Re: lgetxattr()/getxattr() return different values on a file labelled with selinux disabled

[Stephen Smalley <sds@tycho.nsa.gov>]

On 03/15/2013 11:24 AM, Thomas COUDRAY wrote:
> 2013/3/15 Stephen Smalley <sds@tycho.nsa.gov>:
>> f is truly a regular file and not a symlink pointing to a regular file?
>
> f is a truly regular file.
>
>> before_t and after_t are both defined in the policy?
>
> Only before_t was defined in the policy.

If not defined in policy, then kernel should remap to unlabeled sid 
context.

> When I define after_t in the policy, both commands return the same
> label (after_t).
> But I wouldn't expect this to make a difference in the output of both
> commands (as the only visible difference is lgetxattr() vs getxattr())

getxattr security.* results are supplied by the security module rather 
than the filesystem to allow the value to be canonicalized.  But this 
should happen the same for lgetxattr and getxattr; those should only 
differ if the file is a symlink.

>> before_t and after_t are not type aliases of each other?
>
> They are not.
>
>> What are the credentials (capabilities and SELinux security
>> context/permissions) of the process running the ls and getfattr commands?
>
> It has unconfined_u:unconfined_r:before_t label with before_t type.
> Same as the file f.
> The process has full SELinux rights on both command and file.

Did it run as root?  Does it have :capability2 mac_override permission?

>> Any relevant messages from SELinux in dmesg output?
>
> No avc warnings in dmesg and audit.log. All looks good.

What about SELinux: messages?  e.g. SELinux:  Context ... is not valid 
(left unmapped).