Re: [PATCH] ptrace: add ability to retrieve signals without removing them from a queue

[Pavel Emelyanov <xemul@parallels.com>]

On 02/19/2013 11:34 PM, Oleg Nesterov wrote:
> On 02/19, Pavel Emelyanov wrote:
>>
>> On 02/19/2013 04:15 PM, Michael Kerrisk (man-pages) wrote:
>>> On Wed, Feb 13, 2013 at 4:16 PM, Andrey Vagin <avagin@openvz.org> wrote:
>>>> This patch adds a new ptrace request PTRACE_PEEKSIGINFO.
>>>>
>>>> This request is used to retrieve information about a signal with the
>>>> specified sequence number. A siginfo_t structure is copied from the child
>>>> to location data in the parent.
>>>>
>>>> The low 16 bits of addr contains a sequence number of signal in a queue:
>>>
>>> I think 16 bits is probably not enough.... Already, on the "out of the
>>> box" system that I have at hand, one can queue more than 2^16-1
>>> signals:
>>>
>>>     $ cat /proc/$$/status | grep SigQ
>>>     SigQ:	2/126065
>>
>> Yup :( Well, actually it would be enough to have only 1 bit as "flags"
>> and the rest (31 at least) for the number. But taking into account
>> Oleg's wish to have an ability to extend the amount of flags
> 
> I am not sure this is really needed, and we can add more PTRACE_PEEK_
> codes later. I am fine either way, we can even add PEEK_PRIVATE/SHARED
> right now.
> 
> But, given that every PEEK does list_for_each() until it finds the
> necessary sequence number, I am wondering how this O(n**2) will work
> if you want to dump 126065 signals ;)

Isn't it the great reason for making the addr point to a structure, that
would look like

struct siginfo_peek_arg {
	unsigned flags; /* all bits but 0th, that selects between private/shared
			   queues, should be zero */
	unsigned int off; /* from which siginfo to start */
	unsigned int nr; /* how may siginfos to take */
};

? :)

> Oleg.
> 
> .

Thanks,
Pavel