iio: gyro: mpu3050: use-after-free via trigger left on iio_trigger_list after unbind
Hi Kernel Maintainers,
I hit the following report while testing current upstream kernel:
KASAN: slab-use-after-free in mpu3050_drdy_trigger_set_state
on commit: a90ee4305c4a (2026-10-07)
Root cause: mpu3050_trigger_probe() registers the data-ready trigger with the non-devm iio_trigger_register(), with drvdata pointing at indio_dev:
static int mpu3050_trigger_probe(struct iio_dev *indio_dev, int irq)
{
...
mpu3050->trig->dev.parent = dev;
mpu3050->trig->ops = &mpu3050_trigger_ops;
iio_trigger_set_drvdata(mpu3050->trig, indio_dev);
ret = iio_trigger_register(mpu3050->trig); /* device_add() pins trig; no unregister on remove */
if (ret)
goto err_iio_trigger;
indio_dev->trig = iio_trigger_get(mpu3050->trig);
return 0;
}mpu3050_common_remove() tears down everything else, but never calls iio_trigger_unregister() on mpu3050->trig:
void mpu3050_common_remove(struct device *dev)
{
struct iio_dev *indio_dev = dev_get_drvdata(dev);
struct mpu3050 *mpu3050 = iio_priv(indio_dev);
iio_device_unregister(indio_dev); /* devm frees indio_dev's storage once this returns */
pm_runtime_get_sync(dev);
pm_runtime_put_noidle(dev);
pm_runtime_disable(dev);
if (mpu3050->irq)
free_irq(mpu3050->irq, mpu3050->trig);
iio_triggered_buffer_cleanup(indio_dev);
mpu3050_power_down(mpu3050);
/* trig is never unregistered: still on the global iio_trigger_list */
}So after unbind, the trigger is still alive and discoverable by name on the global trigger list, with drvdata still pointing at the now-freed indio_dev. Any other IIO device with a triggered buffer can select it by name via trigger/current_trigger and enable its buffer, which reaches the leaked trigger's ->set_trigger_state() op:
static int mpu3050_drdy_trigger_set_state(struct iio_trigger *trig,
bool enable)
{
struct iio_dev *indio_dev = iio_trigger_get_drvdata(trig);
struct mpu3050 *mpu3050 = iio_priv(indio_dev); /* indio_dev is freed -- UAF read */
...The reproducer and .config files are here.
https://gist.github.com/dingiso/985aa3eb2786b62f80e2e6780e9af090
I'm happy to test debug patches or provide additional information.
[ 24.039594][ T8117] BUG: KASAN: slab-use-after-free in mpu3050_drdy_trigger_set_state (./include/linux/iio/iio.h:980 drivers/iio/gyro/mpu3050-core.c:957) [ 24.039630][ T8117] Read of size 8 at addr ffff888030621388 by task repro/8117 [ 24.039634][ T8117] [ 24.039644][ T8117] Tainted: [E]=UNSIGNED_MODULE [ 24.039645][ T8117] Hardware name: QEMU Ubuntu 24.04 PC v2 (i440FX + PIIX, arch_caps fix, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014 [ 24.039648][ T8117] Call Trace: [ 24.039650][ T8117] <TASK> [ 24.039651][ T8117] dump_stack_lvl (lib/dump_stack.c:94 lib/dump_stack.c:120) [ 24.039663][ T8117] print_report (mm/kasan/report.c:378 mm/kasan/report.c:482) [ 24.039682][ T8117] kasan_report (mm/kasan/report.c:595) [ 24.039690][ T8117] mpu3050_drdy_trigger_set_state (./include/linux/iio/iio.h:980 drivers/iio/gyro/mpu3050-core.c:957) [ 24.039709][ T8117] iio_trigger_attach_poll_func (drivers/iio/industrialio-trigger.c:308) [ 24.039715][ T8117] __iio_update_buffers (drivers/iio/industrialio-buffer.c:1207 drivers/iio/industrialio-buffer.c:1324) [ 24.039747][ T8117] enable_store (drivers/iio/industrialio-buffer.c:1414) [ 24.039765][ T8117] kernfs_fop_write_iter (fs/kernfs/file.c:345) [ 24.039769][ T8117] vfs_write (fs/read_write.c:595 fs/read_write.c:687) [ 24.039786][ T8117] ksys_write (fs/read_write.c:739) [ 24.039795][ T8117] do_syscall_64 (arch/x86/entry/syscall_64.c:61 arch/x86/entry/syscall_64.c:84) [ 24.039799][ T8117] entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:121) [ 24.039803][ T8117] RIP: 0033:0x7fe293245473 [ 24.039807][ T8117] Code: 8b 15 21 2a 0e 00 f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb b7 0f 1f 00 64 8b 04 25 18 00 00 00 85 c0 75 14 b8 01 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 55 c3 0f 1f 40 00 48 83 ec 28 48 89 54 24 18 All code ======== 0: 8b 15 21 2a 0e 00 mov 0xe2a21(%rip),%edx # 0xe2a27 6: f7 d8 neg %eax 8: 64 89 02 mov %eax,%fs:(%rdx) b: 48 c7 c0 ff ff ff ff mov $0xffffffffffffffff,%rax 12: eb b7 jmp 0xffffffffffffffcb 14: 0f 1f 00 nopl (%rax) 17: 64 8b 04 25 18 00 00 mov %fs:0x18,%eax 1e: 00 1f: 85 c0 test %eax,%eax 21: 75 14 jne 0x37 23: b8 01 00 00 00 mov $0x1,%eax 28: 0f 05 syscall 2a:* 48 3d 00 f0 ff ff cmp $0xfffffffffffff000,%rax <-- trapping instruction 30: 77 55 ja 0x87 32: c3 ret 33: 0f 1f 40 00 nopl 0x0(%rax) 37: 48 83 ec 28 sub $0x28,%rsp 3b: 48 89 54 24 18 mov %rdx,0x18(%rsp)
Code starting with the faulting instruction
===========================================
0: 48 3d 00 f0 ff ff cmp $0xfffffffffffff000,%rax 6: 77 55 ja 0x5d 8: c3 ret 9: 0f 1f 40 00 nopl 0x0(%rax) d: 48 83 ec 28 sub $0x28,%rsp 11: 48 89 54 24 18 mov %rdx,0x18(%rsp) [ 24.039810][ T8117] RSP: 002b:00007fff7856e628 EFLAGS: 00000246 ORIG_RAX: 0000000000000001 [ 24.039815][ T8117] RAX: ffffffffffffffda RBX: 00007fff7856e6e0 RCX: 00007fe293245473 [ 24.039818][ T8117] RDX: 0000000000000001 RSI: 000056386e819159 RDI: 0000000000000003 [ 24.039820][ T8117] RBP: 0000000000000003 R08: 0000000000004000 R09: 00007fe2932e90c0 [ 24.039822][ T8117] R10: 0000000000000000 R11: 0000000000000246 R12: 000056386e819159 [ 24.039824][ T8117] R13: 00007fff7856ea60 R14: 00007fff7856ea60 R15: 0000563885c1635b [ 24.039827][ T8117] </TASK> [ 24.039829][ T8117] [ 24.039830][ T8117] Freed by task 8117: [ 24.039832][ T8117] kasan_save_stack (mm/kasan/common.c:57) [ 24.039835][ T8117] kasan_save_track (mm/kasan/common.c:78) [ 24.039838][ T8117] kasan_save_free_info (mm/kasan/generic.c:584) [ 24.039841][ T8117] __kasan_slab_free (mm/kasan/common.c:264 mm/kasan/common.c:296) [ 24.039844][ T8117] kfree (./include/linux/kasan.h:235 mm/slub.c:2748 mm/slub.c:6509 mm/slub.c:6802) [ 24.039849][ T8117] device_release (drivers/base/core.c:2645) [ 24.039854][ T8117] kobject_put (lib/kobject.c:689 lib/kobject.c:720 ./include/linux/kref.h:65 lib/kobject.c:737) [ 24.039859][ T8117] release_nodes (drivers/base/devres.c:546) [ 24.039862][ T8117] devres_release_group (drivers/base/devres.c:739) [ 24.039865][ T8117] i2c_device_remove (drivers/i2c/i2c-core-base.c:637) [ 24.039871][ T8117] device_release_driver_internal (drivers/base/dd.c:1349 drivers/base/dd.c:1372) [ 24.039874][ T8117] bus_remove_device (drivers/base/bus.c:664) [ 24.039878][ T8117] device_del (drivers/base/core.c:3965) [ 24.039882][ T8117] device_unregister (drivers/base/core.c:4006) [ 24.039885][ T8117] delete_device_store (drivers/i2c/i2c-core-base.c:1369) [ 24.039889][ T8117] kernfs_fop_write_iter (fs/kernfs/file.c:345) [ 24.039892][ T8117] vfs_write (fs/read_write.c:595 fs/read_write.c:687) [ 24.039895][ T8117] ksys_write (fs/read_write.c:739) [ 24.039899][ T8117] do_syscall_64 (arch/x86/entry/syscall_64.c:61 arch/x86/entry/syscall_64.c:84) [ 24.039901][ T8117] entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:121)
(Full KASAN report, including the trailing slab/page-owner and memory-state info after this point, is in the gist as kasan_report_full.txt. Lines starting with "?" above are stack-scanned frames, not reliable return addresses, and have been omitted here.)
Best,
Dingisoul
Quoted 59 lines · On Sat, 10 Oct 2026 16:36:01 -0400 Dingisoul <[email protected]> wrote:
Hi Kernel Maintainers,
I hit the following report while testing current upstream kernel:
KASAN: slab-use-after-free in mpu3050_drdy_trigger_set_stateon commit:
a90ee4305c4a(2026-10-07)Root cause:
mpu3050_trigger_probe()registers the data-ready trigger with the non-devmiio_trigger_register(), with drvdata pointing atindio_dev:static int mpu3050_trigger_probe(struct iio_dev *indio_dev, int irq) { ... mpu3050->trig->dev.parent = dev; mpu3050->trig->ops = &mpu3050_trigger_ops; iio_trigger_set_drvdata(mpu3050->trig, indio_dev); ret = iio_trigger_register(mpu3050->trig); /* device_add() pins trig; no unregister on remove */ if (ret) goto err_iio_trigger; indio_dev->trig = iio_trigger_get(mpu3050->trig); return 0; }
mpu3050_common_remove()tears down everything else, but never callsiio_trigger_unregister()onmpu3050->trig:void mpu3050_common_remove(struct device *dev) { struct iio_dev *indio_dev = dev_get_drvdata(dev); struct mpu3050 *mpu3050 = iio_priv(indio_dev); iio_device_unregister(indio_dev); /* devm frees indio_dev's storage once this returns */ pm_runtime_get_sync(dev); pm_runtime_put_noidle(dev); pm_runtime_disable(dev); if (mpu3050->irq) free_irq(mpu3050->irq, mpu3050->trig); iio_triggered_buffer_cleanup(indio_dev); mpu3050_power_down(mpu3050); /* trig is never unregistered: still on the global iio_trigger_list */ }So after unbind, the trigger is still alive and discoverable by name on the global trigger list, with drvdata still pointing at the now-freed
indio_dev. Any other IIO device with a triggered buffer can select it by name via trigger/current_trigger and enable its buffer, which reaches the leaked trigger's ->set_trigger_state() op:static int mpu3050_drdy_trigger_set_state(struct iio_trigger *trig, bool enable) { struct iio_dev *indio_dev = iio_trigger_get_drvdata(trig); struct mpu3050 *mpu3050 = iio_priv(indio_dev); /* indio_dev is freed -- UAF read */ ...
Quoted 5 lines
The reproducer and .config files are here.
https://gist.github.com/dingiso/985aa3eb2786b62f80e2e6780e9af090I'm happy to test debug patches or provide additional information.
Hi, thanks for the bug report!
I've sent a patch addressing the issue, it would be great if you could test it.
https://lore.kernel.org/linux-iio/[email protected]/
Thanks.
Signature · 2 lines
Kind regards, Joshua Crofts
Hi Joshua, Sashiko,
Sashiko's concern is really about not being able to tell whether the trigger was ever actually registered before calling iio_trigger_unregister(). There's a precedent for this exact bug class:
commit 3c5eed894efd9 ("iio: temperature: tmp006: use devm_iio_trigger_register") hit the same devm_iio_trigger_alloc()-but-plain-register() mismatch in tmp006, and fixed it by switching to devm_iio_trigger_register() instead of adding a manual unregister call.
https://github.com/torvalds/linux/commit/3c5eed894efd9
devm_iio_trigger_register() only installs its cleanup action if iio_trigger_register() actually succeeds:
int devm_iio_trigger_register(struct device *dev,
struct iio_trigger *trig_info)
{
int ret;
ret = iio_trigger_register(trig_info);
if (ret)
return ret;
return devm_add_action_or_reset(dev, devm_iio_trigger_unreg, trig_info);
}so it can never unregister a trigger that was never registered -- no flag/pointer tracking needed.
Suggested fix, in mpu3050_trigger_probe():
- ret = iio_trigger_register(mpu3050->trig); + ret = devm_iio_trigger_register(&indio_dev->dev, mpu3050->trig);
(same scope devm_iio_trigger_alloc() already uses a few lines above). With this, both iio_trigger_unregister() calls in your patch can be dropped.
I rebuilt with just this one-line change and reran my reproducer:
the trigger registers normally on probe, and after delete_device it's actually gone instead of left dangling, so the attack just gets -EINVAL instead of a UAF. No KASAN report, no warnings.
Happy to test further revisions.
Best,
Dingisoul
Quoted 28 lines · On Mon, 12 Oct 2026 16:09:04 -0400 Dingisoul <[email protected]> wrote:
Hi Joshua, Sashiko,
Sashiko's concern is really about not being able to tell whether the trigger was ever actually registered before calling
iio_trigger_unregister(). There's a precedent for this exact bug class:
commit3c5eed894efd9("iio: temperature: tmp006: usedevm_iio_trigger_register") hit the same devm_iio_trigger_alloc()-but-plain-register() mismatch in tmp006, and fixed it by switching todevm_iio_trigger_register()instead of adding a manual unregister call.https://github.com/torvalds/linux/commit/3c5eed894efd9
devm_iio_trigger_register()only installs its cleanup action ifiio_trigger_register()actually succeeds:int devm_iio_trigger_register(struct device *dev, struct iio_trigger *trig_info) { int ret; ret = iio_trigger_register(trig_info); if (ret) return ret; return devm_add_action_or_reset(dev, devm_iio_trigger_unreg, trig_info); }
Quoted 7 lines
so it can never unregister a trigger that was never registered -- no flag/pointer tracking needed.
Suggested fix, in
mpu3050_trigger_probe():- ret = iio_trigger_register(mpu3050->trig); + ret = devm_iio_trigger_register(&indio_dev->dev, mpu3050->trig);
Ideally you shouldn't mix devm_* and non-devm_* function, this can cause headaches down the road. Either we do the manual dance or we convert the entire driver to devm_*. If you check the file in the commit you referenced above, you'll notice that all the other functions are devm_*.
Bartosz Golaszewski had a talk on this at OSS Prague (recording still isn't available, so I'm attaching the presentation instead).
https://hosted-files.sched.co/osselceu2026/0c/OSS%202026%20Prague.pdf
Signature · 2 lines
Kind regards, Joshua Crofts