Re: [syzbot] [io-uring?] KCSAN: data-race in io_wq_activate_free_worker / io_wq_worker_running

[Marco Elver <elver@google.com>]

On Wed, 13 Sept 2023 at 14:13, Pavel Begunkov <asml.silence@gmail.com> wrote:
>
> On 9/13/23 12:29, syzbot wrote:
> > Hello,
> >
> > syzbot found the following issue on:
> >
> > HEAD commit:    f97e18a3f2fb Merge tag 'gpio-updates-for-v6.6' of git://gi..
> > git tree:       upstream
> > console output: https://syzkaller.appspot.com/x/log.txt?x=12864667a80000
> > kernel config:  https://syzkaller.appspot.com/x/.config?x=fe440f256d065d3b
> > dashboard link: https://syzkaller.appspot.com/bug?extid=a36975231499dc24df44
> > compiler:       Debian clang version 15.0.6, GNU ld (GNU Binutils for Debian) 2.40
> >
> > Unfortunately, I don't have any reproducer for this issue yet.
> >
> > Downloadable assets:
> > disk image: https://storage.googleapis.com/syzbot-assets/b1781aaff038/disk-f97e18a3.raw.xz
> > vmlinux: https://storage.googleapis.com/syzbot-assets/5b915468fd6d/vmlinux-f97e18a3.xz
> > kernel image: https://storage.googleapis.com/syzbot-assets/abc8ece931f3/bzImage-f97e18a3.xz
> >
> > IMPORTANT: if you fix the issue, please add the following tag to the commit:
> > Reported-by: syzbot+a36975231499dc24df44@syzkaller.appspotmail.com
> >
> > ==================================================================
> > BUG: KCSAN: data-race in io_wq_activate_free_worker / io_wq_worker_running
> >
> > write to 0xffff888127f736c4 of 4 bytes by task 4731 on cpu 1:
> >   io_wq_worker_running+0x64/0xa0 io_uring/io-wq.c:668
> >   schedule_timeout+0xcc/0x230 kernel/time/timer.c:2167
> >   io_wq_worker+0x4b2/0x840 io_uring/io-wq.c:633
> >   ret_from_fork+0x2e/0x40 arch/x86/kernel/process.c:145
> >   ret_from_fork_asm+0x11/0x20 arch/x86/entry/entry_64.S:304
> >
> > read to 0xffff888127f736c4 of 4 bytes by task 4719 on cpu 0:
> >   io_wq_get_acct io_uring/io-wq.c:168 [inline]
> >   io_wq_activate_free_worker+0xfa/0x280 io_uring/io-wq.c:267
> >   io_wq_enqueue+0x262/0x450 io_uring/io-wq.c:914
>
> 1) the worst case scenario we'll choose a wrong type of
> worker, which is inconsequential.
>
> 2) we're changing the IO_WORKER_F_RUNNING bit, but checking
> for IO_WORKER_F_BOUND. The latter one is set at the very
> beginning, it would require compiler to be super inventive
> to actually hit the problem.
>
> I don't believe it's a problem, but it'll nice to attribute
> it properly, READ_ONCE?, or split IO_WORKER_F_BOUND out into
> a separate field.

It's a simple bit flag set & read, I'd go for READ_ONCE() (and
WRITE_ONCE() - but up to you, these bitflag sets & reads have been ok
with just the READ_ONCE(), and KCSAN currently doesn't care if there's
a WRITE_ONCE() or not).

> value changed: 0x0000000d -> 0x0000000b

This is interesting though - it says that it observed 2 bits being
flipped. We don't see where IO_WORKER_F_FREE was unset though.