/include/uapi/linux/bpf.h
> > > > +++ b/include/uapi/linux/bpf.h
> > > > @@ -2457,7 +2457,8 @@ union bpf_attr {
> > > >  	FN(spin_lock),			\
> > > >  	FN(spin_unlock),		\
> > > >  	FN(sk_fullsock),		\
> > > > -	FN(tcp_sock),
> > > > +	FN(tcp_sock),			\
> > > > +	FN(progenyof),
> > > Please add doc like other helpers do.
> >
> > Oops, good catch, thanks! Will send v2 soon!!
> >
> > >
> > > >
> > > >  /* integer value in 'imm' field of BPF_CALL instruction selects which helper
> > > >   * function eBPF program intends to call
> > > > diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
> > > > index ef88b167959d..69e209fbd128 100644
> > > > --- a/kernel/bpf/core.c
> > > > +++ b/kernel/bpf/core.c
> > > > @@ -2015,6 +2015,7 @@ const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
> > > >  const struct bpf_func_proto bpf_get_current_comm_proto __weak;
> > > >  const struct bpf_func_proto bpf_get_current_cgroup_id_proto __weak;
> > > >  const struct bpf_func_proto bpf_get_local_storage_proto __weak;
> > > > +const struct bpf_func_proto bpf_progenyof_proto __weak;
> > > >
> > > >  const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
> > > >  {
> > > > diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> > > > index a411fc17d265..3899787e8dbf 100644
> > > > --- a/kernel/bpf/helpers.c
> > > > +++ b/kernel/bpf/helpers.c
> > > > @@ -18,6 +18,7 @@
> > > >  #include <linux/sched.h>
> > > >  #include <linux/uidgid.h>
> > > >  #include <linux/filter.h>
> > > > +#include <linux/init_task.h>
> > > >
> > > >  /* If kernel subsystem is allowing eBPF programs to call this function,
> > > >   * inside its own verifier_ops->get_func_proto() callback it should return
> > > > @@ -364,3 +365,31 @@ const struct bpf_func_proto bpf_get_local_storage_proto = {
> > > >  };
> > > >  #endif
> > > >  #endif
> > > > +
> > > > +BPF_CALL_1(bpf_progenyof, int, pid)
> > > > +{
> > > > +	int result = 0;
> > > > +	struct task_struct *task = current;
> > > > +
> > > > +	if (unlikely(!task))
> > > hmm.... Could current be NULL?
> >
> > Wasn't sure about this but added as bpf_get_current_pid_tgid,
> > bpf_get_current_uid_gid, and bpf_get_current_comm check for this. Texted Alexei
> > about this and he told me this is probably not necessary anymore, but I
> > guess it doesn't hurt leaving it?
> >
> > >
> > > > +		return -EINVAL;
> > > > +
> > > > +	rcu_read_lock();
> > > > +	while (task != &init_task) {
> > > I don't know the details of init_task, so qq:
> > > Could the passed in "pid" be the init_task->pid?
> > > If possible, what is the expected "result"?
> > >
> >
> > Yep! init_task doesn't set a pid for what I could see, so I guess it
> > will be PID=0. The test in the last patch check bpf_progenyof(0) :)
> >
> > bpf_progenyof with 0 or 1 will always return 1
> the test in patch 3 commit message has this though:
> "- progenyof(0) == 0"
>
> so the intention for progenyof(0) is to always return 0 or 1?
>
> A random ps output from my vm:
> [root@arch-fb-vm1 bpf]# ps -eaf | head -3
> UID        PID  PPID  C STIME TTY          TIME CMD
> root         1     0  0 11:45 ?        00:00:12 /sbin/init
> root         2     0  0 11:45 ?        00:00:00 [kthreadd]
>
> I was asking because,
> after reading the loop, it seems all tasks tracing back to init_task.
> so my intuitive thinking is progenyof(init_task.pid) should always
> return 1.  If it is otherwise, some comments and doc would be useful
> to explain why treating init_task.pid differently.
>

My bad, that was a typo. bpf_progenyof(1) returns 1, and bpf_progenyof(0)
returns 0. Sorry for the confusion.

This is a good point! I chose to return 0 for this case because of
init_task being an implementation detail and PID 0 not having much
meaning for most users, but you are right that I should document it as
an exception if we keep it as is

That being said I think changing the behaviour to make progenyof(0) return 1
makes more sense from a semantics perspective!

> >
> > > > +		if (task->pid == pid) {
> > > > +			result = 1;
> > > > +			break;
> > > > +		}
> > > > +		task = rcu_dereference(task->real_parent);
> > > > +	}
> > > > +	rcu_read_unlock();
> > > > +
> > > > +	return result;
> > > > +}
> > > > +
> > > > +const struct bpf_func_proto bpf_progenyof_proto = {
> > > > +	.func		= bpf_progenyof,
> > > > +	.gpl_only	= false,
> > > > +	.ret_type	= RET_INTEGER,
> > > > +	.arg1_type	= ARG_ANYTHING,
> > > > +};
> > > > diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
> > > > index f1a86a0d881d..8602ae83c799 100644
> > > > --- a/kernel/trace/bpf_trace.c
> > > > +++ b/kernel/trace/bpf_trace.c
> > > > @@ -600,6 +600,8 @@ tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
> > > >  		return &bpf_get_prandom_u32_proto;
> > > >  	case BPF_FUNC_probe_read_str:
> > > >  		return &bpf_probe_read_str_proto;
> > > > +	case BPF_FUNC_progenyof:
> > > > +		return &bpf_progenyof_proto;
> > > >  #ifdef CONFIG_CGROUPS
> > > >  	case BPF_FUNC_get_current_cgroup_id:
> > > >  		return &bpf_get_current_cgroup_id_proto;
> > > > --
> > > > 2.17.1
> > > >