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

[Song Liu <songliubraving@fb.com>]

> On Jun 26, 2020, at 1:17 PM, Andrii Nakryiko <andrii.nakryiko@gmail.com> wrote:
> 
> On Thu, Jun 25, 2020 at 5:14 PM Song Liu <songliubraving@fb.com> 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@fb.com>
>> ---
> 
> 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@fb.com>

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?

> 
>> +       }
>> +
>> +exit_put:
>> +       put_callchain_entry(rctx);
>> +
>> +       return entry;
>> +}
>> +
> 
> [...]
> 
>> +BPF_CALL_4(bpf_get_task_stack, struct task_struct *, task, void *, buf,
>> +          u32, size, u64, flags)
>> +{
>> +       struct pt_regs *regs = task_pt_regs(task);
>> +
>> +       return __bpf_get_stack(regs, task, buf, size, flags);
>> +}
> 
> 
> So this takes advantage of BTF and having a direct task_struct
> pointer. But for kprobes/tracepoint I think it would also be extremely
> helpful to be able to request stack trace by PID. How about one more
> helper which will wrap this one with get/put task by PID, e.g.,
> bpf_get_pid_stack(int pid, void *buf, u32 size, u64 flags)? Would that
> be a problem?

That should work. Let me add that in a follow up patch.