linux-kernel - Re: [PATCH bpf-next 2/2] selftests/bpf: Extend test fs

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: <E80730DC-9389-41F7-A46D-136495C1E82D@fb.com>
Date: Wed, 16 Oct 2024 16:38:03 +0000
From: Song Liu <songliubraving@...a.com>
To: Christian Brauner <brauner@...nel.org>
CC: Jan Kara <jack@...e.cz>, Song Liu <songliubraving@...a.com>,
        Christoph
 Hellwig <hch@...radead.org>, Song Liu <song@...nel.org>,
        bpf
	<bpf@...r.kernel.org>,
        Linux-Fsdevel <linux-fsdevel@...r.kernel.org>,
        LKML
	<linux-kernel@...r.kernel.org>,
        Kernel Team <kernel-team@...a.com>,
        Andrii
 Nakryiko <andrii@...nel.org>,
        Eduard Zingerman <eddyz87@...il.com>,
        Alexei
 Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Martin
 KaFai Lau <martin.lau@...ux.dev>,
        Al Viro <viro@...iv.linux.org.uk>, KP Singh
	<kpsingh@...nel.org>,
        Matt Bobrowski <mattbobrowski@...gle.com>
Subject: Re: [PATCH bpf-next 2/2] selftests/bpf: Extend test fs_kfuncs to
 cover security.bpf xattr names



> On Oct 16, 2024, at 7:51 AM, Christian Brauner <brauner@...nel.org> wrote:
> 
> On Wed, Oct 16, 2024 at 03:51:55PM +0200, Jan Kara wrote:
>> On Tue 15-10-24 05:52:02, Song Liu wrote:
>>>> On Oct 14, 2024, at 10:25 PM, Christoph Hellwig <hch@...radead.org> wrote:
>>>> On Tue, Oct 15, 2024 at 05:21:48AM +0000, Song Liu wrote:
>>>>>>> Extend test_progs fs_kfuncs to cover different xattr names. Specifically:
>>>>>>> xattr name "user.kfuncs", "security.bpf", and "security.bpf.xxx" can be
>>>>>>> read from BPF program with kfuncs bpf_get_[file|dentry]_xattr(); while
>>>>>>> "security.bpfxxx" and "security.selinux" cannot be read.
>>>>>> 
>>>>>> So you read code from untrusted user.* xattrs?  How can you carve out
>>>>>> that space and not known any pre-existing userspace cod uses kfuncs
>>>>>> for it's own purpose?
>>>>> 
>>>>> I don't quite follow the comment here. 
>>>>> 
>>>>> Do you mean user.* xattrs are untrusted (any user can set it), so we 
>>>>> should not allow BPF programs to read them? Or do you mean xattr 
>>>>> name "user.kfuncs" might be taken by some use space?
>>>> 
>>>> All of the above.
>>> 
>>> This is a selftest, "user.kfunc" is picked for this test. The kfuncs
>>> (bpf_get_[file|dentry]_xattr) can read any user.* xattrs. 
>>> 
>>> Reading untrusted xattrs from trust BPF LSM program can be useful. 
>>> For example, we can sign a binary with private key, and save the
>>> signature in the xattr. Then the kernel can verify the signature
>>> and the binary matches the public key. If the xattr is modified by
>>> untrusted user space, the BPF program will just deny the access.
>> 
>> So I tend to agree with Christoph that e.g. for the above LSM usecase you
>> mention, using user. xattr space is a poor design choice because you have
>> to very carefully validate any xattr contents (anybody can provide
>> malicious content) and more importantly as different similar usecases
>> proliferate the chances of name collisions and resulting funcionality
>> issues increase. It is similar as if you decided to store some information
>> in a specially named file in each directory. If you choose special enough
>> name, it will likely work but long-term someone is going to break you :)

Yes, with user.* xattr, name conflicts is always an issue. That's why we 
are adding the security.bpf prefix in this set. 

However, besides name conflicts, I don't think there are many more issues
with using user. xattrs. VFS code does not block any access to the
security.* xattrs. It is up to the LSMs to block read/write of certain
xattrs. IOW, if the LSM writer decide to use user.xxx for some use cases, 
it is up to LSM writer to protect this xattr from unauthorized access 
(via security_inode_setxattr hook). 

>> 
>> I think that getting user.* xattrs from bpf hooks can still be useful for
>> introspection and other tasks so I'm not convinced we should revert that
>> functionality but maybe it is too easy to misuse? I'm not really decided.
> 
> Reading user.* xattr is fine. If an LSM decides to built a security
> model around it then imho that's their business and since that happens
> in out-of-tree LSM programs: shrug.

Thanks,
Song