From mboxrd@z Thu Jan 1 00:00:00 1970 From: ebiederm@xmission.com (Eric W. Biederman) Subject: Re: [PATCH 00/23] Removal of binary sysctl support Date: Sun, 08 Nov 2009 19:44:02 -0800 Message-ID: References: <200911081306.20201.arnd@arndb.de> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: linux-kernel@vger.kernel.org, netdev@vger.kernel.org, David Miller , Stephen Rothwell To: Arnd Bergmann Return-path: In-Reply-To: <200911081306.20201.arnd@arndb.de> (Arnd Bergmann's message of "Sun\, 8 Nov 2009 13\:06\:20 +0000") Sender: linux-kernel-owner@vger.kernel.org List-Id: netdev.vger.kernel.org Arnd Bergmann writes: > On Sunday 08 November 2009 12:16:43 Eric W. Biederman wrote: >> This patchset reimplements sys_sysctl as a compatibility wrapper >> around /proc/sys. After which it removes all of the code to all over >> the kernel that is used today to implement the binary sysctls. >> >> I am posting this patchset to give everyone a heads up what is in >> flight. >> >> I intend to carry all of these patches in my sysctl tree. > > Very nice patches again! > > Looking at what you did, I had two ideas how to move on from there, > which may be part of your plans already: > > 1. Make it possible to build sysctl_binary.c as a loadable module > so you can get a smaller kernel without losing the option to use > binary sysctl altogether. This of course requires a small portion > to remain in the kernel, to provide the actual syscall entry point > and load the module on demand. I can see how this could make sense from a distribution perspective. > 2. On top of that, put the same code into glibc so that you don't > even have to load the module when you're running a new glibc version. > Since the binary sysctl ABI is stable (as in stiff and dead), there > should be no need to synchronize any extensions to it betwen kernel > and libc. I don't expect we will need to move this to glibc. There are so few users of sys_sysctl now, that I hardly expect it to be worth it to move this code out of the kernel. For me the big problem with sys_sysctl is solved by this patchset. It no longer constitutes a maintenance burden on the rest of the sysctl code, and the other sysctl users. Now the implementation of /proc/sys can be implemented and optimized without dealing with any sys_sysctl baggage. All of that said if someone is interested in tweaking sysctl_binary.c to make it easier to deal with sys_sysctl going away I don't have any problems. Eric