[RFC 20/20] ima: Setup securityfs_ns for IMA namespace

[Stefan Berger]

On 12/1/21 12:56, James Bottomley wrote:
> On Tue, 2021-11-30 at 11:06 -0500, Stefan Berger wrote:
> [...]
>> +
>> +/*
>> + * Fix the ownership (uid/gid) of the dentry's that couldn't be set
>> at the
>> + * time of their creation because the user namespace wasn't
>> configured, yet.
>> + */
>> +static void ima_fs_ns_fixup_uid_gid(struct ima_namespace *ns)
>> +{
>> +	struct inode *inode;
>> +	size_t i;
>> +
>> +	if (ns->file_ownership_fixes_done ||
>> +	    ns->user_ns->uid_map.nr_extents == 0)
>> +		return;
>> +
>> +	ns->file_ownership_fixes_done = true;
>> +	for (i = 0; i < IMAFS_DENTRY_LAST; i++) {
>> +		if (!ns->dentry[i])
>> +			continue;
>> +		inode = ns->dentry[i]->d_inode;
>> +		inode->i_uid = make_kuid(ns->user_ns, 0);
>> +		inode->i_gid = make_kgid(ns->user_ns, 0);
>> +	}
>> +}
>> +
>> +/* Fix the permissions when a file is opened */
>> +int ima_fs_ns_permission(struct user_namespace *mnt_userns, struct
>> inode *inode,
>> +			 int mask)
>> +{
>> +	ima_fs_ns_fixup_uid_gid(get_current_ns());
>> +	return generic_permission(mnt_userns, inode, mask);
>> +}
>> +
>> +const struct inode_operations ima_fs_ns_inode_operations = {
>> +	.lookup		= simple_lookup,
>> +	.permission	= ima_fs_ns_permission,
>> +};
>> +
> In theory this uid/gid shifting should have already been done for you
> and all of the above code should be unnecessary.  What is supposed to
> happen is that the mount of securityfs_ns in the new user namespace
> should pick up a superblock s_user_ns for that new user namespace.  Now
> inode_alloc() uses i_uid_write(inode, 0) which maps back through the
> s_user_ns to obtain the owner of the user namespace.
>
> What can happen is that if you do the inode allocation before (or even
> without) writing to the uid_map file, it maps back through an empty map
> and ends up with -1 for i_uid ... is this what you're seeing?

I tried this with runc and a user namespace active mapping uid 1000 on 
the host to uid 0 in the container. There I run into the problem that 
all of the files and directories without the above work-around are 
mapped to 'nobody', just like all the files in sysfs in this case are 
also mapped to nobody. This code resolved the issue.


sh-5.1# ls -l /sys/
total 0
drwxr-xr-x.   2 nobody nobody  0 Dec  1 18:06 block
drwxr-xr-x.  28 nobody nobody  0 Dec  1 18:06 bus
drwxr-xr-x.  54 nobody nobody  0 Dec  1 18:06 class
drwxr-xr-x.   4 nobody nobody  0 Dec  1 18:06 dev
drwxr-xr-x.  15 nobody nobody  0 Dec  1 18:06 devices
drwxrwxrwt.   2 root   root   40 Dec  1 18:06 firmware
drwxr-xr-x.   9 nobody nobody  0 Dec  1 18:06 fs
drwxr-xr-x.  16 nobody nobody  0 Dec  1 18:06 kernel
drwxr-xr-x. 161 nobody nobody  0 Dec  1 18:06 module
drwxr-xr-x.   3 nobody nobody  0 Dec  1 18:06 power

sh-5.1# ls -l /sys/kernel/security/
total 0
lr--r--r--. 1 nobody nobody 0 Dec  1 18:06 ima -> integrity/ima
drwxr-xr-x. 3 nobody nobody 0 Dec  1 18:06 integrity

sh-5.1# ls -l /sys/kernel/security/ima/
total 0
-r--r-----. 1 root root 0 Dec  1 18:06 ascii_runtime_measurements
-r--r-----. 1 root root 0 Dec  1 18:06 binary_runtime_measurements
-rw-------. 1 root root 0 Dec  1 18:06 policy
-r--r-----. 1 root root 0 Dec  1 18:06 runtime_measurements_count
-r--r-----. 1 root root 0 Dec  1 18:06 violations

The nobody's are obviously sufficient to cd into the directories, but 
for file accesses I wanted to see root and no changes to permissions.

     Stefan

>
> James
>
>