Skip to contentSkip to conversation

USB: iowarrior: kill async writes before freeing on disconnect

PATCH In review Cen Zhang 2 participants 3 messages Jun 18 USB

lore
Share & tools
lore

Cen Zhang

author
#0 lore raw

PATCH 1/1 USB: iowarrior: kill async writes before freeing on disconnect

IOWarrior write() uses anchored interrupt URBs for async-capable devices and passes struct iowarrior as the completion context. release() can close the file while those URBs are still outstanding; it only stops the interrupt-in URB when the device is still present.

disconnect() already kills the submitted write URBs when the device is still opened, but skips that step when opened is clear and then frees struct iowarrior. That allows a close-before-disconnect sequence to free the embedded anchor and completion context while giveback is still using them.

The buggy scenario involves two paths, with each column showing the order within that path:

file write and close path: disconnect and giveback path:
1. write() submits an anchored 1. USB disconnect sets present to 0.
interrupt-out URB.
2. release() sets opened to 0 2. Because opened is 0, disconnect
and returns without draining skips the write-anchor drain and
dev->submitted. frees dev.
3. The write URB remains in 3. Giveback unanchors the URB and
flight after close. runs iowarrior_write_callback()
with the freed dev.

Kill the submitted write URBs in the opened == 0 disconnect branch before dropping the device mutex and freeing the device. usb_kill_anchored_urbs() waits for any completion window covered by the anchor, so the embedded anchor and callback context remain live until write giveback is done.

Validation reproduced this kernel report:
KASAN slab-use-after-free in iowarrior_write_callback+0x7d/0xf0

RIP: 0010:pv_native_safe_halt+0xf/0x20
Write of size 4
Call trace:
  dump_stack_lvl+0x66/0xa0
  print_report+0xce/0x630
  fixup_red_left+0x9/0x30
  complete_report_info+0x83/0x110
  iowarrior_write_callback+0x7d/0xf0 (drivers/usb/misc/iowarrior.c:220)
  kasan_report+0xe0/0x110
  kasan_check_range+0x105/0x1b0
  __usb_hcd_giveback_urb+0x112/0x1d0
  dummy_timer+0xaaa/0x19a0
  dummy_timer+0x4/0x19a0
  srso_alias_return_thunk+0x5/0xfbef5
  __hrtimer_run_queues+0xeb/0x510
  __hrtimer_run_queues+0x102/0x510
  hrtimer_run_softirq+0xd0/0x130
  handle_softirqs+0x155/0x650
  do_raw_spin_unlock+0x9a/0x100
  do_raw_spin_unlock+0x8b/0x100
  __irq_exit_rcu+0xc4/0x160
  irq_exit_rcu+0xe/0x20
  sysvec_apic_timer_interrupt+0x6c/0x80
  asm_sysvec_apic_timer_interrupt+0x1a/0x20
Allocated by task stack:
  kasan_save_stack+0x33/0x60
  kasan_save_track+0x14/0x30
  __kasan_kmalloc+0x8f/0xa0
  __device_attach_driver+0xf1/0x1a0
  bus_for_each_drv+0xf9/0x160
  __device_attach+0x133/0x2a0
  device_add+0x9b9/0xc10
  usb_set_configuration+0xb64/0xf20
  usb_new_device+0x492/0x870
  hub_event+0x1b10/0x29c0
  process_one_work+0x4d7/0xb90
  worker_thread+0x2d8/0x570
  kthread+0x1ad/0x1f0
  ret_from_fork+0x3c9/0x540
  ret_from_fork_asm+0x1a/0x30
Fixes 946b960d13c1 ("USB: add driver for iowarrior devices.")
Assisted-by Codex:gpt-5.5
---
 drivers/usb/misc/iowarrior.c | 1 +
 1 file changed, 1 insertion(+)
drivers/usb/misc/iowarrior.c +1 -0
index 22504c0a2841..07406ec7aabe 100644
@@ -917,6 +917,7 @@ static void iowarrior_disconnect(struct usb_interface *interface)
 		mutex_unlock(&dev->mutex);
 	} else {
 		/* no process is using the device, cleanup now */
+		usb_kill_anchored_urbs(&dev->submitted);
 		mutex_unlock(&dev->mutex);
 		iowarrior_delete(dev);
 	}
Signature · 1 line
2.43.0
#1 lore raw
Quoted 4 lines · Cen Zhang

IOWarrior write() uses anchored interrupt URBs for async-capable devices and passes struct iowarrior as the completion context. release() can close the file while those URBs are still outstanding; it only stops the interrupt-in URB when the device is still present.

I believe I fixed this here:

	https://lore.kernel.org/lkml/[email protected]/

but there may have been some confusion about who was going to apply it so the fix is not yet in Greg's tree.

Johan

author
#2 lore raw

Hi

Quoted 7 lines

I believe I fixed this here:

        https://lore.kernel.org/lkml/[email protected]/

but there may have been some confusion about who was going to apply it so the fix is not yet in Greg's tree.

Thanks, I missed your earlier patch. I will test and verify it on my side if needed.

Best regards,
Cen Zhang