Re: [PATCH bpf-next 1/3] bpf: introduce helper bpf_get_task_stack_trace()

[Song Liu <songliubraving@fb.com>]

> On Jun 23, 2020, at 11:45 AM, Andrii Nakryiko <andrii.nakryiko@gmail.com> wrote:
> 
> On Tue, Jun 23, 2020 at 12:08 AM Song Liu <songliubraving@fb.com> wrote:
>> 
>> This helper can be used with bpf_iter__task to dump all /proc/*/stack to
>> a seq_file.
>> 
>> Signed-off-by: Song Liu <songliubraving@fb.com>
>> ---
>> include/uapi/linux/bpf.h       | 10 +++++++++-
>> kernel/trace/bpf_trace.c       | 21 +++++++++++++++++++++
>> scripts/bpf_helpers_doc.py     |  2 ++
>> tools/include/uapi/linux/bpf.h | 10 +++++++++-
>> 4 files changed, 41 insertions(+), 2 deletions(-)
>> 
>> diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
>> index 19684813faaed..a30416b822fe3 100644
>> --- a/include/uapi/linux/bpf.h
>> +++ b/include/uapi/linux/bpf.h
>> @@ -3252,6 +3252,13 @@ union bpf_attr {
>>  *             case of **BPF_CSUM_LEVEL_QUERY**, the current skb->csum_level
>>  *             is returned or the error code -EACCES in case the skb is not
>>  *             subject to CHECKSUM_UNNECESSARY.
>> + *
>> + * int bpf_get_task_stack_trace(struct task_struct *task, void *entries, u32 size)
>> + *     Description
>> + *             Save a task stack trace into array *entries*. This is a wrapper
>> + *             over stack_trace_save_tsk().
>> + *     Return
>> + *             Number of trace entries stored.
>>  */
>> #define __BPF_FUNC_MAPPER(FN)          \
>>        FN(unspec),                     \
>> @@ -3389,7 +3396,8 @@ union bpf_attr {
>>        FN(ringbuf_submit),             \
>>        FN(ringbuf_discard),            \
>>        FN(ringbuf_query),              \
>> -       FN(csum_level),
>> +       FN(csum_level),                 \
>> +       FN(get_task_stack_trace),
> 
> We have get_stackid and get_stack, I think to stay consistent it
> should be named get_task_stack
> 
>> 
>> /* integer value in 'imm' field of BPF_CALL instruction selects which helper
>>  * function eBPF program intends to call
>> diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
>> index e729c9e587a07..2c13bcb5c2bce 100644
>> --- a/kernel/trace/bpf_trace.c
>> +++ b/kernel/trace/bpf_trace.c
>> @@ -1488,6 +1488,23 @@ static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
>>        .arg4_type      = ARG_ANYTHING,
>> };
>> 
>> +BPF_CALL_3(bpf_get_task_stack_trace, struct task_struct *, task,
>> +          void *, entries, u32, size)
> 
> See get_stack definition, this one needs to support flags as well. And
> we should probably support BPF_F_USER_BUILD_ID as well. And
> BPF_F_USER_STACK is also a good idea, I presume?

This will be a different direction that is similar to stackmap implementation.
Current version follows the implementation behind /proc/<pid>/stack . Let me 
check which of them is better. 

Thanks,
Song