Re: [PATCH 11/13] AppArmor: LSM interface, and security module initialization

[John Johansen <john.johansen@canonical.com>]

On 07/15/2010 10:27 AM, Serge E. Hallyn wrote:
> Quoting John Johansen (john.johansen@canonical.com):
>> AppArmor hooks to interface with the LSM, module parameters and module
>> initialization.
>>
>> Signed-off-by: John Johansen <john.johansen@canonical.com>
> 
> Thanks, John - looks good overall.  Comments:
> 
> ...
> 
>> +static int apparmor_ptrace_access_check(struct task_struct *child,
>> +					unsigned int mode)
>> +{
>> +	int rc;
>> +
>> +	rc = cap_ptrace_access_check(child, mode);
>> +	if (rc)
>> +		return rc;
>> +
>> +	return aa_ptrace(current, child, mode);
>> +}
>> +
>> +static int apparmor_ptrace_traceme(struct task_struct *parent)
>> +{
> 
> Just curious - why aren't you calling cap_ptrace_traceme() first here?
> 
err, we should be.  I'm not sure where that got dropped.  I'll go through
and re audit all of these.

thanks
>> +	return aa_ptrace(parent, current, PTRACE_MODE_ATTACH);
>> +}
>> +
>> +/* Derived from security/commoncap.c:cap_capget */
>> +static int apparmor_capget(struct task_struct *target, kernel_cap_t *effective,
>> +			   kernel_cap_t *inheritable, kernel_cap_t *permitted)
>> +{
>> +	struct aa_profile *profile;
>> +	const struct cred *cred;
>> +
>> +	rcu_read_lock();
>> +	cred = __task_cred(target);
>> +	profile = aa_cred_profile(cred);
>> +
>> +	*effective = cred->cap_effective;
>> +	*inheritable = cred->cap_inheritable;
>> +	*permitted = cred->cap_permitted;
>> +
>> +	if (!unconfined(profile))
>> +		*effective = cap_intersect(*effective, profile->caps.allow);
> 
> Should you mask permitted too?  Otherwise you might confuse a userspace
> lib which assumes it's caller previously culled pE, and that it can
> nwo refill it from pP.
> 
yes indeed thanks

>> +	rcu_read_unlock();
>> +
>> +	return 0;
>> +}
>> +
>> +static int apparmor_capable(struct task_struct *task, const struct cred *cred,
>> +			    int cap, int audit)
>> +{
>> +	struct aa_profile *profile;
>> +	/* cap_capable returns 0 on success, else -EPERM */
>> +	int error = cap_capable(task, cred, cap, audit);
> 
> jinkeys, it might be just me, but i'd have spend 2 mins less looking
> at this if you'd done
> 
> 	if (error)
> 		return error;
> 
> here, simplifying the condition below.
> 
>> +
>> +	profile = aa_cred_profile(cred);
>> +	if (!error  && !unconfined(profile))
>> +		error = aa_capable(task, profile, cap, audit);
>> +
>> +	return error;
>> +}
> 
yeah, that is better

thanks Serge