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

From: Pavel Emelyanov <xemul@parallels.com>
To: Oleg Nesterov <oleg@redhat.com>
Cc: mtk.manpages@gmail.com, Andrey Vagin <avagin@openvz.org>,
	linux-kernel@vger.kernel.org, criu@openvz.org,
	linux-api@vger.kernel.org, Roland McGrath <roland@redhat.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	"Paul E. McKenney" <paulmck@linux.vnet.ibm.com>,
	David Howells <dhowells@redhat.com>,
	Dave Jones <davej@redhat.com>,
	Linus Torvalds <torvalds@linux-foundation.org>
Subject: Re: [PATCH] ptrace: add ability to retrieve signals without removing them from a queue
Date: Thu, 21 Feb 2013 14:30:26 +0400	[thread overview]
Message-ID: <5125F742.2050707@parallels.com> (raw)
In-Reply-To: <20130220223703.GB16194@redhat.com>

On 02/21/2013 02:37 AM, Oleg Nesterov wrote:
> On 02/19, Pavel Emelyanov wrote:
>>
>> On 02/19/2013 11:34 PM, Oleg Nesterov wrote:
>>> 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 */
>> };
> 
> I am fine either way, to me everything looks better than signalfd
> hacks.
> 
> But if you meant "avoid n^2", this won't help? You can't do
> copy_siginfo_to_user() under ->siglock, so you need to restart
> list_for_each() anyway.

Or allocate intermediate buffer putting the siginfo's there.

> Oleg.

Thanks,
Pavel