[BUG] cannot mount subvolume with selinux context

[BUG] cannot mount subvolume with selinux context

[Eryu Guan]

Hi,

Description of the problem:

mount btrfs with selinux context, then create a subvolume, the new
subvolume cannot be mounted, even with the same context.

mkfs -t btrfs /dev/sda5
mount -o context=system_u:object_r:nfs_t:s0 /dev/sda5 /mnt/btrfs
btrfs subvolume create /mnt/btrfs/subvol
mount -o subvol=subvol,context=system_u:object_r:nfs_t:s0 /dev/sda5 /mnt/test

The last mount fails, and dmesg shows:
SELinux: mount invalid.  Same superblock, different security settings for (dev sda5, type btrfs)

But in fact, the security settings are the same.

from fs/super.c

struct dentry *
mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
{
        struct dentry *root;
        struct super_block *sb;
        char *secdata = NULL;
        int error = -ENOMEM;

        if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
                secdata = alloc_secdata();
                if (!secdata)
                        goto out;

                error = security_sb_copy_data(data, secdata);
                if (error)
                        goto out_free_secdata;
        }

        root = type->mount(type, flags, name, data);

The security_sb_copy_data() takes out selinux context data to
"secdata", then mount_subvol() calls mount_fs() (via vfs_kern_mount())
again without selinux context, so mount_subvol() fails, which fails
the whole mount.

Not sure what's the proper fix. Zach suggestted that the fix will
probably be to rework the vfs functions a bit as he said in rh
bugzilla[1].

Thanks,
Eryu

[1] https://bugzilla.redhat.com/show_bug.cgi?id=1130860

Re: [BUG] cannot mount subvolume with selinux context

[Zach Brown]

On Tue, Aug 19, 2014 at 11:32:16AM +0800, Eryu Guan wrote:
> Hi,
> 
> Description of the problem:
> 
> mount btrfs with selinux context, then create a subvolume, the new
> subvolume cannot be mounted, even with the same context.
> 
> mkfs -t btrfs /dev/sda5
> mount -o context=system_u:object_r:nfs_t:s0 /dev/sda5 /mnt/btrfs
> btrfs subvolume create /mnt/btrfs/subvol
> mount -o subvol=subvol,context=system_u:object_r:nfs_t:s0 /dev/sda5 /mnt/test

Submit a xfstest?

> The security_sb_copy_data() takes out selinux context data to
> "secdata", then mount_subvol() calls mount_fs() (via vfs_kern_mount())
> again without selinux context, so mount_subvol() fails, which fails
> the whole mount.
> 
> Not sure what's the proper fix. Zach suggestted that the fix will
> probably be to rework the vfs functions a bit as he said in rh
> bugzilla[1].

Yeah, I have no idea what'd be preferred here:

 - rework the vfs _kern_ mount api to offer one that doesn't mess with
   selinux mount options
 - add a flag to have the second _kern_ mount ignore selinux (but not
   MS_KERNMOUNT?)
 - binary data and fs selinux handling?  (like nfs)

- z

Re: [BUG] cannot mount subvolume with selinux context

[Eryu Guan]

On Tue, Aug 19, 2014 at 10:28:54AM -0700, Zach Brown wrote:
> On Tue, Aug 19, 2014 at 11:32:16AM +0800, Eryu Guan wrote:
> > Hi,
> > 
> > Description of the problem:
> > 
> > mount btrfs with selinux context, then create a subvolume, the new
> > subvolume cannot be mounted, even with the same context.
> > 
> > mkfs -t btrfs /dev/sda5
> > mount -o context=system_u:object_r:nfs_t:s0 /dev/sda5 /mnt/btrfs
> > btrfs subvolume create /mnt/btrfs/subvol
> > mount -o subvol=subvol,context=system_u:object_r:nfs_t:s0 /dev/sda5 /mnt/test
> 
> Submit a xfstest?

Sure, will do.

Thanks,
Eryu
> 
> > The security_sb_copy_data() takes out selinux context data to
> > "secdata", then mount_subvol() calls mount_fs() (via vfs_kern_mount())
> > again without selinux context, so mount_subvol() fails, which fails
> > the whole mount.
> > 
> > Not sure what's the proper fix. Zach suggestted that the fix will
> > probably be to rework the vfs functions a bit as he said in rh
> > bugzilla[1].
> 
> Yeah, I have no idea what'd be preferred here:
> 
>  - rework the vfs _kern_ mount api to offer one that doesn't mess with
>    selinux mount options
>  - add a flag to have the second _kern_ mount ignore selinux (but not
>    MS_KERNMOUNT?)
>  - binary data and fs selinux handling?  (like nfs)
> 
> - z

Re: [BUG] cannot mount subvolume with selinux context

[Qu Wenruo]

-------- Original Message --------
Subject: Re: [BUG] cannot mount subvolume with selinux context
From: Eryu Guan <guaneryu@gmail.com>
To: Zach Brown <zab@zabbo.net>
Date: 2014年08月20日 11:57
> On Tue, Aug 19, 2014 at 10:28:54AM -0700, Zach Brown wrote:
>> On Tue, Aug 19, 2014 at 11:32:16AM +0800, Eryu Guan wrote:
>>> Hi,
>>>
>>> Description of the problem:
>>>
>>> mount btrfs with selinux context, then create a subvolume, the new
>>> subvolume cannot be mounted, even with the same context.
>>>
>>> mkfs -t btrfs /dev/sda5
>>> mount -o context=system_u:object_r:nfs_t:s0 /dev/sda5 /mnt/btrfs
>>> btrfs subvolume create /mnt/btrfs/subvol
>>> mount -o subvol=subvol,context=system_u:object_r:nfs_t:s0 /dev/sda5 /mnt/test
>> Submit a xfstest?
> Sure, will do.
>
> Thanks,
> Eryu
>>> The security_sb_copy_data() takes out selinux context data to
>>> "secdata", then mount_subvol() calls mount_fs() (via vfs_kern_mount())
>>> again without selinux context, so mount_subvol() fails, which fails
>>> the whole mount.
>>>
>>> Not sure what's the proper fix. Zach suggestted that the fix will
>>> probably be to rework the vfs functions a bit as he said in rh
>>> bugzilla[1].
>> Yeah, I have no idea what'd be preferred here:
>>
>>   - rework the vfs _kern_ mount api to offer one that doesn't mess with
>>     selinux mount options
>>   - add a flag to have the second _kern_ mount ignore selinux (but not
>>     MS_KERNMOUNT?)
>>   - binary data and fs selinux handling?  (like nfs)
In fact, we can just make btrfs deal with "subvol=" mount option in a 
new method.
Current, btrfs handle "subvol=" by call vfs_kern_mount again and use vfs 
level mount_subtree() to do the path
search thing.

But on the other hand, btrfs does not call vfs_kern_mount() when 
handling default subvolume or "subvolid=" mount,
so, I think we can do all the path search inside btrfs instead of reuse 
vfs level functions, and convert "subvol="
mount option to "subvolid=", which should be selinux friendly now.
(And in this method mount_subvol() should be called just before 
get_default_root()).

If I am wrong, please tell me.

BTW, it seems that if mainline kernel accept the patchset which convert 
"subvolid=" to "subvol=", it will make the
bug more seriously. :-(
Thank goddness, the successor patch uses get_path()....

Thanks,
Qu
>>
>> - z
> --
> To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html