linux-kernel - Re: [PATCH 0/1] ns: introduce proc_get_ns_by

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <87si5xxczd.fsf@x220.int.ebiederm.org>
Date:	Tue, 29 Sep 2015 14:05:42 -0500
From:	ebiederm@...ssion.com (Eric W. Biederman)
To:	Oleg Nesterov <oleg@...hat.com>
Cc:	Konstantin Khlebnikov <khlebnikov@...dex-team.ru>,
	linux-api@...r.kernel.org, containers@...ts.linux-foundation.org,
	linux-kernel@...r.kernel.org,
	Roman Gushchin <klamm@...dex-team.ru>,
	Serge Hallyn <serge.hallyn@...ntu.com>,
	Chen Fan <chen.fan.fnst@...fujitsu.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Linus Torvalds <torvalds@...ux-foundation.org>,
	Stéphane Graber <stgraber@...ntu.com>
Subject: Re: [PATCH 0/1] ns: introduce proc_get_ns_by_fd()

Oleg Nesterov <oleg@...hat.com> writes:

> On 09/29, Eric W. Biederman wrote:
>>
>> Oleg Nesterov <oleg@...hat.com> writes:
>>
>> > OK, I won't insist, this too looks better to me than proc_ns_fdget(&fd_ref).
>> >
>> > And in any case fcheck_files() makes more sense than fdget(), somehow I did
>> > not think about this when I sent 1/1.
>> >
>> > Hmm. and after the quick look at cleanup_net() I can't understand whether
>> > get_net_ns_by_fd() can use ns_by_fd_rcu() + maybe_get_net(to_net_ns()) or
>> > not... Can it?
>>
>> Some of those places need a reference that allows them to sleep, and the
>> code is shared with the legacy pid case so with an addition of get_net
>> we can use ns_by_fd_rcu().   There are cases like setns that could
>> use ns_by_fd_rcu() with code reording.
>>
>> We can implement get_net_ns_by_fd as:
>> struct net *get_net_ns_by_fd(int fd)
>> {
>>         struct net *net;
>>
>> 	rcu_read_lock();
>> 	net = net_ns_by_fd_rcu(fd);
>>         if (!IS_ERR(net))
>>         	get_net(net);
>>         rcu_read_unlock();
>>
>> 	return net;
>> }
>>
>> Which means we can achieve code sharing with the pure rcu version
>> as a base.
>
> Yes, this is what I meant... but don't we need maybe_get_net() ?

Let me see.

The call chain is a little interesting so let's trace it.

thread1                               thread2
  net = net_ns_by_fd_rcu(fd);           fput(file)
  get_net(net);                           ____fput
                                            __fput
                                              dput
                                                dentry_kill
                                                  __dentry_kill
                                                    dentry_iput
                                                      iput
                                                        iput_final
                                                          evict
                                                            ->evict_inode (nsfs_evict)
                                                              netns_put
                                                                 put_net
                                                                   __put_net
                                                                      queue_work(net_cleanup_work)
                                                                      ...
                                                                      cleanup_net
                                                                        synchronize_rcu.


Given that someone in another thread can put the file descriptor and I
don't see anything explicitly waiting until after the rcu critical
section ends to put the network namespace I guess we do need
maybe_get_net.

In practice I think the delay logic in fput will act as an rcu barrier,
but that is not guaranteed so we had better not count on it.  Sigh.

Anyway this is a long ways from a syscall to translate pids.

Eric

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/