Re: [BUG] Kmemleak, possibly hiddev_connect(), in 6.3.0+ torvalds tree commit gfc4354c6e5c2

[Greg Kroah-Hartman]

On Mon, May 08, 2023 at 08:51:55AM +0200, Greg Kroah-Hartman wrote:
> On Mon, May 08, 2023 at 08:30:07AM +0200, Mirsad Goran Todorovac wrote:
> > Hi,
> > 
> > There seems to be a kernel memory leak in the USB keyboard driver.
> > 
> > The leaked memory allocs are 96 and 512 bytes.
> > 
> > The platform is Ubuntu 22.04 LTS on a assembled AMD Ryzen 9 with X670E PG
> > Lightning mobo,
> > and Genius SlimStar i220 GK-080012 keyboard.
> > 
> > (Logitech M100 HID mouse is not affected by the bug.)
> > 
> > BIOS is:
> > 
> >      *-firmware
> >           description: BIOS
> >           vendor: American Megatrends International, LLC.
> >           physical id: 0
> >           version: 1.21
> >           date: 04/26/2023
> >           size: 64KiB
> > 
> > The kernel is 6.3.0-torvalds-<id>-13466-gfc4354c6e5c2.
> > 
> > The keyboard is recognised as Chicony:
> > 
> >                  *-usb
> >                       description: Keyboard
> >                       product: CHICONY USB Keyboard
> >                       vendor: CHICONY
> >                       physical id: 2
> >                       bus info: usb@5:2
> >                       logical name: input35
> >                       logical name: /dev/input/event4
> >                       logical name: input35::capslock
> >                       logical name: input35::numlock
> >                       logical name: input35::scrolllock
> >                       logical name: input36
> >                       logical name: /dev/input/event5
> >                       logical name: input37
> >                       logical name: /dev/input/event6
> >                       logical name: input38
> >                       logical name: /dev/input/event8
> >                       version: 2.30
> >                       capabilities: usb-2.00 usb
> >                       configuration: driver=usbhid maxpower=100mA
> > speed=1Mbit/s
> > 
> > The bug is easily reproduced by unplugging the USB keyboard, waiting about a
> > couple of seconds,
> > and then reconnect and scan for memory leaks twice.
> > 
> > The kmemleak log is as follows [edited privacy info]:
> > 
> > root@hostname:/home/username# cat /sys/kernel/debug/kmemleak
> > unreferenced object 0xffff8dd020037c00 (size 96):
> >   comm "systemd-udevd", pid 435, jiffies 4294892550 (age 8909.356s)
> >   hex dump (first 32 bytes):
> >     5d 8e 4e b9 ff ff ff ff 00 00 00 00 00 00 00 00 ].N.............
> >     00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
> >   backtrace:
> >     [<ffffffffb81a74be>] __kmem_cache_alloc_node+0x22e/0x2b0
> >     [<ffffffffb8127b6e>] kmalloc_trace+0x2e/0xa0
> >     [<ffffffffb87543d9>] class_create+0x29/0x80
> >     [<ffffffffb8880d24>] usb_register_dev+0x1d4/0x2e0
> 
> As the call to class_create() in this path is now gone in 6.4-rc1, can
> you retry that release to see if this is still there or not?

No, wait, it's still there, I was looking at a development branch of
mine that isn't sent upstream yet.  And syzbot just reported the same
thing:
	https://lore.kernel.org/r/00000000000058d15f05fb264013@xxxxxxxxxx

So something's wrong here, let me dig into it tomorrow when I get a
chance...

thanks,

greg k-h