On Mon, Apr 25, 2022 at 2:18 PM Lukas Wunner <lukas@xxxxxxxxx> wrote: > > On Mon, Apr 25, 2022 at 07:41:46AM -0700, Jakub Kicinski wrote: > > On Sat, 23 Apr 2022 18:07:23 +0200 Lukas Wunner wrote: > > > > Looking at the original report it looks like the issue could be > > > > resolved with a more usb-specific change: e.g. it looks like > > > > usbnet_defer_kevent() is not acquiring a dev reference as it should. > > > > > > > > Have you considered that path? > > > > > > First of all, the diffstat of the patch shows this is an opportunity > > > to reduce LoC as well as simplify and speed up device teardown. > > > > > > Second, the approach you're proposing won't work if a driver calls > > > netif_carrier_on/off() after unregister_netdev(). > > > > > > It seems prudent to prevent such a misbehavior in *any* driver, > > > not just usbnet. usbnet may not be the only one doing it wrong. > > > Jann pointed out that there are more syzbot reports related > > > to a UAF in linkwatch: > > > > > > https://lore.kernel.org/netdev/?q=__linkwatch_run_queue+syzbot > > > > > > Third, I think an API which schedules work, invisibly to the driver, > > > is dangerous and misguided. If it is illegal to call > > > netif_carrier_on/off() for an unregistered but not yet freed netdev, > > > catch that in core networking code and don't expect drivers to respect > > > a rule which isn't even documented. > > > > Doesn't mean we should make it legal. We can add a warning to catch > > abuses. > > That would be inconsequent, considering that netif_carrier_on/off() > do not warn for a reg_state of NETREG_UNINITIALIZED. > Yes, only 1500 calls to audit ;) I guess we could start adding WARN_ON_ONCE(), then wait for a few syzbot/users reports to fix offenders... commit b47300168e770b60ab96c8924854c3b0eb4260eb Author: David S. Miller <davem@xxxxxxxxxxxxx> Date: Wed Nov 19 15:33:54 2008 -0800 net: Do not fire linkwatch events until the device is registered. Several device drivers try to do things like netif_carrier_off() before register_netdev() is invoked. This is bogus, but too many drivers do this to fix them all up in one go. Reported-by: Folkert van Heusden <folkert@xxxxxxxxxxxxxx> Signed-off-by: David S. Miller <davem@xxxxxxxxxxxxx>