On Tue, 12 Nov 2024 17:04:04 +0100, Takashi Iwai wrote:
> On Wed, 06 Nov 2024 03:15:49 +0100,
> Edward Adam Davis wrote:
> >
> > task 1: snd ctrl will add card_dev ref count and can't call close to dec it,
> > it is blocked waiting for task 2 to release the USB dev lock.
> >
> > task 2: usb dev lock has been locked by hung task (here is usb_disconnect),
> > it is hung waiting for task 1 to exit and release card_dev.
> >
> > Adjust the USB lock acquisition method to non-blocking in ioctl to avoid
> > hang when the USB connection is closed.
>
> I'm afraid that this change would break things too badly.
> i.e. changing the blocking behavior to non-blocking is no-go.
>
> > Reported-and-tested-by: syzbot+73582d08864d8268b6fd@xxxxxxxxxxxxxxxxxxxxxxxxx
> > Closes: https://syzkaller.appspot.com/bug?extid=73582d08864d8268b6fd
>
> This particular syzkaller entry can be fixed rather by replacing
> snd_card_free() in snd_usx2y_disconnect() with
> snd_card_free_when_closed() like other USB audio drivers, something
> like below.
>
> Judging from the git log, it had been with snd_card_free_in_thread(),
> but was switch to snd_card_free() around year 2005. Meanwhile the
> handling of async card release got improved, and it's very likely OK
> to use snd_card_free_when_closed() there with the recent kernel.
The snd_card instance will be released in snd_card_do_free().
So, if snd_card_free_when_closed() is used to replace snd_card_free(), who will release the snd_card instance?
BR,
Edward
>
>
> thanks,
>
> Takashi
>
> -- 8< --
> --- a/sound/usb/usx2y/usbusx2y.c
> +++ b/sound/usb/usx2y/usbusx2y.c
> @@ -422,7 +422,7 @@ static void snd_usx2y_disconnect(struct usb_interface *intf)
> }
> if (usx2y->us428ctls_sharedmem)
> wake_up(&usx2y->us428ctls_wait_queue_head);
> - snd_card_free(card);
> + snd_card_free_when_closed(card);
> }
>
> static int snd_usx2y_probe(struct usb_interface *intf,
[Index of Archives]
[Pulseaudio]
[Linux Audio Users]
[ALSA Devel]
[Fedora Desktop]
[Fedora SELinux]
[Big List of Linux Books]
[Yosemite News]
[KDE Users]