Message-ID: <Zt_yk0LZ8r8N2MZu@krava>
Date: Tue, 10 Sep 2024 09:17:39 +0200
From: Jiri Olsa <olsajiri@...il.com>
To: Andrii Nakryiko <andrii.nakryiko@...il.com>
Cc: Oleg Nesterov <oleg@...hat.com>, Peter Zijlstra <peterz@...radead.org>,
	Alexei Starovoitov <ast@...nel.org>,
	Daniel Borkmann <daniel@...earbox.net>,
	Andrii Nakryiko <andrii@...nel.org>, bpf@...r.kernel.org,
	Martin KaFai Lau <kafai@...com>, Song Liu <songliubraving@...com>,
	Yonghong Song <yhs@...com>,
	John Fastabend <john.fastabend@...il.com>,
	KP Singh <kpsingh@...omium.org>,
	Stanislav Fomichev <sdf@...gle.com>, Hao Luo <haoluo@...gle.com>,
	Steven Rostedt <rostedt@...dmis.org>,
	Masami Hiramatsu <mhiramat@...nel.org>,
	linux-kernel@...r.kernel.org, linux-trace-kernel@...r.kernel.org
Subject: Re: [PATCHv3 1/7] uprobe: Add support for session consumer

On Mon, Sep 09, 2024 at 04:44:09PM -0700, Andrii Nakryiko wrote:
> On Mon, Sep 9, 2024 at 12:46 AM Jiri Olsa <jolsa@...nel.org> wrote:
> >
> > Adding support for uprobe consumer to be defined as session and have
> > new behaviour for consumer's 'handler' and 'ret_handler' callbacks.
> >
> > The session means that 'handler' and 'ret_handler' callbacks are
> > connected in a way that allows to:
> >
> >   - control execution of 'ret_handler' from 'handler' callback
> >   - share data between 'handler' and 'ret_handler' callbacks
> >
> > The session is enabled by setting new 'session' bool field to true
> > in uprobe_consumer object.
> >
> > We use return_consumer object to keep track of consumers with
> > 'ret_handler'. This object also carries the shared data between
> > 'handler' and and 'ret_handler' callbacks.
> 
> and and

ok

> 
> >
> > The control of 'ret_handler' callback execution is done via return
> > value of the 'handler' callback. This patch adds new 'ret_handler'
> > return value (2) which means to ignore ret_handler callback.
> >
> > Actions on 'handler' callback return values are now:
> >
> >   0 - execute ret_handler (if it's defined)
> >   1 - remove uprobe
> >   2 - do nothing (ignore ret_handler)
> >
> > The session concept fits to our common use case where we do filtering
> > on entry uprobe and based on the result we decide to run the return
> > uprobe (or not).
> >
> > It's also convenient to share the data between session callbacks.
> >
> > Signed-off-by: Jiri Olsa <jolsa@...nel.org>
> > ---
> 
> Just minor things:
> 
> Acked-by: Andrii Nakryiko <andrii@...nel.org>
> 
> >  include/linux/uprobes.h                       |  17 ++-
> >  kernel/events/uprobes.c                       | 132 ++++++++++++++----
> >  kernel/trace/bpf_trace.c                      |   6 +-
> >  kernel/trace/trace_uprobe.c                   |  12 +-
> >  .../selftests/bpf/bpf_testmod/bpf_testmod.c   |   2 +-
> >  5 files changed, 133 insertions(+), 36 deletions(-)
> >
> 
> [...]
> 
> >  enum rp_check {
> > diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
> > index 4b7e590dc428..9e971f86afdf 100644
> > --- a/kernel/events/uprobes.c
> > +++ b/kernel/events/uprobes.c
> > @@ -67,6 +67,8 @@ struct uprobe {
> >         loff_t                  ref_ctr_offset;
> >         unsigned long           flags;
> 
> we should shorten flags to unsigned int, we use one bit out of it
> 
> >
> > +       unsigned int            consumers_cnt;
> > +
> 
> and then this won't increase the size of the struct unnecessarily

right, makes sense

> 
> >         /*
> >          * The generic code assumes that it has two members of unknown type
> >          * owned by the arch-specific code:
> > @@ -826,8 +828,12 @@ static struct uprobe *alloc_uprobe(struct inode *inode, loff_t offset,
> >
> 
> [...]
> 
> >         current->utask->auprobe = NULL;
> >
> > -       if (need_prep && !remove)
> > -               prepare_uretprobe(uprobe, regs); /* put bp at return */
> > +       if (ri && !remove)
> > +               prepare_uretprobe(uprobe, regs, ri); /* put bp at return */
> > +       else
> > +               kfree(ri);
> 
> maybe `else if (ri) kfree(ri)` to avoid unnecessary calls to kfree
> when we only have uprobes?

there's null check in kfree, but it's true that we can skip the
whole call and there's the else condition line already, ok

> 
> >
> >         if (remove && has_consumers) {
> >                 down_read(&uprobe->register_rwsem);
> > @@ -2160,15 +2230,25 @@ static void handler_chain(struct uprobe *uprobe, struct pt_regs *regs)
> >  static void
> >  handle_uretprobe_chain(struct return_instance *ri, struct pt_regs *regs)
> >  {
> > +       struct return_consumer *ric = NULL;
> >         struct uprobe *uprobe = ri->uprobe;
> >         struct uprobe_consumer *uc;
> > -       int srcu_idx;
> > +       int srcu_idx, iter = 0;
> 
> iter -> next_ric_idx  or just ric_idx?

sure, ric_idx seems ok to me

thanks,
jirka

> 
> >
> >         srcu_idx = srcu_read_lock(&uprobes_srcu);
> >         list_for_each_entry_srcu(uc, &uprobe->consumers, cons_node,
> >                                  srcu_read_lock_held(&uprobes_srcu)) {
> > +               /*
> > +                * If we don't find return consumer, it means uprobe consumer
> > +                * was added after we hit uprobe and return consumer did not
> > +                * get registered in which case we call the ret_handler only
> > +                * if it's not session consumer.
> > +                */
> > +               ric = return_consumer_find(ri, &iter, uc->id);
> > +               if (!ric && uc->session)
> > +                       continue;
> >                 if (uc->ret_handler)
> > -                       uc->ret_handler(uc, ri->func, regs);
> > +                       uc->ret_handler(uc, ri->func, regs, ric ? &ric->cookie : NULL);
> >         }
> >         srcu_read_unlock(&uprobes_srcu, srcu_idx);
> >  }
> 
> [...]

Powered by blists - more mailing lists