From mboxrd@z Thu Jan 1 00:00:00 1970 From: ebiederm-aS9lmoZGLiVWk0Htik3J/w@public.gmane.org (Eric W. Biederman) Subject: Re: [Devel] [RFC PATCH 0/5] Resend -v2 - Use procfs to change a syscall behavior Date: Wed, 09 Jul 2008 19:14:37 -0700 Message-ID: References: <20080708112422.164370000@bull.net> <20080709221028.GA4926@martell.zuzino.mipt.ru> <20080710013915.GB8327@martell.zuzino.mipt.ru> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20080710013915.GB8327-QDJVlCTZ4KWTKS93B3g+7KFoa47nwP16@public.gmane.org> (Alexey Dobriyan's message of "Thu, 10 Jul 2008 05:39:15 +0400") List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: containers-bounces-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org Errors-To: containers-bounces-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org To: Alexey Dobriyan Cc: containers-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org, Nadia.Derbey-6ktuUTfB/bM@public.gmane.org List-Id: containers.vger.kernel.org Alexey Dobriyan writes: > In case of ->did_exec the only clean interface I see is: > > tsk->did_exec = !!tsk_img->did_exec; > > It would be pretty silly to wrap this one line in a system call (two > actually -- one in, one out), since you're going to restore some more > fields of such variety anyway (like ->pdeath_signal). There I agree the granularity seems small enough to be a major pain for the implementation. > Given the diversity of kernel internal data structures and all sorts of > links between them, the only system call suitable is ioctl(2), not all > this zoo of system calls proposed. They are all extendable and without > rules, but ioctl(2) is also without rules. At least for processes my gut reaction is to look at binary formats and coredumps. Something with at least that large of a granularity seems to make most sense. > This is all said in assumption that serializing kernel-internal data for > checkpoint/restart to userspace is acceptable for mainline. > I don't think it is. I don't believe that serializing kernel-internal data is acceptable for mainline. I believe that serializing user-visible data is acceptable. Note: user-visible data does not mean user-manipulatable data. On a socket you may not save the skbs but you can save the pending packets for example. Assuming the transition cost is not too high. Eric