[PATCH] iio: core: Fix BUG() on rmmod of iio-drivers caused by double list_del()

* [PATCH] iio: core: Fix BUG() on rmmod of iio-drivers caused by double list_del()
@ 2021-05-14 19:59 Hans de Goede
  2021-05-15  6:57 ` Alexandru Ardelean
  0 siblings, 1 reply; 3+ messages in thread
From: Hans de Goede @ 2021-05-14 19:59 UTC (permalink / raw)
  To: Jonathan Cameron
  Cc: Hans de Goede, Lars-Peter Clausen, linux-iio, Alexandru Ardelean

Removing an iio-driver module (or unbinding the driver) causes a BUG() to
trigger when the kernel is build with list-debugging enabled. This is
caused by a double list_del() introduced by commit 8dedcc3eee3a ("iio:
core: centralize ioctl() calls to the main chardev").

This patch introduces an ioctl_handlers list to which ioctl handlers are
added. The cleanup of these however happens twice.

The 2 registration functions:
iio_device_ioctl_handler_register()
iio_device_register_eventset()

Have matching unregister functions which do a list_del on the head added
to the list; and iio_device_unregister() has:

    list_for_each_entry_safe(h, t, &iio_dev_opaque->ioctl_handlers, entry)
        list_del(&h->entry);

This is a problem because iio_device_unregister() does this before calling
iio_buffers_free_sysfs_and_mask() which calls
iio_device_ioctl_handler_unregister() which calls list_del on the entry
added by iio_buffers_alloc_sysfs_and_mask() a second time causing this:

[root@fedora ~]# rmmod bmc150_accel_i2c
Segmentation fault

[  160.627546] ------------[ cut here ]------------
[  160.627549] kernel BUG at lib/list_debug.c:45!
[  160.629125] invalid opcode: 0000 [#1] SMP PTI
[  160.629140] CPU: 0 PID: 1139 Comm: rmmod Tainted: G            E     5.13.0-rc1+ #341
[  160.629146] Hardware name: LENOVO 80M1/Mini, BIOS C7CN31WW 08/05/2016
[  160.629149] RIP: 0010:__list_del_entry_valid.cold+0xf/0x47
...
[  160.629202] Call Trace:
[  160.629209]  iio_device_ioctl_handler_unregister+0xe/0x90 [industrialio]
[  160.629226]  iio_device_unregister_trigger_consumer+0x21d/0x2f0 [industrialio]
[  160.629239]  bmc150_accel_core_remove+0x20/0xd0 [bmc150_accel_core]
[  160.629246]  i2c_device_remove+0x25/0xb0
[  160.629254]  __device_release_driver+0x180/0x240
[  160.629261]  device_release_driver+0x26/0x40
[  160.629267]  bus_remove_device+0xf2/0x160
[  160.629272]  device_del+0x18c/0x3e0
[  160.629280]  device_unregister+0x16/0x60
...

Since iio_device_ioctl_handler_unregister/iio_device_unregister_eventset()
already do the list_del() there is no need for the
list_for_each_entry_safe() in iio_device_unregister() which also does the
list_del(). Remove the list_for_each_entry_safe() to fix the double
list_del() issue.

Fixes: 8dedcc3eee3a ("iio: core: centralize ioctl() calls to the main chardev")
Cc: Alexandru Ardelean <alexandru.ardelean@analog.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
---
 drivers/iio/industrialio-core.c | 6 ------
 1 file changed, 6 deletions(-)

diff --git a/drivers/iio/industrialio-core.c b/drivers/iio/industrialio-core.c
index d92c58a94fe4..9e59f5da3d28 100644
--- a/drivers/iio/industrialio-core.c
+++ b/drivers/iio/industrialio-core.c
@@ -1926,9 +1926,6 @@ EXPORT_SYMBOL(__iio_device_register);
  **/
 void iio_device_unregister(struct iio_dev *indio_dev)
 {
-	struct iio_dev_opaque *iio_dev_opaque = to_iio_dev_opaque(indio_dev);
-	struct iio_ioctl_handler *h, *t;
-
 	cdev_device_del(&indio_dev->chrdev, &indio_dev->dev);
 
 	mutex_lock(&indio_dev->info_exist_lock);
@@ -1939,9 +1936,6 @@ void iio_device_unregister(struct iio_dev *indio_dev)
 
 	indio_dev->info = NULL;
 
-	list_for_each_entry_safe(h, t, &iio_dev_opaque->ioctl_handlers, entry)
-		list_del(&h->entry);
-
 	iio_device_wakeup_eventset(indio_dev);
 	iio_buffer_wakeup_poll(indio_dev);
 
-- 
2.31.1


^ permalink raw reply related	[flat|nested] 3+ messages in thread