Skip to contentSkip to conversation

media: em28xx: NULL pointer dereference in em28xx_close_extension() on disconnect/unload

REGRESSION Paweł Ziemniak 2 participants 3 messages Nov 18 – Nov 19

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

Hi,

Commit 826915b6b65e ("media: em28xx: fix use-after-free of dev_next->devlist on disconnect", CVE-2026-89891) introduces a 100% reproducible kernel NULL pointer dereference in em28xx_close_extension() on real dual-TS hardware (Hauppauge WinTV-dualHD, USB ID 2040:8265) whenever the device disconnects or the em28xx module is unloaded.

#regzbot introduced: 826915b6b65e

Impact
------
When triggered by a USB disconnect, the Oops kills the kworker thread on usb_hub_wq (hub_event -> usb_disconnect -> usb_disable_device -> usb_unbind_interface -> em28xx_usb_disconnect -> em28xx_close_extension) while the USB device/hub lock is held. This permanently deadlocks USB enumeration/lsusb and container/device teardown until a full host reboot.

When triggered by modprobe -r em28xx_dvb em28xx_rc em28xx, modprobe is killed (SIGKILL) leaving em28xx stuck in initstate "going" (refcnt -1).

Hardware & Kernels Affected
---------------------------
- Hardware: Dell Inc. OptiPlex 3070 (Intel Cannon Lake PCH xHCI 0000:00:14.0)
- USB Device: Hauppauge WinTV-dualHD (2040:8265, has_dual_ts = 1, is_audio_only = 0)
- Observed identically on Proxmox VE kernels (in-tree em28xx module; taint flags are only from ZFS/r8125 OOT modules and are unrelated):
  - 7.0.14-23-pve (2026-10-27, usb_hub_wq disconnect)
  - 7.0.14-26-pve (2026-10-28, usb_hub_wq disconnect)
  - 7.0.14-28-pve (2026-11-07, usb_hub_wq disconnect)
  - 7.0.14-29-pve (2026-11-18, modprobe -r reproducer)

Reproducer
----------
With a Hauppauge WinTV-dualHD (2040:8265) attached (and no userspace holding /dev/dvb/*):

  # modprobe -r em28xx_dvb em28xx_rc em28xx
(or physically unplugging / resetting the USB stick).

Root Cause Analysis
-------------------
For normal dual-TS devices (has_dual_ts = 1, is_audio_only = 0), only the primary dev is inserted into em28xx_devlist; dev->dev_next->devlist is never added to em28xx_devlist (and remains zero-initialized with next == NULL, prev == NULL).

Disassembly of the crashing Code bytes at em28xx_close_extension+0x80/0x140:

  mov    0x17a0(%r12),%rbx   ; rbx = dev->dev_next (ffff8ba4513fe000) test   %rbx,%rbx           ; if (!dev->dev_next)
  je     +0x4b
  mov    0x1f0(%rbx),%rax    ; rax = dev->dev_next->devlist.prev == 0x0 mov    0x1e8(%rbx),%rdx    ; rdx = dev->dev_next->devlist.next == 0x0 lea    0x1e8(%rbx),%rdi    ; rdi = &dev->dev_next->devlist (ffff8ba4513fe1e8) cmp    (%rax),%rdi         ; <-- #PF CR2=0x0 in list_del(&dev->dev_next->devlist)

Because dev->dev_next->devlist.next and .prev are NULL (0x0), list_del(&dev->dev_next->devlist) unconditionally dereferences NULL inside __list_del_entry_valid().

If em28xx_devlist is only supposed to hold primary devices (as addressed by commit 95f76f51937f "media: em28xx: defer audio-only extension registration"), list_del(&dev->dev_next->devlist) should either be removed or guarded (e.g., initialized with INIT_LIST_HEAD(&dev_next->devlist) and checked with !list_empty(&dev->dev_next->devlist) or dev->dev_next->devlist.next).

Trace 1: Spontaneous USB Disconnect (deadlocks usb_hub_wq)
----------------------------------------------------------
usb 1-2: USB disconnect, device number 2
em28xx 1-2:1.0: Disconnecting em28xx #1
em28xx 1-2:1.0: Disconnecting em28xx
em28xx 1-2:1.0: Closing DVB extension
em28xx 1-2:1.0: Closing DVB extension
em28xx 1-2:1.0: Closing input extension
em28xx 1-2:1.0: Closing input extension

BUG: kernel NULL pointer dereference, address: 0000000000000000
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 0 P4D 0
Oops: Oops: 0000 [#1] SMP PTI
CPU: 4 UID: 0 PID: 3086307 Comm: kworker/4:1 Tainted: P S  OE       7.0.14-28-pve #2 PREEMPT(lazy)
Tainted: [P]=PROPRIETARY_MODULE, [S]=CPU_OUT_OF_SPEC, [O]=OOT_MODULE, [E]=UNSIGNED_MODULE
Hardware name: Dell Inc. OptiPlex 3070/02N3WF, BIOS 1.33.0 03/26/2025
Workqueue: usb_hub_wq hub_event
RIP: 0010:em28xx_close_extension+0x80/0x140 [em28xx]
Code: 81 fb 50 82 3c c1 75 cd 49 8b 9c 24 a0 17 00 00 48 85 db 74 4b 48 8b 83 f0 01 00 00 48 8b 93 e8 01 00 00 48 8d bb e8 01 00 00 <48> 3b 38 0f 85 fa 26 00 00 48 3b 7a 08 0f 85 f0 26 00 00 48 89 42
RSP: 0000:ffffcf6822097af8 EFLAGS: 00010286
RAX: 0000000000000000 RBX: ffff8ba4513fe000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff8ba4513fe1e8
RBP: ffffcf6822097b08 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000000 R12: ffff8ba45576c000
R13: ffff8ba4461dd8e0 R14: ffffffffc13c8590 R15: ffff8ba4461dd850
FS:  0000000000000000(0000) GS:ffff8babd7908000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 00000005f5b86004 CR4: 00000000003726f0
Call Trace:
 <TASK>
 em28xx_usb_disconnect.cold+0x71/0xbf [em28xx]
 usb_unbind_interface+0x9b/0x2e0
 device_remove+0x68/0x80
 device_release_driver_internal+0x206/0x270
 device_release_driver+0x12/0x20
 bus_remove_device+0xfd/0x1c0
 ? device_remove_attrs+0xb6/0x100
 device_del+0x160/0x3c0
 usb_disable_device+0xea/0x240
 usb_disconnect+0xe6/0x2e0
 hub_event+0xe8c/0x1a40
 process_one_work+0x1a9/0x3c0
 worker_thread+0x1b8/0x360
 ? __pfx_worker_thread+0x10/0x10
 kthread+0xf7/0x130
 ? __pfx_kthread+0x10/0x10
 ret_from_fork+0x2da/0x3a0
 ? __pfx_kthread+0x10/0x10
 ret_from_fork_asm+0x1a/0x30
 </TASK>

note: kworker/4:1[3086307] exited with irqs disabled

Trace 2: Deterministic Reproducer (modprobe -r em28xx_dvb em28xx_rc em28xx)
-----------------------------------------------------------------------------
em28xx 1-2:1.0: Closing DVB extension
em28xx 1-2:1.0: Closing DVB extension
em28xx: Removed (Em28xx dvb Extension) extension
em28xx 1-2:1.0: Closing input extension
em28xx 1-2:1.0: Closing input extension
em28xx: Removed (Em28xx Input Extension) extension
usbcore: deregistering interface driver em28xx
em28xx 1-2:1.0: Disconnecting em28xx #1
em28xx 1-2:1.0: Disconnecting em28xx

BUG: kernel NULL pointer dereference, address: 0000000000000000
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 0 P4D 0
Oops: Oops: 0000 [#1] SMP PTI
CPU: 3 UID: 0 PID: 151085 Comm: modprobe Tainted: P S         OE      7.0.14-29-pve #1 PREEMPT(lazy)
Tainted: [P]=PROPRIETARY_MODULE, [S]=CPU_OUT_OF_SPEC, [O]=OOT_MODULE, [E]=UNSIGNED_MODULE
Hardware name: Dell Inc. OptiPlex 3070/02N3WF, BIOS 1.33.0 03/26/2025
RIP: 0010:em28xx_close_extension+0x80/0x140 [em28xx]
Code: 81 fb 50 f2 14 c1 75 cd 49 8b 9c 24 a0 17 00 00 48 85 db 74 4b 48 8b 83 f0 01 00 00 48 8b 93 e8 01 00 00 48 8d bb e8 01 00 00 <48> 3b 38 0f 85 fa 26 00 00 48 3b 7a 08 0f 85 f0 26 00 00 48 89 42
RSP: 0018:ffffcadcb18d3b80 EFLAGS: 00010286
RAX: 0000000000000000 RBX: ffff8a28c5d92000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff8a28c5d921e8
RBP: ffffcadcb18d3b90 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000000 R12: ffff8a28c5d96000
R13: ffff8a28df3b60e0 R14: ffffffffc114f590 R15: ffff8a28df3b6050
FS:  0000747342ef1640(0000) GS:ffff8a304b888000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 000000049f936006 CR4: 00000000003726f0
Call Trace:
 <TASK>
 em28xx_usb_disconnect.cold+0x71/0xbf [em28xx]
 usb_unbind_interface+0x9b/0x2e0
 device_remove+0x68/0x80
 device_release_driver_internal+0x206/0x270
 driver_detach+0x4a/0xa0
 bus_remove_driver+0x83/0x120
 driver_unregister+0x2f/0x60
 usb_deregister+0x81/0x109
 em28xx_usb_driver_exit+0x10/0xf90 [em28xx]
 __do_sys_delete_module.isra.0+0x1c3/0x2f0
 __x64_sys_delete_module+0x12/0x20
 x64_sys_call+0x20c3/0x2390
 do_syscall_64+0x10b/0x14c0
 entry_SYSCALL_64_after_hwframe+0x76/0x7e
 </TASK>

note: modprobe[151085] exited with irqs disabled

Happy to test a patch on this hardware.

Regards,

  Paweł Ziemniak

#1 lore raw
Quoted 8 lines · Paweł Ziemniak

Commit 826915b6b65e ("media: em28xx: fix use-after-free of dev_next-

Quoted 1 line

devlist on disconnect",

CVE-2026-89891) introduces a 100% reproducible kernel NULL pointer dereference in
em28xx_close_extension() on real dual-TS hardware (Hauppauge WinTV- dualHD, USB ID 2040:8265)
whenever the device disconnects or the em28xx module is unloaded.

This is not my area of expertise, so the following might be sending you in the wrong direction. But FWIW:

From a very quick look this seems to be the problem that was reported earlier; a fix is discussed here:

[PATCH] media: em28xx: fix NULL pointer dereference on disconnect of dual-TS devices
https://lore.kernel.org/all/[email protected]/

If your's is a different problem, please let us know.

Ciao, Thorsten

#2 lore raw

Thank you, that seems to be the same problem, I'm sorry I missed that.

I backported and tested commit b0bb9a4fe0d3 from linux-next ("media: em28xx: fix NULL pointer dereference on disconnect of dual-TS devices") against 7.0.14-29-pve on my Hauppauge WinTV-dualHD (2040:8265).

Unloading/reloading (modprobe -r em28xx_dvb em28xx_rc em28xx + modprobe em28xx) and USB disconnect now complete cleanly with no Oops and no USB hang.

#regzbot fix: media: em28xx: fix NULL pointer dereference on disconnect of dual-TS devices

Regards,

  Paweł Ziemniak

W dniu 2026-11-19 o 06:44, Thorsten Leemhuis pisze:

Quoted 21 lines · Paweł Ziemniak
Quoted 7 lines

Commit 826915b6b65e ("media: em28xx: fix use-after-free of dev_next-

Quoted 1 line

devlist on disconnect",

CVE-2026-89891) introduces a 100% reproducible kernel NULL pointer dereference in
em28xx_close_extension() on real dual-TS hardware (Hauppauge WinTV- dualHD, USB ID 2040:8265)
whenever the device disconnects or the em28xx module is unloaded.

This is not my area of expertise, so the following might be sending you in the wrong direction. But FWIW:

From a very quick look this seems to be the problem that was reported
earlier; a fix is discussed here:

[PATCH] media: em28xx: fix NULL pointer dereference on disconnect of dual-TS devices
https://lore.kernel.org/all/[email protected]/

If your's is a different problem, please let us know.

Ciao, Thorsten