From mboxrd@z Thu Jan 1 00:00:00 1970 From: ebiederm-aS9lmoZGLiVWk0Htik3J/w@public.gmane.org (Eric W. Biederman) Subject: Re: [PATCH] proc: Do not invalidate dentries with submounts Date: Thu, 06 Dec 2007 09:05:02 -0700 Message-ID: References: <20071205143518.GA24886@iris.sw.ru> <4757CF70.3020508@sw.ru> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: In-Reply-To: <4757CF70.3020508-3ImXcnM4P+0@public.gmane.org> (Denis V. Lunev's message of "Thu, 06 Dec 2007 13:31:12 +0300") 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: "Denis V. Lunev" Cc: linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, marcus-sJr3legBufCzQB+pC5nmwQ@public.gmane.org, containers-qjLDD68F18O7TbgM5vRIOg@public.gmane.org, "Denis V. Lunev" , Andrew Morton List-Id: containers.vger.kernel.org "Denis V. Lunev" writes: > you have changed the behavior of revalidation by shadows. I think it > will be better to restore it and keep new one for shadows (and below) > only, which has been done by my yesterday patch. - I think it is better to move forward rather then back. - The old proc dentry caching behavior is actually too aggressive, and has problem corner cases. Keeping the dentries when we have something mounted on top is a trade off that is the least of two evils. - My change fixes the mount leak on all of /proc not just on /proc/generic. What you did is a hack that restored the old slightly buggy behavior. Which is fine if we can't find anything better. It is not code that is on the path towards a /proc that properly caches it's dentries. With the old behavior a random user space application can open a file or a directory in /proc pinning it's dcache entry. Then the module supplying that open file can be removed and reinserted. Until the user space application removes reference to that /proc file all you will be able to find is the version of the file from before /proc was removed. That sounds like a way to trigger nasty behavior to me. I would like to remove that possibility from the kernel if I can. Eric