[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ff23bcb5-d2e8-4b1b-a669-feab4a97994a@redhat.com>
Date: Tue, 10 Sep 2024 11:58:56 +0200
From: Paolo Abeni <pabeni@...hat.com>
To: Oliver Neukum <oneukum@...e.com>, davem@...emloft.net,
edumazet@...gle.com, kuba@...nel.org, netdev@...r.kernel.org,
linux-usb@...r.kernel.org, linux-kernel@...r.kernel.org
Cc: stable@...r.kernel.org
Subject: Re: [PATCHv2 net] usbnet: fix cyclical race on disconnect with work
queue
On 9/5/24 15:46, Oliver Neukum wrote:
> The work can submit URBs and the URBs can schedule the work.
> This cycle needs to be broken, when a device is to be stopped.
> Use a flag to do so.
> This is a design issue as old as the driver.
>
> Signed-off-by: Oliver Neukum <oneukum@...e.com>
> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
> CC: stable@...r.kernel.org
> ---
>
> v2: fix PM reference issue
>
> drivers/net/usb/usbnet.c | 37 ++++++++++++++++++++++++++++---------
> include/linux/usb/usbnet.h | 17 +++++++++++++++++
> 2 files changed, 45 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/net/usb/usbnet.c b/drivers/net/usb/usbnet.c
> index 18eb5ba436df..2506aa8c603e 100644
> --- a/drivers/net/usb/usbnet.c
> +++ b/drivers/net/usb/usbnet.c
> @@ -464,10 +464,15 @@ static enum skb_state defer_bh(struct usbnet *dev, struct sk_buff *skb,
> void usbnet_defer_kevent (struct usbnet *dev, int work)
> {
> set_bit (work, &dev->flags);
> - if (!schedule_work (&dev->kevent))
> - netdev_dbg(dev->net, "kevent %s may have been dropped\n", usbnet_event_names[work]);
> - else
> - netdev_dbg(dev->net, "kevent %s scheduled\n", usbnet_event_names[work]);
> + if (!usbnet_going_away(dev)) {
> + if (!schedule_work(&dev->kevent))
> + netdev_dbg(dev->net,
> + "kevent %s may have been dropped\n",
> + usbnet_event_names[work]);
> + else
> + netdev_dbg(dev->net,
> + "kevent %s scheduled\n", usbnet_event_names[work]);
> + }
> }
> EXPORT_SYMBOL_GPL(usbnet_defer_kevent);
>
> @@ -535,7 +540,8 @@ static int rx_submit (struct usbnet *dev, struct urb *urb, gfp_t flags)
> tasklet_schedule (&dev->bh);
> break;
> case 0:
> - __usbnet_queue_skb(&dev->rxq, skb, rx_start);
> + if (!usbnet_going_away(dev))
> + __usbnet_queue_skb(&dev->rxq, skb, rx_start);
> }
> } else {
> netif_dbg(dev, ifdown, dev->net, "rx: stopped\n");
> @@ -843,9 +849,18 @@ int usbnet_stop (struct net_device *net)
>
> /* deferred work (timer, softirq, task) must also stop */
> dev->flags = 0;
> - del_timer_sync (&dev->delay);
> - tasklet_kill (&dev->bh);
> + del_timer_sync(&dev->delay);
> + tasklet_kill(&dev->bh);
> cancel_work_sync(&dev->kevent);
> +
> + /* We have cyclic dependencies. Those calls are needed
> + * to break a cycle. We cannot fall into the gaps because
> + * we have a flag
> + */
> + tasklet_kill(&dev->bh);
> + del_timer_sync(&dev->delay);
> + cancel_work_sync(&dev->kevent);
I guess you do the shutdown twice because a running tasklet or timer
could re-schedule the others? If so, what prevent the rescheduling to
happen in the 2nd iteration? why can't you add usbnet_going_away()
checks on tasklet and timer reschedule point?
Thanks,
Paolo
Powered by blists - more mailing lists