diff mbox series

[v2,bpf-next,2/4] bpf: allow bpf_d_path in sleepable bpf_iter program

Message ID 20201215233702.3301881-3-songliubraving@fb.com
State Superseded
Headers show
Series introduce bpf_iter for task_vma | expand

Commit Message

Song Liu Dec. 15, 2020, 11:37 p.m. UTC
task_file and task_vma iter programs have access to file->f_path. Enable
bpf_d_path to print paths of these file.

bpf_iter programs are generally called in sleepable context. However, it
is still necessary to diffientiate sleepable and non-sleepable bpf_iter
programs: sleepable programs have access to bpf_d_path; non-sleepable
programs have access to bpf_spin_lock.

Signed-off-by: Song Liu <songliubraving@fb.com>
---
 kernel/trace/bpf_trace.c | 5 +++++
 1 file changed, 5 insertions(+)

Comments

Yonghong Song Dec. 16, 2020, 5:41 p.m. UTC | #1
On 12/15/20 3:37 PM, Song Liu wrote:
> task_file and task_vma iter programs have access to file->f_path. Enable

> bpf_d_path to print paths of these file.

> 

> bpf_iter programs are generally called in sleepable context. However, it

> is still necessary to diffientiate sleepable and non-sleepable bpf_iter

> programs: sleepable programs have access to bpf_d_path; non-sleepable

> programs have access to bpf_spin_lock.

> 

> Signed-off-by: Song Liu <songliubraving@fb.com>


Agreed. So far bpf_iter programs all called from process context.

Acked-by: Yonghong Song <yhs@fb.com>
KP Singh Dec. 16, 2020, 6:15 p.m. UTC | #2
On Wed, Dec 16, 2020 at 1:06 AM Song Liu <songliubraving@fb.com> wrote:
>

> task_file and task_vma iter programs have access to file->f_path. Enable

> bpf_d_path to print paths of these file.

>

> bpf_iter programs are generally called in sleepable context. However, it

> is still necessary to diffientiate sleepable and non-sleepable bpf_iter

> programs: sleepable programs have access to bpf_d_path; non-sleepable

> programs have access to bpf_spin_lock.

>

> Signed-off-by: Song Liu <songliubraving@fb.com>

> ---

>  kernel/trace/bpf_trace.c | 5 +++++

>  1 file changed, 5 insertions(+)

>

> diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c

> index 4be771df5549a..9e5f9b968355f 100644

> --- a/kernel/trace/bpf_trace.c

> +++ b/kernel/trace/bpf_trace.c

> @@ -1191,6 +1191,11 @@ BTF_SET_END(btf_allowlist_d_path)

>

>  static bool bpf_d_path_allowed(const struct bpf_prog *prog)

>  {

> +       if (prog->type == BPF_PROG_TYPE_TRACING &&

> +           prog->expected_attach_type == BPF_TRACE_ITER &&

> +           prog->aux->sleepable)

> +               return true;


For the sleepable/non-sleepable we have been (until now) checking
this in bpf_tracing_func_proto (or bpf_lsm_func_proto)

eg.

case BPF_FUNC_copy_from_user:
return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;

But even beyond that, I don't think this is needed.

We have originally exposed the helper to both sleepable and
non-sleepable LSM and tracing programs with an allow list.

For LSM the allow list is bpf_lsm_is_sleepable_hook) but
that's just an initial allow list and thus causes some confusion
w.r.t to sleep ability (maybe we should add a comment there).

Based on the current logic, my understanding is that
it's okay to use the helper in the allowed hooks in both
"lsm.s/" and "lsm/" (and the same for
BPF_PROG_TYPE_TRACING).

We would have required sleepable only if this helper called "dput"
(which can sleep).

> +

>         if (prog->type == BPF_PROG_TYPE_LSM)

>                 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);

>

> --

> 2.24.1

>
KP Singh Dec. 16, 2020, 6:31 p.m. UTC | #3
On Wed, Dec 16, 2020 at 7:15 PM KP Singh <kpsingh@kernel.org> wrote:
>

> On Wed, Dec 16, 2020 at 1:06 AM Song Liu <songliubraving@fb.com> wrote:

> >

> > task_file and task_vma iter programs have access to file->f_path. Enable

> > bpf_d_path to print paths of these file.

> >

> > bpf_iter programs are generally called in sleepable context. However, it

> > is still necessary to diffientiate sleepable and non-sleepable bpf_iter

> > programs: sleepable programs have access to bpf_d_path; non-sleepable

> > programs have access to bpf_spin_lock.

> >

> > Signed-off-by: Song Liu <songliubraving@fb.com>

> > ---

> >  kernel/trace/bpf_trace.c | 5 +++++

> >  1 file changed, 5 insertions(+)

> >

> > diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c

> > index 4be771df5549a..9e5f9b968355f 100644

> > --- a/kernel/trace/bpf_trace.c

> > +++ b/kernel/trace/bpf_trace.c

> > @@ -1191,6 +1191,11 @@ BTF_SET_END(btf_allowlist_d_path)

> >

> >  static bool bpf_d_path_allowed(const struct bpf_prog *prog)

> >  {

> > +       if (prog->type == BPF_PROG_TYPE_TRACING &&

> > +           prog->expected_attach_type == BPF_TRACE_ITER &&

> > +           prog->aux->sleepable)

> > +               return true;

>


Another try to send it on the list.

> For the sleepable/non-sleepable we have been (until now) checking

> this in bpf_tracing_func_proto (or bpf_lsm_func_proto)

>

> eg.

>

> case BPF_FUNC_copy_from_user:

> return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;

>

> But even beyond that, I don't think this is needed.

>

> We have originally exposed the helper to both sleepable and

> non-sleepable LSM and tracing programs with an allow list.

>

> For LSM the allow list is bpf_lsm_is_sleepable_hook) but

> that's just an initial allow list and thus causes some confusion

> w.r.t to sleep ability (maybe we should add a comment there).

>

> Based on the current logic, my understanding is that

> it's okay to use the helper in the allowed hooks in both

> "lsm.s/" and "lsm/" (and the same for

> BPF_PROG_TYPE_TRACING).

>

> We would have required sleepable only if this helper called "dput"

> (which can sleep).

>

> > +

> >         if (prog->type == BPF_PROG_TYPE_LSM)

> >                 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);

> >

> > --

> > 2.24.1

> >
Jiri Olsa Jan. 25, 2021, 12:52 p.m. UTC | #4
On Tue, Dec 15, 2020 at 03:37:00PM -0800, Song Liu wrote:
> task_file and task_vma iter programs have access to file->f_path. Enable

> bpf_d_path to print paths of these file.

> 

> bpf_iter programs are generally called in sleepable context. However, it

> is still necessary to diffientiate sleepable and non-sleepable bpf_iter

> programs: sleepable programs have access to bpf_d_path; non-sleepable

> programs have access to bpf_spin_lock.

> 

> Signed-off-by: Song Liu <songliubraving@fb.com>

> ---

>  kernel/trace/bpf_trace.c | 5 +++++

>  1 file changed, 5 insertions(+)

> 

> diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c

> index 4be771df5549a..9e5f9b968355f 100644

> --- a/kernel/trace/bpf_trace.c

> +++ b/kernel/trace/bpf_trace.c

> @@ -1191,6 +1191,11 @@ BTF_SET_END(btf_allowlist_d_path)

>  

>  static bool bpf_d_path_allowed(const struct bpf_prog *prog)

>  {

> +	if (prog->type == BPF_PROG_TYPE_TRACING &&

> +	    prog->expected_attach_type == BPF_TRACE_ITER &&

> +	    prog->aux->sleepable)

> +		return true;

> +

>  	if (prog->type == BPF_PROG_TYPE_LSM)

>  		return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);

>  

> -- 

> 2.24.1

> 


would be great to have this merged for bpftrace

Tested-by: Jiri Olsa <jolsa@redhat.com>


thanks,
jirka
diff mbox series

Patch

diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
index 4be771df5549a..9e5f9b968355f 100644
--- a/kernel/trace/bpf_trace.c
+++ b/kernel/trace/bpf_trace.c
@@ -1191,6 +1191,11 @@  BTF_SET_END(btf_allowlist_d_path)
 
 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
 {
+	if (prog->type == BPF_PROG_TYPE_TRACING &&
+	    prog->expected_attach_type == BPF_TRACE_ITER &&
+	    prog->aux->sleepable)
+		return true;
+
 	if (prog->type == BPF_PROG_TYPE_LSM)
 		return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);