Re: [PATCH 0/7] Uprobes: Support SDT markers having reference count (semaphore)

[Ravi Bangoria]

Hi Masami,

On 06/16/2018 07:20 PM, Masami Hiramatsu wrote:
> Hi Ravi,
> 
> Sorry for replying later.

No issues :)

> 
> On Mon, 11 Jun 2018 10:01:58 +0530
> Ravi Bangoria <ravi.bangoria@xxxxxxxxxxxxx> wrote:
> 
>> Hi Masami,
>>
>>>>> Hmm, it sounds simple... maybe we can increment refctr in install_breakpoint/
>>>>> remove_breakpoint?
>>>>
>>>> Not really, it would be simpler if I can put it inside install_breakpoint().
>>>> Consider an mmap() case. Probed instruction resides in the text section whereas
>>>> reference counter resides in the data section. These sections gets mapped using
>>>> separate mmap() calls. So, when process mmaps the text section we will change the
>>>> instruction, but section holding the reference counter may not have been mapped
>>>> yet in the virtual memory. If so, we will fail to update the reference counter.
>>>
>>> Got it. 
>>> In such case, maybe we can hook the target page mmapped and do install_breakpoint()
>>> at that point. Since the instruction is protected by a refctr, unless mmap the
>>> page on where the refctr is, the program doesn't reach the tracepoint. Is that right?
>>>
>>
>> You mean, when mmap(text) happens, save the target page somewhere and when
>> mmap(data) happens, update both instruction and ref_ctr?
> 
> Yes. I think you can just clone the target(text) page but not install
> breakpoint, and if mmap(data) happens, update both.

I'm preparing a prototype according to this. The only difference in my approach is,
I'm not delaying instruction update. I.e. let instruction update happen as it is,
just mark that uprobe as delayed. Whenever vma holding reference counter gets mapped, 
update the reference counter.

Will post the series soon.

Thanks for the reply,
Ravi