Linux Input/HID development
 help / color / mirror / Atom feed
* [PATCH] HID: corsair: cancel worker after unregistering LED, not before
@ 2026-09-04 14:36 Danish Khateeb
  2026-09-04 14:53 ` sashiko-bot
  2026-09-04 16:15 ` [PATCH v2] HID: corsair: use disable_work_sync() to tear down the LED workers Danish Khateeb
  0 siblings, 2 replies; 4+ messages in thread
From: Danish Khateeb @ 2026-09-04 14:36 UTC (permalink / raw)
  To: Jiri Kosina, Benjamin Tissoires
  Cc: Chen Changcheng, linux-input, linux-kernel

corsair_remove() tears down each LED with:

	removed = true;
	cancel_work_sync(&led->work);
	led_classdev_unregister(&led->cdev);
	kfree(...);

led_classdev_unregister() calls led_set_brightness(led_cdev, LED_OFF),
which reaches the driver's k90_brightness_set() and re-arms the worker
with schedule_work(). That happens after cancel_work_sync() has already
run, so the work item is queued again and the structure holding it is
then freed while it is still linked into the worklist.

The corruption surfaces on the next worklist insertion rather than in
the driver itself:

  BUG: KASAN: slab-use-after-free in __list_add_valid_or_report+0x186/0x210
  Read of size 8 at addr ffff888102a1ead0 by task kworker/0:1/11
    __queue_work+0xade/0x1350
    queue_work_on+0xb6/0xc0
    led_classdev_unregister+0x26b/0x340
    corsair_remove+0x1c2/0x2d0
    hid_device_remove+0xba/0x1e0
    usbhid_disconnect+0xa0/0xe0
  Allocated by task 129:
    corsair_probe+0x560/0xd50
  Freed by task 11:
    corsair_remove+0xec/0x2d0

k90 is freed in k90_cleanup_macro_functions(), and the stale
record_led.work entry left on the worklist is then tripped over by the
queue_work() that k90_cleanup_backlight() performs immediately
afterwards. The syzbot report appears to be the same bug on the other
LED, surfacing later when the worker itself runs and reads led->removed
from freed memory, but that variant was not reproduced here.

Cancel the worker after led_classdev_unregister() so that nothing can
re-arm it before the memory goes away. The removed flag is still set
first, so a worker already past the flag check returns without
dereferencing led->cdev.dev, which is what commit eb51c9f8cb4f ("HID:
corsair: cancel worker before unregistering LED to fix use-after-free")
set out to prevent. The fail_sysfs error path in
k90_init_macro_functions() already uses this ordering.

Reproduced by binding a configfs USB gadget carrying the K90's IDs
(1b1c:1b02) to dummy_hcd and then unbinding it; the splat above is from
the first unbind. With this patch the same sequence runs cleanly under
KASAN.

Fixes: eb51c9f8cb4f ("HID: corsair: cancel worker before unregistering LED to fix use-after-free")
Reported-by: syzbot+abcedffc9201f2bb66c2@syzkaller.appspotmail.com
Link: https://syzkaller.appspot.com/bug?extid=abcedffc9201f2bb66c2
Assisted-by: LLM
Signed-off-by: Danish Khateeb <danishkhateeb03@gmail.com>
---

Reproducer, for anyone wanting to confirm the fix. Needs CONFIG_HID_CORSAIR,
CONFIG_USB_DUMMY_HCD, CONFIG_USB_CONFIGFS_F_HID and KASAN. It fakes a K90 with
a configfs gadget on dummy_hcd and unbinds it; unpatched, the splat above fires
on the first unbind. syzbot has no reproducer for this one.

  #!/bin/sh
  # Reproducer for the hid-corsair teardown use-after-free.
  #
  # Emulates a Corsair K90 (1b1c:1b02) with a configfs USB gadget bound to
  # dummy_hcd, so hid-corsair probes and registers its two LEDs. Unbinding the
  # gadget drives corsair_remove(), which is the path under test:
  #
  #	removed = true;
  #	cancel_work_sync(&work);        <- work cancelled
  #	led_classdev_unregister(&cdev); <- led_set_brightness(LED_OFF) re-queues it
  #	kfree(k90);                     <- freed while still on the worklist
  #
  # The splat surfaces on the *next* worklist insertion, as a KASAN
  # slab-use-after-free in __list_add_valid_or_report() under __queue_work().
  #
  # Run inside a KASAN guest. Fires on the first unbind.
  set -e

  G=/sys/kernel/config/usb_gadget/k90

  mountpoint -q /sys/kernel/config || mount -t configfs none /sys/kernel/config

  UDC=$(ls /sys/class/udc | head -1)
  [ -n "$UDC" ] || { echo "no UDC found (need CONFIG_USB_DUMMY_HCD)"; exit 1; }
  echo "using UDC: $UDC"

  cleanup() {
  	[ -d "$G" ] || return 0
  	echo "" > "$G/UDC" 2>/dev/null || true
  	rm -f "$G"/configs/c.1/hid.usb0 2>/dev/null || true
  	rmdir "$G"/configs/c.1/strings/0x409 "$G"/configs/c.1 2>/dev/null || true
  	rmdir "$G"/functions/hid.usb0 "$G"/strings/0x409 "$G" 2>/dev/null || true
  }
  cleanup

  mkdir -p "$G"
  cd "$G"
  echo 0x1b1c > idVendor          # USB_VENDOR_ID_CORSAIR
  echo 0x1b02 > idProduct         # USB_DEVICE_ID_CORSAIR_K90

  mkdir -p strings/0x409
  echo "0001"    > strings/0x409/serialnumber
  echo "Corsair" > strings/0x409/manufacturer
  echo "K90"     > strings/0x409/product

  mkdir -p functions/hid.usb0
  echo 1 > functions/hid.usb0/protocol      # keyboard
  echo 1 > functions/hid.usb0/subclass
  echo 8 > functions/hid.usb0/report_length
  # Standard HID boot-keyboard report descriptor, 63 bytes.
  # NOTE: octal escapes, not \xHH -- Debian's /bin/sh is dash, whose printf does
  # not implement \xHH and would write the escapes out as literal ASCII text.
  printf '\005\001\011\006\241\001\005\007\031\340\051\347\025\000\045\001\165\001\225\010\201\002\225\001\165\010\201\003\225\005\165\001\005\010\031\001\051\005\221\002\225\001\165\003\221\003\225\006\165\010\025\000\045\145\005\007\031\000\051\145\201\000\300' \
  	> functions/hid.usb0/report_desc

  desc_sz=$(wc -c < functions/hid.usb0/report_desc)
  [ "$desc_sz" -eq 63 ] || { echo "BAD DESCRIPTOR ($desc_sz bytes, want 63)"; exit 1; }

  mkdir -p configs/c.1/strings/0x409
  echo "c1" > configs/c.1/strings/0x409/configuration
  ln -s functions/hid.usb0 configs/c.1/

  echo "binding gadget..."
  echo "$UDC" > UDC
  sleep 2

  dev=$(ls /sys/bus/hid/drivers/corsair/ 2>/dev/null | grep ':' || true)
  [ -n "$dev" ] || { echo "hid-corsair did not bind - check dmesg"; cleanup; exit 1; }
  echo "bound: $dev"
  echo "LEDs: $(ls /sys/class/leds 2>/dev/null | grep "$dev" | tr '\n' ' ')"

  echo "--- unbinding (triggers corsair_remove) ---"
  echo "" > UDC
  sleep 2

  cleanup
  echo "done (no splat)"

 drivers/hid/hid-corsair.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/hid/hid-corsair.c b/drivers/hid/hid-corsair.c
index 278c6efb565d..ac12877a6db7 100644
--- a/drivers/hid/hid-corsair.c
+++ b/drivers/hid/hid-corsair.c
@@ -524,8 +524,8 @@ static void k90_cleanup_backlight(struct hid_device *dev)
 
 	if (drvdata->backlight) {
 		drvdata->backlight->removed = true;
-		cancel_work_sync(&drvdata->backlight->work);
 		led_classdev_unregister(&drvdata->backlight->cdev);
+		cancel_work_sync(&drvdata->backlight->work);
 		kfree(drvdata->backlight->cdev.name);
 		kfree(drvdata->backlight);
 	}
@@ -540,8 +540,8 @@ static void k90_cleanup_macro_functions(struct hid_device *dev)
 		sysfs_remove_group(&dev->dev.kobj, &k90_attr_group);
 
 		k90->record_led.removed = true;
-		cancel_work_sync(&k90->record_led.work);
 		led_classdev_unregister(&k90->record_led.cdev);
+		cancel_work_sync(&k90->record_led.work);
 		kfree(k90->record_led.cdev.name);
 
 		kfree(k90);
-- 
2.55.0


^ permalink raw reply related	[flat|nested] 4+ messages in thread
end of thread, other threads:[~2026-09-04 16:15 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-09-04 14:36 [PATCH] HID: corsair: cancel worker after unregistering LED, not before Danish Khateeb
2026-09-04 14:53 ` sashiko-bot
2026-09-04 16:13   ` Danish Khateeb
2026-09-04 16:15 ` [PATCH v2] HID: corsair: use disable_work_sync() to tear down the LED workers Danish Khateeb

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox