Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 8 Oct 2020 23:27:53 +0100
From:   Matthew Wilcox <willy@infradead.org>
To:     syzbot <syzbot+cdcbdc0bd42e559b52b9@syzkaller.appspotmail.com>
Cc:     axboe@kernel.dk, io-uring@vger.kernel.org,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
        syzkaller-bugs@googlegroups.com, viro@zeniv.linux.org.uk
Subject: Re: inconsistent lock state in xa_destroy
Message-ID: <20201008222753.GP20115@casper.infradead.org>
References: <00000000000045ac4605b12a1720@google.com>
 <000000000000c35f0805b12f5099@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <000000000000c35f0805b12f5099@google.com>
Precedence: bulk


If I understand the lockdep report here, this actually isn't an XArray
issue, although I do think there is one.

On Thu, Oct 08, 2020 at 02:14:20PM -0700, syzbot wrote:
> ================================
> WARNING: inconsistent lock state
> 5.9.0-rc8-next-20201008-syzkaller #0 Not tainted
> --------------------------------
> inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
> swapper/0/0 [HC0[0]:SC1[1]:HE0:SE0] takes:
> ffff888025f65018 (&xa->xa_lock#7){+.?.}-{2:2}, at: xa_destroy+0xaa/0x350 lib/xarray.c:2205
> {SOFTIRQ-ON-W} state was registered at:
>   lock_acquire+0x1f2/0xaa0 kernel/locking/lockdep.c:5419
>   __raw_spin_lock include/linux/spinlock_api_smp.h:142 [inline]
>   _raw_spin_lock+0x2a/0x40 kernel/locking/spinlock.c:151
>   spin_lock include/linux/spinlock.h:354 [inline]
>   io_uring_add_task_file fs/io_uring.c:8607 [inline]

You're using the XArray in a non-interrupt-disabling mode.

>  _raw_spin_lock_irqsave+0x94/0xd0 kernel/locking/spinlock.c:159
>  xa_destroy+0xaa/0x350 lib/xarray.c:2205
>  __io_uring_free+0x60/0xc0 fs/io_uring.c:7693
>  io_uring_free include/linux/io_uring.h:40 [inline]
>  __put_task_struct+0xff/0x3f0 kernel/fork.c:732
>  put_task_struct include/linux/sched/task.h:111 [inline]
>  delayed_put_task_struct+0x1f6/0x340 kernel/exit.c:172
>  rcu_do_batch kernel/rcu/tree.c:2484 [inline]

But you're calling xa_destroy() from in-interrupt context.
So (as far as lockdep is concerned), no matter what I do in
xa_destroy(), this potential deadlock is there.  You'd need to be
using xa_init_flags(XA_FLAGS_LOCK_IRQ) if you actually needed to call
xa_destroy() here.

Fortunately, it seems you don't need to call xa_destroy() at all, so
that problem is solved, but the patch I have here wouldn't help.