Re: [PATCH v2 bpf-next 2/4] bpf: introduce helper bpf_get_task_stak()

From: Andrii Nakryiko
Date: Fri Jun 26 2020 - 20:06:44 EST


On Fri, Jun 26, 2020 at 4:47 PM Song Liu <songliubraving@xxxxxx> wrote:
>
>
>
> > On Jun 26, 2020, at 3:51 PM, Andrii Nakryiko <andrii.nakryiko@xxxxxxxxx> wrote:
> >
> > On Fri, Jun 26, 2020 at 3:45 PM Song Liu <songliubraving@xxxxxx> wrote:
> >>
> >>
> >>
> >>> On Jun 26, 2020, at 1:17 PM, Andrii Nakryiko <andrii.nakryiko@xxxxxxxxx> wrote:
> >>>
> >>> On Thu, Jun 25, 2020 at 5:14 PM Song Liu <songliubraving@xxxxxx> wrote:
> >>>>
> >>>> Introduce helper bpf_get_task_stack(), which dumps stack trace of given
> >>>> task. This is different to bpf_get_stack(), which gets stack track of
> >>>> current task. One potential use case of bpf_get_task_stack() is to call
> >>>> it from bpf_iter__task and dump all /proc/<pid>/stack to a seq_file.
> >>>>
> >>>> bpf_get_task_stack() uses stack_trace_save_tsk() instead of
> >>>> get_perf_callchain() for kernel stack. The benefit of this choice is that
> >>>> stack_trace_save_tsk() doesn't require changes in arch/. The downside of
> >>>> using stack_trace_save_tsk() is that stack_trace_save_tsk() dumps the
> >>>> stack trace to unsigned long array. For 32-bit systems, we need to
> >>>> translate it to u64 array.
> >>>>
> >>>> Signed-off-by: Song Liu <songliubraving@xxxxxx>
> >>>> ---
> >>>
> >>> Looks great, I just think that there are cases where user doesn't
> >>> necessarily has valid task_struct pointer, just pid, so would be nice
> >>> to not artificially restrict such cases by having extra helper.
> >>>
> >>> Acked-by: Andrii Nakryiko <andriin@xxxxxx>
> >>
> >> Thanks!
> >>
> >>>
> >>>> include/linux/bpf.h | 1 +
> >>>> include/uapi/linux/bpf.h | 35 ++++++++++++++-
> >>>> kernel/bpf/stackmap.c | 79 ++++++++++++++++++++++++++++++++--
> >>>> kernel/trace/bpf_trace.c | 2 +
> >>>> scripts/bpf_helpers_doc.py | 2 +
> >>>> tools/include/uapi/linux/bpf.h | 35 ++++++++++++++-
> >>>> 6 files changed, 149 insertions(+), 5 deletions(-)
> >>>>
> >>>
> >>> [...]
> >>>
> >>>> + /* stack_trace_save_tsk() works on unsigned long array, while
> >>>> + * perf_callchain_entry uses u64 array. For 32-bit systems, it is
> >>>> + * necessary to fix this mismatch.
> >>>> + */
> >>>> + if (__BITS_PER_LONG != 64) {
> >>>> + unsigned long *from = (unsigned long *) entry->ip;
> >>>> + u64 *to = entry->ip;
> >>>> + int i;
> >>>> +
> >>>> + /* copy data from the end to avoid using extra buffer */
> >>>> + for (i = entry->nr - 1; i >= (int)init_nr; i--)
> >>>> + to[i] = (u64)(from[i]);
> >>>
> >>> doing this forward would be just fine as well, no? First iteration
> >>> will cast and overwrite low 32-bits, all the subsequent iterations
> >>> won't even overlap.
> >>
> >> I think first iteration will write zeros to higher 32 bits, no?
> >
> > Oh, wait, I completely misread what this is doing. It up-converts from
> > 32-bit to 64-bit, sorry. Yeah, ignore me on this :)
> >
> > But then I have another question. How do you know that entry->ip has
> > enough space to keep the same number of 2x bigger entries?
>
> The buffer is sized for sysctl_perf_event_max_stack u64 numbers.
> stack_trace_save_tsk() will put at most stack_trace_save_tsk unsigned
> long in it (init_nr == 0). So the buffer is big enough.
>

Awesome, thanks for clarification!

> Thanks,
> Song