[PATCH 69/90] LSM: Use full security context in security_inode_setsecctx

[Tetsuo Handa]

On 2019/04/23 5:45, Casey Schaufler wrote:
> On 4/22/2019 6:13 AM, Tetsuo Handa wrote:
>> On 2019/04/19 9:45, Casey Schaufler wrote:
>>> +    hlist_for_each_entry(hp, &security_hook_heads.inode_setsecctx, list) {
>>> +        if (strncmp(ctx, hp->lsm, strlen(hp->lsm))) {
>>> +            WARN_ONCE(1, "security_inode_setsecctx form1 error\n");
>>> +            rc = -EINVAL;
>>> +            break;
>>> +        }
>> Will you avoid using WARN*() ?
>> Since syzbot tests using panic_on_warn == 1, this WARN_ONCE() will act as panic().
> 
> If syzbot hits any of the WARN_ONCE()s in security_inode_setsecctx()
> I want it to panic and generate a report. A badly formatted inode secctx
> would indicate that kernfs isn't getting the string from
> security_inode_getsecctx() or that it is getting corrupted somehow. In
> either case, it would be a bug that needs fixing. I used WARN instead of
> BUG for the kernfs people, who might break something by accident.

Since the code continues with -EINVAL error, I assumed that this is not
a bad situation. But if this can't be triggered by invalid input from
userspace, BUG() is better.

> 
> If there's a strong objection to WARN_ONCE() in general, I can pull it.
>