From: Oleg Nesterov <oleg@redhat.com>
To: "Eric W. Biederman" <ebiederm@xmission.com>,
Pavel Emelyanov <xemul@openvz.org>,
Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>,
Andrew Morton <akpm@linux-foundation.org>,
Linux Containers <containers@lists.osdl.org>,
linux-kernel@vger.kernel.org
Subject: Re: [RESEND PATCH] pid_ns: Fix proc_flush_task() accessing freed proc_mnt
Date: Fri, 25 Jun 2010 14:21:18 +0200 [thread overview]
Message-ID: <20100625122118.GA5341@redhat.com> (raw)
In-Reply-To: <20100625102303.GG3773@hawkmoon.kerlabs.com>
On 06/25, Louis Rilling wrote:
>
> On 24/06/10 21:18 +0200, Oleg Nesterov wrote:
> >
> > and this adds the extra code to alloc/free pidmap.
>
> Hopefully this is not much in alloc_pidmap() since we can expect that nr_pids
> and last_pid are in the same cache line.
This also adds atomic op. But I mostly dislike the pid-ns-specific
complications itself (including the mount-after-the-first-alloc_pid
dependancy), not the minor perfomance penalty.
But see below...
> > And, this subjective, yes, but it looks a bit strange that upid->nr
> > has a reference to proc_mnt.
>
> I presume that you wanted to say upid->ns.
I meant ns->nr_pids ;)
> This last point is what made me worry about your approach so far, although I did
> not take time to spot the precise issues. Unfortunately I don't see what the
> checks you added in proc_self_readlink(), proc_self_follow_link() and
> proc_pid_lookup() buy. What does prevent destroy_pid_namespace() from running
> concurrently? Maybe RCU could help in those cases?
Very good point. And the strong argument against this approach.
> Moreover, I think that proc_pid_readdir() should get some check too.
Well, it checks ->reaper != NULL, that is why I don't verify ns.
But yes, we have the same race (races) you pointed out here.
> void pid_ns_release_proc(struct pid_namespace *ns)
> {
> struct inode *root_inode;
>
> if (ns->proc_mnt) {
> root_inode = ns->proc_mnt->mnt_sb->s_root->d_inode;
>
> mutex_lock(&root_inode->i_mutex);
> ns->proc_mnt->mnt_sb->s_fs_info = NULL;
> PROC_I(root_inode)->pid = NULL;
> mutex_unlock(&root_inode->i_mutex);
>
> mntput(ns->proc_mnt);
> }
> }
>
> This would also solve the issue for proc_pid_lookup() btw.
Looks like you are right, but this doesn't help proc_self_readlink().
I think we can fix all these problems, but I no longer think this
approach can pretend to simplify the code. No, it will make the
code more complex/ugly and potentially more buggy.
Louis, thank you very much.
Oleg.
next prev parent reply other threads:[~2010-06-25 12:23 UTC|newest]
Thread overview: 80+ messages / expand[flat|nested] mbox.gz Atom feed top
2010-06-16 16:34 [PATCH] procfs: Do not release pid_ns->proc_mnt too early Louis Rilling
2010-06-17 9:53 ` Pavel Emelyanov
2010-06-17 13:41 ` Eric W. Biederman
2010-06-17 14:20 ` Louis Rilling
2010-06-17 21:36 ` Oleg Nesterov
2010-06-18 8:27 ` Louis Rilling
2010-06-18 16:27 ` Oleg Nesterov
2010-06-21 11:11 ` Louis Rilling
2010-06-21 12:58 ` Eric W. Biederman
2010-06-21 14:15 ` Louis Rilling
2010-06-21 14:26 ` Eric W. Biederman
2010-06-17 21:20 ` Oleg Nesterov
2010-06-18 8:20 ` Louis Rilling
2010-06-18 11:15 ` Oleg Nesterov
2010-06-18 16:08 ` Oleg Nesterov
2010-06-18 17:33 ` Louis Rilling
2010-06-18 17:55 ` Oleg Nesterov
2010-06-18 21:23 ` Oleg Nesterov
2010-06-19 19:08 ` [PATCH 0/4] pid_ns_prepare_proc/unshare cleanups Oleg Nesterov
2010-06-19 19:09 ` [PATCH 1/4] procfs: proc_get_sb: consolidate/cleanup root_inode->pid logic Oleg Nesterov
2010-06-19 19:10 ` [PATCH 2/4] procfs: kill the global proc_mnt variable Oleg Nesterov
2010-06-19 19:10 ` [PATCH 3/4] procfs: move pid_ns_prepare_proc() from copy_process() to create_pid_namespace() Oleg Nesterov
2010-06-19 19:11 ` [PATCH RESEND 4/4] sys_unshare: simplify the not-really-implemented CLONE_THREAD/SIGHAND/VM code Oleg Nesterov
2010-06-20 8:42 ` [PATCH 0/6] Unshare support for the pid namespace Eric W. Biederman
2010-06-20 8:44 ` [PATCH 1/6] pid: Remove the child_reaper special case in init/main.c Eric W. Biederman
2010-06-20 18:29 ` Oleg Nesterov
2010-06-20 20:27 ` Oleg Nesterov
2010-06-20 8:45 ` [PATCH 2/6] pidns: Call pid_ns_prepare_proc from create_pid_namespace Eric W. Biederman
2010-06-20 18:19 ` Oleg Nesterov
2010-06-20 8:45 ` [PATCH 3/6] procfs: kill the global proc_mnt variable Eric W. Biederman
2010-06-20 8:47 ` [PATCH 4/6] pidns: Don't allow new pids after the namespace is dead Eric W. Biederman
2010-06-20 18:44 ` Oleg Nesterov
2010-06-20 8:48 ` [PATCH 5/6] pidns: Use task_active_pid_ns where appropriate Eric W. Biederman
2010-06-20 8:49 ` [PATCH 6/6] pidns: Support unsharing the pid namespace Eric W. Biederman
2010-06-20 20:14 ` Oleg Nesterov
2010-06-20 20:42 ` Oleg Nesterov
2010-06-21 1:53 ` Eric W. Biederman
2010-06-20 18:03 ` [PATCH 0/6] Unshare support for " Oleg Nesterov
2010-06-20 18:05 ` [PATCH 0/2] pid_ns_release_proc() fixes Oleg Nesterov
2010-06-20 18:06 ` [PATCH 1/2] pid_ns: move destroy_pid_namespace() into workqueue context Oleg Nesterov
2010-06-20 18:06 ` [PATCH 2/2] pid_ns: refactor the buggy pid_ns_release_proc() logic Oleg Nesterov
2010-06-20 21:00 ` [PATCH 0/6] Unshare support for the pid namespace Eric W. Biederman
2010-06-20 21:48 ` Oleg Nesterov
2010-06-20 21:56 ` Oleg Nesterov
2011-01-26 15:57 ` Daniel Lezcano
2010-06-23 20:36 ` [PATCH 0/1] pid_ns: move pid_ns_release_proc() from proc_flush_task() to zap_pid_ns_processes() Oleg Nesterov
2010-06-23 20:37 ` [PATCH 1/1] " Oleg Nesterov
2010-06-24 6:36 ` Sukadev Bhattiprolu
2010-06-24 12:59 ` Oleg Nesterov
2010-06-24 7:06 ` Eric W. Biederman
2010-06-24 13:01 ` Oleg Nesterov
2010-06-24 8:37 ` [PATCH] pid_ns: Fix proc_flush_task() accessing freed proc_mnt Louis Rilling
2010-06-24 17:08 ` [RESEND PATCH] " Louis Rilling
2010-06-24 19:18 ` Oleg Nesterov
2010-06-25 10:23 ` Louis Rilling
2010-06-25 12:21 ` Oleg Nesterov [this message]
2010-06-25 18:37 ` Sukadev Bhattiprolu
2010-06-25 19:29 ` Oleg Nesterov
2010-06-25 21:26 ` Sukadev Bhattiprolu
2010-06-25 21:27 ` Oleg Nesterov
2010-06-25 22:07 ` Sukadev Bhattiprolu
2010-07-09 4:36 ` [RFC][PATCH 1/2] pidns: Add a flag to indicate a pid namespace is dead Eric W. Biederman
2010-07-09 4:39 ` [RFC][PATCH 2/2] pidns: Remove proc flush races when a pid namespaces are exiting Eric W. Biederman
2010-07-09 12:14 ` Louis Rilling
2010-07-09 13:05 ` Eric W. Biederman
2010-07-09 14:13 ` Louis Rilling
2010-07-09 15:58 ` [PATCH 01/24] pidns: Remove races by stopping the caching of proc_mnt Eric W. Biederman
2010-07-09 22:13 ` Serge E. Hallyn
2010-07-11 14:14 ` Louis Rilling
2010-07-11 14:25 ` Eric W. Biederman
2010-07-12 18:09 ` [PATCH] pidns: Fix wait for zombies to be reaped in zap_pid_ns_processes Eric W. Biederman
2010-07-13 21:42 ` Louis Rilling
2010-07-13 22:34 ` Serge E. Hallyn
2010-07-14 1:47 ` Eric W. Biederman
2010-10-30 7:07 ` Sukadev Bhattiprolu
2010-07-14 20:53 ` Sukadev Bhattiprolu
2010-07-14 21:35 ` Eric W. Biederman
2010-06-21 11:09 ` [PATCH] procfs: Do not release pid_ns->proc_mnt too early Louis Rilling
2010-06-21 11:15 ` Louis Rilling
2010-06-21 14:38 ` Oleg Nesterov
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20100625122118.GA5341@redhat.com \
--to=oleg@redhat.com \
--cc=akpm@linux-foundation.org \
--cc=containers@lists.osdl.org \
--cc=ebiederm@xmission.com \
--cc=linux-kernel@vger.kernel.org \
--cc=sukadev@linux.vnet.ibm.com \
--cc=xemul@openvz.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).