Re: [PATCH 1/2] fcntl: fix potential deadlocks for &fown_struct.lock

[Desmond Cheong Zhi Xi <desmondcheongzx@gmail.com>]

On 2/7/21 10:27 pm, Jeff Layton wrote:
> On Fri, 2021-07-02 at 21:55 +0800, Desmond Cheong Zhi Xi wrote:
>> On 2/7/21 7:44 pm, Jeff Layton wrote:
>>> On Fri, 2021-07-02 at 17:18 +0800, Desmond Cheong Zhi Xi wrote:
>>>> Syzbot reports a potential deadlock in do_fcntl:
>>>>
>>>> ========================================================
>>>> WARNING: possible irq lock inversion dependency detected
>>>> 5.12.0-syzkaller #0 Not tainted
>>>> --------------------------------------------------------
>>>> syz-executor132/8391 just changed the state of lock:
>>>> ffff888015967bf8 (&f->f_owner.lock){.+..}-{2:2}, at: f_getown_ex fs/fcntl.c:211 [inline]
>>>> ffff888015967bf8 (&f->f_owner.lock){.+..}-{2:2}, at: do_fcntl+0x8b4/0x1200 fs/fcntl.c:395
>>>> but this lock was taken by another, HARDIRQ-safe lock in the past:
>>>>    (&dev->event_lock){-...}-{2:2}
>>>>
>>>> and interrupts could create inverse lock ordering between them.
>>>>
>>>> other info that might help us debug this:
>>>> Chain exists of:
>>>>     &dev->event_lock --> &new->fa_lock --> &f->f_owner.lock
>>>>
>>>>    Possible interrupt unsafe locking scenario:
>>>>
>>>>          CPU0                    CPU1
>>>>          ----                    ----
>>>>     lock(&f->f_owner.lock);
>>>>                                  local_irq_disable();
>>>>                                  lock(&dev->event_lock);
>>>>                                  lock(&new->fa_lock);
>>>>     <Interrupt>
>>>>       lock(&dev->event_lock);
>>>>
>>>>    *** DEADLOCK ***
>>>>
>>>> This happens because there is a lock hierarchy of
>>>> &dev->event_lock --> &new->fa_lock --> &f->f_owner.lock
>>>> from the following call chain:
>>>>
>>>>     input_inject_event():
>>>>       spin_lock_irqsave(&dev->event_lock,...);
>>>>       input_handle_event():
>>>>         input_pass_values():
>>>>           input_to_handler():
>>>>             evdev_events():
>>>>               evdev_pass_values():
>>>>                 spin_lock(&client->buffer_lock);
>>>>                 __pass_event():
>>>>                   kill_fasync():
>>>>                     kill_fasync_rcu():
>>>>                       read_lock(&fa->fa_lock);
>>>>                       send_sigio():
>>>>                         read_lock_irqsave(&fown->lock,...);
>>>>
>>>> However, since &dev->event_lock is HARDIRQ-safe, interrupts have to be
>>>> disabled while grabbing &f->f_owner.lock, otherwise we invert the lock
>>>> hierarchy.
>>>>
>>>> Hence, we replace calls to read_lock/read_unlock on &f->f_owner.lock,
>>>> with read_lock_irq/read_unlock_irq.
>>>>
>>>
>>> Patches look reasonable overall, but why does this one use read_lock_irq
>>> and the other one use read_lock_irqsave? Don't we need to *_irqsasve in
>>> both patches?
>>>
>>>
>>
>> My thinking was that the functions f_getown_ex and f_getowner_uids are
>> only called from do_fcntl, and f_getown is only called from do_fnctl and
>> sock_ioctl. do_fnctl itself is only called from syscalls.
>>
>> For sock_ioctl, the chain is
>>     compat_sock_ioctl():
>>       compat_sock_ioctl_trans():
>>         sock_ioctl()
>>
>> For both paths, it doesn't seem that interrupts are disabled, so I used
>> the *irq variants.
>>
>> But of course, I might be very mistaken on this, and I'd be happy to
>> make the change to *_irqsave.
>>
>> Also, on further inspection, if these calls should be changed to
>> *_irqsave, then I believe the call to write_lock_irq in f_modown (called
>> from do_fcntl() --> f_setown() --> __f_setown() --> f_modown()) should
>> also be changed to *_irqsave.
>>
>> There's also a call to write_lock_irq(&fa->fa_lock) in
>> fasync_remove_entry and fasync_insert_entry. Whether these should be
>> changed as well isn't as clear to me, but since it's safe to do, perhaps
>> it makes sense to use *_irqsave for them too. Thoughts?
>>
> 
> 
> I think your reasoning is probably valid here and we don't need to
> save/restore. It wasn't obvious to me until you pointed it out though.
> It might be worth a comment, or maybe even this at the top of both
> functions:
> 
>      WARN_ON_ONCE(irqs_disabled());
> 

Adding the WARN_ON_ONCE makes sense. I'll test it with Syzbot then 
prepare a v2 series.

> I'll pick these into linux-next soon and plan to merge them for v5.15.
> Let me know if you think they need to go in sooner.
> 
> 

Sounds good to me. Thanks for the feedback, Jeff.

>>>> Reported-and-tested-by: syzbot+e6d5398a02c516ce5e70@syzkaller.appspotmail.com
>>>> Signed-off-by: Desmond Cheong Zhi Xi <desmondcheongzx@gmail.com>
>>>> ---
>>>>    fs/fcntl.c | 13 +++++++------
>>>>    1 file changed, 7 insertions(+), 6 deletions(-)
>>>>
>>>> diff --git a/fs/fcntl.c b/fs/fcntl.c
>>>> index dfc72f15be7f..cf9e81dfa615 100644
>>>> --- a/fs/fcntl.c
>>>> +++ b/fs/fcntl.c
>>>> @@ -150,7 +150,8 @@ void f_delown(struct file *filp)
>>>>    pid_t f_getown(struct file *filp)
>>>>    {
>>>>    	pid_t pid = 0;
>>>> -	read_lock(&filp->f_owner.lock);
>>>> +
>>>> +	read_lock_irq(&filp->f_owner.lock);
>>>>    	rcu_read_lock();
>>>>    	if (pid_task(filp->f_owner.pid, filp->f_owner.pid_type)) {
>>>>    		pid = pid_vnr(filp->f_owner.pid);
>>>> @@ -158,7 +159,7 @@ pid_t f_getown(struct file *filp)
>>>>    			pid = -pid;
>>>>    	}
>>>>    	rcu_read_unlock();
>>>> -	read_unlock(&filp->f_owner.lock);
>>>> +	read_unlock_irq(&filp->f_owner.lock);
>>>>    	return pid;
>>>>    }
>>>>    
>>>> @@ -208,7 +209,7 @@ static int f_getown_ex(struct file *filp, unsigned long arg)
>>>>    	struct f_owner_ex owner = {};
>>>>    	int ret = 0;
>>>>    
>>>> -	read_lock(&filp->f_owner.lock);
>>>> +	read_lock_irq(&filp->f_owner.lock);
>>>>    	rcu_read_lock();
>>>>    	if (pid_task(filp->f_owner.pid, filp->f_owner.pid_type))
>>>>    		owner.pid = pid_vnr(filp->f_owner.pid);
>>>> @@ -231,7 +232,7 @@ static int f_getown_ex(struct file *filp, unsigned long arg)
>>>>    		ret = -EINVAL;
>>>>    		break;
>>>>    	}
>>>> -	read_unlock(&filp->f_owner.lock);
>>>> +	read_unlock_irq(&filp->f_owner.lock);
>>>>    
>>>>    	if (!ret) {
>>>>    		ret = copy_to_user(owner_p, &owner, sizeof(owner));
>>>> @@ -249,10 +250,10 @@ static int f_getowner_uids(struct file *filp, unsigned long arg)
>>>>    	uid_t src[2];
>>>>    	int err;
>>>>    
>>>> -	read_lock(&filp->f_owner.lock);
>>>> +	read_lock_irq(&filp->f_owner.lock);
>>>>    	src[0] = from_kuid(user_ns, filp->f_owner.uid);
>>>>    	src[1] = from_kuid(user_ns, filp->f_owner.euid);
>>>> -	read_unlock(&filp->f_owner.lock);
>>>> +	read_unlock_irq(&filp->f_owner.lock);
>>>>    
>>>>    	err  = put_user(src[0], &dst[0]);
>>>>    	err |= put_user(src[1], &dst[1]);
>>>
>>
>