linux-kernel - Re: [PATCH linux-next] mqueue: fix IPC namespace use-after-free

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: <874lomhcwb.fsf@redhat.com>
Date:   Tue, 19 Dec 2017 17:59:32 +0100
From:   Giuseppe Scrivano <gscrivan@...hat.com>
To:     Al Viro <viro@...IV.linux.org.uk>
Cc:     Andrew Morton <akpm@...ux-foundation.org>,
        LKML <linux-kernel@...r.kernel.org>, alexander.deucher@....com,
        broonie@...nel.org, chris@...is-wilson.co.uk,
        David Miller <davem@...emloft.net>, deepa.kernel@...il.com,
        Greg KH <gregkh@...uxfoundation.org>,
        luc.vanoostenryck@...il.com, lucien xin <lucien.xin@...il.com>,
        Ingo Molnar <mingo@...nel.org>,
        Neil Horman <nhorman@...driver.com>,
        syzkaller-bugs@...glegroups.com,
        Vladislav Yasevich <vyasevich@...il.com>
Subject: Re: [PATCH linux-next] mqueue: fix IPC namespace use-after-free

Al Viro <viro@...IV.linux.org.uk> writes:

> On Tue, Dec 19, 2017 at 11:48:19AM +0000, Al Viro wrote:
>> On Tue, Dec 19, 2017 at 11:14:40AM +0100, Giuseppe Scrivano wrote:
>> > mqueue_evict_inode() doesn't access the ipc namespace if it was
>> > already freed.  It can happen if in a new IPC namespace the inode was
>> > created without a prior mq_open() which creates the vfsmount used to
>> > access the superblock from mq_clear_sbinfo().
>> > 
>> > Keep a direct pointer to the superblock used by the inodes so we can
>> > correctly reset the reference to the IPC namespace being destroyed.
>> > 
>> > Bug introduced with 9c583773d03633 ("ipc, mqueue: lazy call
>> > kern_mount_data in new namespaces")
>> 
>> And just what will happen in the same scenario if you mount the damn
>> thing in userland without ever calling mq_open(), touch a file there,
>> then unmount and then leave the ipc namespace?
>
> FWIW, the real solution would be to have userland mounts trigger the creation
> of internal one, same as mq_open() would.  Something along these lines
> (completely untested, on top of vfs.git#for-next).  Care to give it some
> beating?

thanks for the patch.  It seems to work after this minor fixup on top of
it:

diff --git a/ipc/mqueue.c b/ipc/mqueue.c
index 30327e201571..636989a44fae 100644
--- a/ipc/mqueue.c
+++ b/ipc/mqueue.c
@@ -360,7 +360,7 @@ static struct vfsmount *mq_internal_mount(void)
 		spin_unlock(&mq_lock);
 		if (!IS_ERR(m))
 			kern_unmount(m);
-			return ns->mq_mnt;
+		return ns->mq_mnt;
 	}
 	if (!IS_ERR(m))
 		ns->mq_mnt = m;
@@ -1560,6 +1560,7 @@ static struct file_system_type mqueue_fs_type = {
 
 int mq_init_ns(struct ipc_namespace *ns)
 {
+	ns->mq_mnt           = NULL;
 	ns->mq_queues_count  = 0;
 	ns->mq_queues_max    = DFLT_QUEUESMAX;
 	ns->mq_msg_max       = DFLT_MSGMAX;


The only issue I've seen with my version is that if I do:

# unshare -im /bin/sh
# mount -t mqueue mqueue /dev/mqueue
# touch /dev/mqueue/foo
# umount /dev/mqueue
# mount -t mqueue mqueue /dev/mqueue

then /dev/mqueue/foo doesn't exist at this point.  Your patch does not
have this problem and /dev/mqueue/foo is again accessible after the
second mount.

Regards,
Giuseppe