Received-SPF: pass (google.com: domain of linux-kernel+bounces-42162-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Message-ID: <dbf0ef61-355b-4dcb-8e51-9298cf847367@I-love.SAKURA.ne.jp>
Date: Mon, 29 Jan 2024 13:46:28 +0900
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH 1/3] LSM: add security_bprm_aborting_creds() hook
Content-Language: en-US
To: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
        Kees Cook <keescook@chromium.org>,
        Alexander Viro <viro@zeniv.linux.org.uk>,
        Christian Brauner <brauner@kernel.org>, Jan Kara <jack@suse.cz>,
        Paul Moore <paul@paul-moore.com>, James Morris <jmorris@namei.org>,
        "Serge E. Hallyn" <serge@hallyn.com>,
        linux-security-module@vger.kernel.org, linux-fsdevel@vger.kernel.org,
        LKML <linux-kernel@vger.kernel.org>
References: <e938c37b-d615-4be4-a2da-02b904b7072f@I-love.SAKURA.ne.jp>
 <613a54d2-9508-4f87-a163-a25a77a101cd@I-love.SAKURA.ne.jp>
 <87frygbx04.fsf@email.froward.int.ebiederm.org>
From: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
In-Reply-To: <87frygbx04.fsf@email.froward.int.ebiederm.org>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit

On 2024/01/29 13:10, Eric W. Biederman wrote:
>> @@ -1519,6 +1519,7 @@ static void free_bprm(struct linux_binprm *bprm)
>>  	}
>>  	free_arg_pages(bprm);
>>  	if (bprm->cred) {
>> +		security_bprm_aborting_creds(bprm);
>>  		mutex_unlock(&current->signal->cred_guard_mutex);
>>  		abort_creds(bprm->cred);
> 
> Why isn't abort_creds calling security_free_cred enough here?

Because security_cred_free() from put_cred_rcu() is called from RCU callback
rather than from current thread doing execve().
TOMOYO wants to restore attributes of current thread doing execve().

> The fact that somewhere Tomoyo is modifying a credential that the rest
> of the kernel sees as read-only, and making it impossible to just
> restore that credential is very concerning from a maintenance
> perspective.

TOMOYO does not use "struct cred"->security.
TOMOYO uses only "struct task_struct"->security.

  struct lsm_blob_sizes tomoyo_blob_sizes __ro_after_init = {
      .lbs_task = sizeof(struct tomoyo_task),
  };

TOMOYO uses security_task_alloc() for allocating "struct task_struct"->security,
security_task_free() for releasing "struct task_struct"->security,
security_bprm_check() for updating "struct task_struct"->security,
security_bprm_committed_creds() for erasing old "struct task_struct"->security,
security_bprm_aborting_creds() for restoring old "struct task_struct"->security.

Commit a6f76f23d297 ("CRED: Make execve() take advantage of copy-on-write
credentials") made TOMOYO impossible to do above. current->in_execve flag was a
hack for emulating security_bprm_aborting_creds() using security_prepare_creds().

> Can't Tomoyo simply allow reading of files that have __FMODE_EXEC
> set when allow_execve is set, without needing to perform a domain
> transition, and later back out that domain transition?

No. That does not match TOMOYO's design.

allow_execve keyword does not imply "allow opening that file for non-execve() purpose".

Also, performing a domain transition before execve() reaches point of no return is
the TOMOYO's design, but COW credentials does not allow such behavior.