Messages in this thread Patch in this message | | | Date | Fri, 17 Nov 2023 10:45:31 +0100 | From | Peter Zijlstra <> | Subject | Re: [syzbot] [kernel?] inconsistent lock state in ptrace_attach |
| |
On Fri, Nov 17, 2023 at 10:38:30AM +0100, Oleg Nesterov wrote: > On 11/17, Peter Zijlstra wrote: > > > > On Fri, Nov 17, 2023 at 12:49:04AM -0800, syzbot wrote: > > > syzbot has bisected this issue to: > > > > > > commit 2d25a889601d2fbc87ec79b30ea315820f874b78 > > > Author: Peter Zijlstra <peterz@infradead.org> > > > Date: Sun Sep 17 11:24:21 2023 +0000 > > > > > > ptrace: Convert ptrace_attach() to use lock guards > > > > > > bisection log: https://syzkaller.appspot.com/x/bisect.txt?x=132b7d84e80000 > > > start commit: f31817cbcf48 Add linux-next specific files for 20231116 > > > git tree: linux-next > > > final oops: https://syzkaller.appspot.com/x/report.txt?x=10ab7d84e80000 > > > console output: https://syzkaller.appspot.com/x/log.txt?x=172b7d84e80000 > > > kernel config: https://syzkaller.appspot.com/x/.config?x=f59345f1d0a928c > > > dashboard link: https://syzkaller.appspot.com/bug?extid=cbb25bb9b4d29a773985 > > > syz repro: https://syzkaller.appspot.com/x/repro.syz?x=1014d797680000 > > > C reproducer: https://syzkaller.appspot.com/x/repro.c?x=11b1ec67680000 > > > > > > Reported-by: syzbot+cbb25bb9b4d29a773985@syzkaller.appspotmail.com > > > Fixes: 2d25a889601d ("ptrace: Convert ptrace_attach() to use lock guards") > > > > Hurmph, let me pull that patch from tip and go have a look. Thanks! > > I guess > > scoped_guard (write_lock, &tasklist_lock) > > is not right... this needs write_lock_irq(). > > I didn't notice this when I reviewed this patch.
Yep, that seems to be the case.
I keep forgetting how to tell the syscall robot to test the below delta, but I'll push the ammended commit to my staging tree and we'll see what happens.
--- --- a/kernel/ptrace.c +++ b/kernel/ptrace.c @@ -461,7 +461,7 @@ static int ptrace_attach(struct task_str return retval; } - scoped_guard (write_lock, &tasklist_lock) { + scoped_guard (write_lock_irq, &tasklist_lock) { if (unlikely(task->exit_state)) return -EPERM; if (task->ptrace)
| |