These changes addresses kernel warnings which shows up after enabling
debug kernel. First one fixes use-after-free warning and second fixes
warning by updating cdev APIs
Changes in v3:
- Clear ept pointers in patch 1
- Remove error check in patch 2
Changes in v2:
- Fix typo in commit message
Sujit Kautkar (2):
rpmsg: glink: Fix use-after-free in qcom_glink_rpdev_release()
rpmsg: glink: Update cdev add/del API in
rpmsg_ctrldev_release_device()
drivers/rpmsg/qcom_glink_native.c | 12 ++++++++++--
drivers/rpmsg/rpmsg_char.c | 10 ++--------
2 files changed, 12 insertions(+), 10 deletions(-)
--
2.31.0
Replace cdev add/del APIs with cdev_device_add/cdev_device_del to avoid
below kernel warning. This correctly takes a reference to the parent
device so the parent will not get released until all references to the
cdev are released.
| ODEBUG: free active (active state 0) object type: timer_list hint: delayed_work_timer_fn+0x0/0x7c
| WARNING: CPU: 7 PID: 19892 at lib/debugobjects.c:488 debug_print_object+0x13c/0x1b0
| CPU: 7 PID: 19892 Comm: kworker/7:4 Tainted: G W 5.4.147-lockdep #1
| ==================================================================
| Hardware name: Google CoachZ (rev1 - 2) with LTE (DT)
| Workqueue: events kobject_delayed_cleanup
| pstate: 60c00009 (nZCv daif +PAN +UAO)
| pc : debug_print_object+0x13c/0x1b0
| lr : debug_print_object+0x13c/0x1b0
| sp : ffffff83b2ec7970
| x29: ffffff83b2ec7970 x28: dfffffd000000000
| x27: ffffff83d674f000 x26: dfffffd000000000
| x25: ffffffd06b8fa660 x24: dfffffd000000000
| x23: 0000000000000000 x22: ffffffd06b7c5108
| x21: ffffffd06d597860 x20: ffffffd06e2c21c0
| x19: ffffffd06d5974c0 x18: 000000000001dad8
| x17: 0000000000000000 x16: dfffffd000000000
| BUG: KASAN: use-after-free in qcom_glink_rpdev_release+0x54/0x70
| x15: ffffffffffffffff x14: 79616c6564203a74
| x13: 0000000000000000 x12: 0000000000000080
| Write of size 8 at addr ffffff83d95768d0 by task kworker/3:1/150
| x11: 0000000000000001 x10: 0000000000000000
| x9 : fc9e8edec0ad0300 x8 : fc9e8edec0ad0300
|
| x7 : 0000000000000000 x6 : 0000000000000000
| x5 : 0000000000000080 x4 : 0000000000000000
| CPU: 3 PID: 150 Comm: kworker/3:1 Tainted: G W 5.4.147-lockdep #1
| x3 : ffffffd06c149574 x2 : ffffff83f77f7498
| x1 : ffffffd06d596f60 x0 : 0000000000000061
| Hardware name: Google CoachZ (rev1 - 2) with LTE (DT)
| Call trace:
| debug_print_object+0x13c/0x1b0
| Workqueue: events kobject_delayed_cleanup
| __debug_check_no_obj_freed+0x25c/0x3c0
| debug_check_no_obj_freed+0x18/0x20
| Call trace:
| slab_free_freelist_hook+0xb4/0x1bc
| kfree+0xe8/0x2d8
| dump_backtrace+0x0/0x27c
| rpmsg_ctrldev_release_device+0x78/0xb8
| device_release+0x68/0x14c
| show_stack+0x20/0x2c
| kobject_cleanup+0x12c/0x298
| kobject_delayed_cleanup+0x10/0x18
| dump_stack+0xe0/0x19c
| process_one_work+0x578/0x92c
| worker_thread+0x804/0xcf8
| print_address_description+0x3c/0x4a8
| kthread+0x2a8/0x314
| ret_from_fork+0x10/0x18
| __kasan_report+0x100/0x124
Signed-off-by: Sujit Kautkar <[email protected]>
---
Changes in v3:
- Remove unecessary error check as per Matthias's comment
Changes in v2:
- Fix typo in commit message
drivers/rpmsg/rpmsg_char.c | 10 ++--------
1 file changed, 2 insertions(+), 8 deletions(-)
diff --git a/drivers/rpmsg/rpmsg_char.c b/drivers/rpmsg/rpmsg_char.c
index 876ce43df732b..a6a33155ca859 100644
--- a/drivers/rpmsg/rpmsg_char.c
+++ b/drivers/rpmsg/rpmsg_char.c
@@ -458,7 +458,7 @@ static void rpmsg_ctrldev_release_device(struct device *dev)
ida_simple_remove(&rpmsg_ctrl_ida, dev->id);
ida_simple_remove(&rpmsg_minor_ida, MINOR(dev->devt));
- cdev_del(&ctrldev->cdev);
+ cdev_device_del(&ctrldev->cdev, &ctrldev->dev);
kfree(ctrldev);
}
@@ -493,19 +493,13 @@ static int rpmsg_chrdev_probe(struct rpmsg_device *rpdev)
dev->id = ret;
dev_set_name(&ctrldev->dev, "rpmsg_ctrl%d", ret);
- ret = cdev_add(&ctrldev->cdev, dev->devt, 1);
+ ret = cdev_device_add(&ctrldev->cdev, &ctrldev->dev);
if (ret)
goto free_ctrl_ida;
/* We can now rely on the release function for cleanup */
dev->release = rpmsg_ctrldev_release_device;
- ret = device_add(dev);
- if (ret) {
- dev_err(&rpdev->dev, "device_add failed: %d\n", ret);
- put_device(dev);
- }
-
dev_set_drvdata(&rpdev->dev, ctrldev);
return ret;
--
2.31.0
qcom_glink_rpdev_release() sets channel->rpdev to NULL. However, with
debug enabled kernel, qcom_glink_rpdev_release() gets delayed due to
delayed kobject release and channel gets released by that time and
triggers below kernel warning. To avoid this use-after-free, clear ept
pointers during ept destroy and channel release and add a new condition
in qcom_glink_rpdev_release() to access channel
| BUG: KASAN: use-after-free in qcom_glink_rpdev_release+0x54/0x70
| Write of size 8 at addr ffffffaba438e8d0 by task kworker/6:1/54
|
| CPU: 6 PID: 54 Comm: kworker/6:1 Not tainted 5.4.109-lockdep #16
| Hardware name: Google Lazor (rev3+) with KB Backlight (DT)
| Workqueue: events kobject_delayed_cleanup
| Call trace:
| dump_backtrace+0x0/0x284
| show_stack+0x20/0x2c
| dump_stack+0xd4/0x170
| print_address_description+0x3c/0x4a8
| __kasan_report+0x144/0x168
| kasan_report+0x10/0x18
| __asan_report_store8_noabort+0x1c/0x24
| qcom_glink_rpdev_release+0x54/0x70
| device_release+0x68/0x14c
| kobject_delayed_cleanup+0x158/0x2cc
| process_one_work+0x7cc/0x10a4
| worker_thread+0x80c/0xcec
| kthread+0x2a8/0x314
| ret_from_fork+0x10/0x18
Signed-off-by: Sujit Kautkar <[email protected]>
---
Changes in v3:
- Clear ept pointers and add extra conditions
Changes in v2:
- Fix typo in commit message
drivers/rpmsg/qcom_glink_native.c | 12 ++++++++++--
1 file changed, 10 insertions(+), 2 deletions(-)
diff --git a/drivers/rpmsg/qcom_glink_native.c b/drivers/rpmsg/qcom_glink_native.c
index e1444fefdd1c0..0c64a6f7a4f09 100644
--- a/drivers/rpmsg/qcom_glink_native.c
+++ b/drivers/rpmsg/qcom_glink_native.c
@@ -269,6 +269,9 @@ static void qcom_glink_channel_release(struct kref *ref)
idr_destroy(&channel->riids);
spin_unlock_irqrestore(&channel->intent_lock, flags);
+ if (channel->rpdev)
+ channel->rpdev->ept = NULL;
+
kfree(channel->name);
kfree(channel);
}
@@ -1214,6 +1217,8 @@ static void qcom_glink_destroy_ept(struct rpmsg_endpoint *ept)
channel->ept.cb = NULL;
spin_unlock_irqrestore(&channel->recv_lock, flags);
+ channel->rpdev->ept = NULL;
+
/* Decouple the potential rpdev from the channel */
channel->rpdev = NULL;
@@ -1371,9 +1376,12 @@ static const struct rpmsg_endpoint_ops glink_endpoint_ops = {
static void qcom_glink_rpdev_release(struct device *dev)
{
struct rpmsg_device *rpdev = to_rpmsg_device(dev);
- struct glink_channel *channel = to_glink_channel(rpdev->ept);
+ struct glink_channel *channel = NULL;
- channel->rpdev = NULL;
+ if (rpdev->ept) {
+ channel = to_glink_channel(rpdev->ept);
+ channel->rpdev = NULL;
+ }
kfree(rpdev);
}
--
2.31.0
Hi Sujit,
On Tue, Nov 02, 2021 at 04:51:49PM -0700, Sujit Kautkar wrote:
> qcom_glink_rpdev_release() sets channel->rpdev to NULL. However, with
> debug enabled kernel, qcom_glink_rpdev_release() gets delayed due to
> delayed kobject release and channel gets released by that time and
> triggers below kernel warning. To avoid this use-after-free, clear ept
> pointers during ept destroy and channel release and add a new condition
> in qcom_glink_rpdev_release() to access channel
>
> | BUG: KASAN: use-after-free in qcom_glink_rpdev_release+0x54/0x70
> | Write of size 8 at addr ffffffaba438e8d0 by task kworker/6:1/54
> |
> | CPU: 6 PID: 54 Comm: kworker/6:1 Not tainted 5.4.109-lockdep #16
> | Hardware name: Google Lazor (rev3+) with KB Backlight (DT)
> | Workqueue: events kobject_delayed_cleanup
> | Call trace:
> | dump_backtrace+0x0/0x284
> | show_stack+0x20/0x2c
> | dump_stack+0xd4/0x170
> | print_address_description+0x3c/0x4a8
> | __kasan_report+0x144/0x168
> | kasan_report+0x10/0x18
> | __asan_report_store8_noabort+0x1c/0x24
> | qcom_glink_rpdev_release+0x54/0x70
> | device_release+0x68/0x14c
> | kobject_delayed_cleanup+0x158/0x2cc
> | process_one_work+0x7cc/0x10a4
> | worker_thread+0x80c/0xcec
> | kthread+0x2a8/0x314
> | ret_from_fork+0x10/0x18
>
> Signed-off-by: Sujit Kautkar <[email protected]>
> ---
> Changes in v3:
> - Clear ept pointers and add extra conditions
>
> Changes in v2:
> - Fix typo in commit message
>
> drivers/rpmsg/qcom_glink_native.c | 12 ++++++++++--
> 1 file changed, 10 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/rpmsg/qcom_glink_native.c b/drivers/rpmsg/qcom_glink_native.c
> index e1444fefdd1c0..0c64a6f7a4f09 100644
> --- a/drivers/rpmsg/qcom_glink_native.c
> +++ b/drivers/rpmsg/qcom_glink_native.c
> @@ -269,6 +269,9 @@ static void qcom_glink_channel_release(struct kref *ref)
> idr_destroy(&channel->riids);
> spin_unlock_irqrestore(&channel->intent_lock, flags);
>
> + if (channel->rpdev)
> + channel->rpdev->ept = NULL;
> +
> kfree(channel->name);
> kfree(channel);
> }
> @@ -1214,6 +1217,8 @@ static void qcom_glink_destroy_ept(struct rpmsg_endpoint *ept)
> channel->ept.cb = NULL;
> spin_unlock_irqrestore(&channel->recv_lock, flags);
>
> + channel->rpdev->ept = NULL;
> +
> /* Decouple the potential rpdev from the channel */
> channel->rpdev = NULL;
>
> @@ -1371,9 +1376,12 @@ static const struct rpmsg_endpoint_ops glink_endpoint_ops = {
> static void qcom_glink_rpdev_release(struct device *dev)
> {
> struct rpmsg_device *rpdev = to_rpmsg_device(dev);
> - struct glink_channel *channel = to_glink_channel(rpdev->ept);
> + struct glink_channel *channel = NULL;
no need to initialize the pointer, it is assigned in the path that uses it.
>
> - channel->rpdev = NULL;
> + if (rpdev->ept) {
> + channel = to_glink_channel(rpdev->ept);
> + channel->rpdev = NULL;
> + }
> kfree(rpdev);
> }
Looks like this is already fixed in -next by:
commit 343ba27b6f9d473ec3e602cc648300eb03a7fa05
Author: Chris Lew <[email protected]>
Date: Thu Jul 30 10:48:15 2020 +0530
rpmsg: glink: Remove channel decouple from rpdev release
If a channel is being rapidly restarting and the kobj release worker is
busy, there is a chance the rpdev_release function will run after the
channel struct itself has been released.
There should not be a need to decouple the channel from rpdev in the
rpdev release since that should only happen from the close commands.
Signed-off-by: Chris Lew <[email protected]>
Signed-off-by: Deepak Kumar Singh <[email protected]>
Signed-off-by: Bjorn Andersson <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
On Tue, Nov 02, 2021 at 04:51:51PM -0700, Sujit Kautkar wrote:
> Replace cdev add/del APIs with cdev_device_add/cdev_device_del to avoid
> below kernel warning. This correctly takes a reference to the parent
> device so the parent will not get released until all references to the
> cdev are released.
>
> | ODEBUG: free active (active state 0) object type: timer_list hint: delayed_work_timer_fn+0x0/0x7c
> | WARNING: CPU: 7 PID: 19892 at lib/debugobjects.c:488 debug_print_object+0x13c/0x1b0
> | CPU: 7 PID: 19892 Comm: kworker/7:4 Tainted: G W 5.4.147-lockdep #1
> | ==================================================================
> | Hardware name: Google CoachZ (rev1 - 2) with LTE (DT)
> | Workqueue: events kobject_delayed_cleanup
> | pstate: 60c00009 (nZCv daif +PAN +UAO)
> | pc : debug_print_object+0x13c/0x1b0
> | lr : debug_print_object+0x13c/0x1b0
> | sp : ffffff83b2ec7970
> | x29: ffffff83b2ec7970 x28: dfffffd000000000
> | x27: ffffff83d674f000 x26: dfffffd000000000
> | x25: ffffffd06b8fa660 x24: dfffffd000000000
> | x23: 0000000000000000 x22: ffffffd06b7c5108
> | x21: ffffffd06d597860 x20: ffffffd06e2c21c0
> | x19: ffffffd06d5974c0 x18: 000000000001dad8
> | x17: 0000000000000000 x16: dfffffd000000000
> | BUG: KASAN: use-after-free in qcom_glink_rpdev_release+0x54/0x70
> | x15: ffffffffffffffff x14: 79616c6564203a74
> | x13: 0000000000000000 x12: 0000000000000080
> | Write of size 8 at addr ffffff83d95768d0 by task kworker/3:1/150
> | x11: 0000000000000001 x10: 0000000000000000
> | x9 : fc9e8edec0ad0300 x8 : fc9e8edec0ad0300
> |
> | x7 : 0000000000000000 x6 : 0000000000000000
> | x5 : 0000000000000080 x4 : 0000000000000000
> | CPU: 3 PID: 150 Comm: kworker/3:1 Tainted: G W 5.4.147-lockdep #1
> | x3 : ffffffd06c149574 x2 : ffffff83f77f7498
> | x1 : ffffffd06d596f60 x0 : 0000000000000061
> | Hardware name: Google CoachZ (rev1 - 2) with LTE (DT)
> | Call trace:
> | debug_print_object+0x13c/0x1b0
> | Workqueue: events kobject_delayed_cleanup
> | __debug_check_no_obj_freed+0x25c/0x3c0
> | debug_check_no_obj_freed+0x18/0x20
> | Call trace:
> | slab_free_freelist_hook+0xb4/0x1bc
> | kfree+0xe8/0x2d8
> | dump_backtrace+0x0/0x27c
> | rpmsg_ctrldev_release_device+0x78/0xb8
> | device_release+0x68/0x14c
> | show_stack+0x20/0x2c
> | kobject_cleanup+0x12c/0x298
> | kobject_delayed_cleanup+0x10/0x18
> | dump_stack+0xe0/0x19c
> | process_one_work+0x578/0x92c
> | worker_thread+0x804/0xcf8
> | print_address_description+0x3c/0x4a8
> | kthread+0x2a8/0x314
> | ret_from_fork+0x10/0x18
> | __kasan_report+0x100/0x124
>
> Signed-off-by: Sujit Kautkar <[email protected]>
Reviewed-by: Matthias Kaehlcke <[email protected]>
The subject is a little confusing. Maybe it should be "Use
cdev_device_{add,del}() instead of open coding".
Quoting Sujit Kautkar (2021-11-02 16:51:51)
> Replace cdev add/del APIs with cdev_device_add/cdev_device_del to avoid
> below kernel warning. This correctly takes a reference to the parent
> device so the parent will not get released until all references to the
> cdev are released.
>
> | ODEBUG: free active (active state 0) object type: timer_list hint: delayed_work_timer_fn+0x0/0x7c
> | WARNING: CPU: 7 PID: 19892 at lib/debugobjects.c:488 debug_print_object+0x13c/0x1b0
> | CPU: 7 PID: 19892 Comm: kworker/7:4 Tainted: G W 5.4.147-lockdep #1
> | ==================================================================
> | Hardware name: Google CoachZ (rev1 - 2) with LTE (DT)
> | Workqueue: events kobject_delayed_cleanup
> | pstate: 60c00009 (nZCv daif +PAN +UAO)
> | pc : debug_print_object+0x13c/0x1b0
> | lr : debug_print_object+0x13c/0x1b0
> | sp : ffffff83b2ec7970
> | x29: ffffff83b2ec7970 x28: dfffffd000000000
> | x27: ffffff83d674f000 x26: dfffffd000000000
> | x25: ffffffd06b8fa660 x24: dfffffd000000000
> | x23: 0000000000000000 x22: ffffffd06b7c5108
> | x21: ffffffd06d597860 x20: ffffffd06e2c21c0
> | x19: ffffffd06d5974c0 x18: 000000000001dad8
> | x17: 0000000000000000 x16: dfffffd000000000
> | BUG: KASAN: use-after-free in qcom_glink_rpdev_release+0x54/0x70
> | x15: ffffffffffffffff x14: 79616c6564203a74
> | x13: 0000000000000000 x12: 0000000000000080
> | Write of size 8 at addr ffffff83d95768d0 by task kworker/3:1/150
> | x11: 0000000000000001 x10: 0000000000000000
> | x9 : fc9e8edec0ad0300 x8 : fc9e8edec0ad0300
> |
> | x7 : 0000000000000000 x6 : 0000000000000000
> | x5 : 0000000000000080 x4 : 0000000000000000
> | CPU: 3 PID: 150 Comm: kworker/3:1 Tainted: G W 5.4.147-lockdep #1
> | x3 : ffffffd06c149574 x2 : ffffff83f77f7498
> | x1 : ffffffd06d596f60 x0 : 0000000000000061
> | Hardware name: Google CoachZ (rev1 - 2) with LTE (DT)
> | Call trace:
> | debug_print_object+0x13c/0x1b0
> | Workqueue: events kobject_delayed_cleanup
> | __debug_check_no_obj_freed+0x25c/0x3c0
> | debug_check_no_obj_freed+0x18/0x20
> | Call trace:
> | slab_free_freelist_hook+0xb4/0x1bc
> | kfree+0xe8/0x2d8
> | dump_backtrace+0x0/0x27c
> | rpmsg_ctrldev_release_device+0x78/0xb8
> | device_release+0x68/0x14c
> | show_stack+0x20/0x2c
> | kobject_cleanup+0x12c/0x298
> | kobject_delayed_cleanup+0x10/0x18
> | dump_stack+0xe0/0x19c
> | process_one_work+0x578/0x92c
> | worker_thread+0x804/0xcf8
> | print_address_description+0x3c/0x4a8
> | kthread+0x2a8/0x314
> | ret_from_fork+0x10/0x18
> | __kasan_report+0x100/0x124
>
> Signed-off-by: Sujit Kautkar <[email protected]>
> ---
Reviewed-by: Stephen Boyd <[email protected]>
On Tue 02 Nov 18:51 CDT 2021, Sujit Kautkar wrote:
I like Stephen's suggestion about modifying the $subject.
Also note that the change isn't in the glink driver, so prefix should
reflect that:
$ git log --oneline --no-decorate -- drivers/rpmsg/rpmsg_char.c
f998d48f9b3c rpmsg: glink: Update cdev add/del API in rpmsg_ctrldev_release_device()
bc774a3887cb rpmsg: char: Remove useless include
964e8bedd5a1 rpmsg: char: Return an error if device already open
...
> Replace cdev add/del APIs with cdev_device_add/cdev_device_del to avoid
> below kernel warning. This correctly takes a reference to the parent
> device so the parent will not get released until all references to the
> cdev are released.
>
> | ODEBUG: free active (active state 0) object type: timer_list hint: delayed_work_timer_fn+0x0/0x7c
> | WARNING: CPU: 7 PID: 19892 at lib/debugobjects.c:488 debug_print_object+0x13c/0x1b0
> | CPU: 7 PID: 19892 Comm: kworker/7:4 Tainted: G W 5.4.147-lockdep #1
> | ==================================================================
> | Hardware name: Google CoachZ (rev1 - 2) with LTE (DT)
> | Workqueue: events kobject_delayed_cleanup
> | pstate: 60c00009 (nZCv daif +PAN +UAO)
> | pc : debug_print_object+0x13c/0x1b0
> | lr : debug_print_object+0x13c/0x1b0
> | sp : ffffff83b2ec7970
> | x29: ffffff83b2ec7970 x28: dfffffd000000000
> | x27: ffffff83d674f000 x26: dfffffd000000000
> | x25: ffffffd06b8fa660 x24: dfffffd000000000
> | x23: 0000000000000000 x22: ffffffd06b7c5108
> | x21: ffffffd06d597860 x20: ffffffd06e2c21c0
> | x19: ffffffd06d5974c0 x18: 000000000001dad8
> | x17: 0000000000000000 x16: dfffffd000000000
> | BUG: KASAN: use-after-free in qcom_glink_rpdev_release+0x54/0x70
> | x15: ffffffffffffffff x14: 79616c6564203a74
> | x13: 0000000000000000 x12: 0000000000000080
> | Write of size 8 at addr ffffff83d95768d0 by task kworker/3:1/150
> | x11: 0000000000000001 x10: 0000000000000000
> | x9 : fc9e8edec0ad0300 x8 : fc9e8edec0ad0300
> |
> | x7 : 0000000000000000 x6 : 0000000000000000
> | x5 : 0000000000000080 x4 : 0000000000000000
> | CPU: 3 PID: 150 Comm: kworker/3:1 Tainted: G W 5.4.147-lockdep #1
> | x3 : ffffffd06c149574 x2 : ffffff83f77f7498
> | x1 : ffffffd06d596f60 x0 : 0000000000000061
> | Hardware name: Google CoachZ (rev1 - 2) with LTE (DT)
> | Call trace:
> | debug_print_object+0x13c/0x1b0
> | Workqueue: events kobject_delayed_cleanup
> | __debug_check_no_obj_freed+0x25c/0x3c0
> | debug_check_no_obj_freed+0x18/0x20
> | Call trace:
> | slab_free_freelist_hook+0xb4/0x1bc
> | kfree+0xe8/0x2d8
> | dump_backtrace+0x0/0x27c
Why is dump_backtrace in the callstack here inbetween
rpmsg_ctrldev_release_device() and kfree()? Isn't the error that we're
calling kfree() on an chunk of memory that contains a live object?
> | rpmsg_ctrldev_release_device+0x78/0xb8
> | device_release+0x68/0x14c
> | show_stack+0x20/0x2c
> | kobject_cleanup+0x12c/0x298
> | kobject_delayed_cleanup+0x10/0x18
> | dump_stack+0xe0/0x19c
> | process_one_work+0x578/0x92c
> | worker_thread+0x804/0xcf8
> | print_address_description+0x3c/0x4a8
> | kthread+0x2a8/0x314
> | ret_from_fork+0x10/0x18
> | __kasan_report+0x100/0x124
>
> Signed-off-by: Sujit Kautkar <[email protected]>
> ---
> Changes in v3:
> - Remove unecessary error check as per Matthias's comment
>
> Changes in v2:
> - Fix typo in commit message
>
> drivers/rpmsg/rpmsg_char.c | 10 ++--------
> 1 file changed, 2 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/rpmsg/rpmsg_char.c b/drivers/rpmsg/rpmsg_char.c
> index 876ce43df732b..a6a33155ca859 100644
> --- a/drivers/rpmsg/rpmsg_char.c
> +++ b/drivers/rpmsg/rpmsg_char.c
> @@ -458,7 +458,7 @@ static void rpmsg_ctrldev_release_device(struct device *dev)
>
> ida_simple_remove(&rpmsg_ctrl_ida, dev->id);
> ida_simple_remove(&rpmsg_minor_ida, MINOR(dev->devt));
> - cdev_del(&ctrldev->cdev);
> + cdev_device_del(&ctrldev->cdev, &ctrldev->dev);
I am not able to find any other instance where cdev_device_del() is
called from the device's release function itself, which tells me that
this probably is not the right thing to do. Instead the appropriate way
seem to put the cdev_device_del() in rpmsg_chrdev_remove().
That said, we already do device_del() in rpmsg_chrdev_remove() so if the
warning is trying to tell us that ctrldev->dev is not deleted I think we
have an unbalanced put_device()?
Regards,
Bjorn
> kfree(ctrldev);
> }
>
> @@ -493,19 +493,13 @@ static int rpmsg_chrdev_probe(struct rpmsg_device *rpdev)
> dev->id = ret;
> dev_set_name(&ctrldev->dev, "rpmsg_ctrl%d", ret);
>
> - ret = cdev_add(&ctrldev->cdev, dev->devt, 1);
> + ret = cdev_device_add(&ctrldev->cdev, &ctrldev->dev);
> if (ret)
> goto free_ctrl_ida;
>
> /* We can now rely on the release function for cleanup */
> dev->release = rpmsg_ctrldev_release_device;
>
> - ret = device_add(dev);
> - if (ret) {
> - dev_err(&rpdev->dev, "device_add failed: %d\n", ret);
> - put_device(dev);
> - }
> -
> dev_set_drvdata(&rpdev->dev, ctrldev);
>
> return ret;
> --
> 2.31.0
>
On Wed, Nov 17, 2021 at 05:29:07PM -0600, Bjorn Andersson wrote:
> On Tue 02 Nov 18:51 CDT 2021, Sujit Kautkar wrote:
>
> I like Stephen's suggestion about modifying the $subject.
> Also note that the change isn't in the glink driver, so prefix should
> reflect that:
>
> $ git log --oneline --no-decorate -- drivers/rpmsg/rpmsg_char.c
> f998d48f9b3c rpmsg: glink: Update cdev add/del API in rpmsg_ctrldev_release_device()
> bc774a3887cb rpmsg: char: Remove useless include
> 964e8bedd5a1 rpmsg: char: Return an error if device already open
> ...
>
> > Replace cdev add/del APIs with cdev_device_add/cdev_device_del to avoid
> > below kernel warning. This correctly takes a reference to the parent
> > device so the parent will not get released until all references to the
> > cdev are released.
> >
> > | ODEBUG: free active (active state 0) object type: timer_list hint: delayed_work_timer_fn+0x0/0x7c
> > | WARNING: CPU: 7 PID: 19892 at lib/debugobjects.c:488 debug_print_object+0x13c/0x1b0
> > | CPU: 7 PID: 19892 Comm: kworker/7:4 Tainted: G W 5.4.147-lockdep #1
> > | ==================================================================
> > | Hardware name: Google CoachZ (rev1 - 2) with LTE (DT)
> > | Workqueue: events kobject_delayed_cleanup
> > | pstate: 60c00009 (nZCv daif +PAN +UAO)
> > | pc : debug_print_object+0x13c/0x1b0
> > | lr : debug_print_object+0x13c/0x1b0
> > | sp : ffffff83b2ec7970
> > | x29: ffffff83b2ec7970 x28: dfffffd000000000
> > | x27: ffffff83d674f000 x26: dfffffd000000000
> > | x25: ffffffd06b8fa660 x24: dfffffd000000000
> > | x23: 0000000000000000 x22: ffffffd06b7c5108
> > | x21: ffffffd06d597860 x20: ffffffd06e2c21c0
> > | x19: ffffffd06d5974c0 x18: 000000000001dad8
> > | x17: 0000000000000000 x16: dfffffd000000000
> > | BUG: KASAN: use-after-free in qcom_glink_rpdev_release+0x54/0x70
> > | x15: ffffffffffffffff x14: 79616c6564203a74
> > | x13: 0000000000000000 x12: 0000000000000080
> > | Write of size 8 at addr ffffff83d95768d0 by task kworker/3:1/150
> > | x11: 0000000000000001 x10: 0000000000000000
> > | x9 : fc9e8edec0ad0300 x8 : fc9e8edec0ad0300
> > |
> > | x7 : 0000000000000000 x6 : 0000000000000000
> > | x5 : 0000000000000080 x4 : 0000000000000000
> > | CPU: 3 PID: 150 Comm: kworker/3:1 Tainted: G W 5.4.147-lockdep #1
> > | x3 : ffffffd06c149574 x2 : ffffff83f77f7498
> > | x1 : ffffffd06d596f60 x0 : 0000000000000061
> > | Hardware name: Google CoachZ (rev1 - 2) with LTE (DT)
> > | Call trace:
> > | debug_print_object+0x13c/0x1b0
> > | Workqueue: events kobject_delayed_cleanup
> > | __debug_check_no_obj_freed+0x25c/0x3c0
> > | debug_check_no_obj_freed+0x18/0x20
> > | Call trace:
> > | slab_free_freelist_hook+0xb4/0x1bc
> > | kfree+0xe8/0x2d8
> > | dump_backtrace+0x0/0x27c
>
> Why is dump_backtrace in the callstack here inbetween
> rpmsg_ctrldev_release_device() and kfree()? Isn't the error that we're
> calling kfree() on an chunk of memory that contains a live object?
When I tried to repro there was no dump_backtrace():
Call trace:
debug_print_object+0x13c/0x1b0
__debug_check_no_obj_freed+0x25c/0x3c0
debug_check_no_obj_freed+0x18/0x20
slab_free_freelist_hook+0xbc/0x1e4
kfree+0xfc/0x2f4
rpmsg_ctrldev_release_device+0x78/0xb8
device_release+0x84/0x168
kobject_cleanup+0x12c/0x298
kobject_delayed_cleanup+0x10/0x18
process_one_work+0x578/0x92c
worker_thread+0x804/0xcf8
kthread+0x2a8/0x314
ret_from_fork+0x10/0x18
My guess is that Sujit added a dump_backtrace() for debugging and it was
still there when the backtrace of the commit message was generated. That
would also explain the two 'Call trace:' entries in the log.
> > | rpmsg_ctrldev_release_device+0x78/0xb8
> > | device_release+0x68/0x14c
> > | show_stack+0x20/0x2c
> > | kobject_cleanup+0x12c/0x298
> > | kobject_delayed_cleanup+0x10/0x18
> > | dump_stack+0xe0/0x19c
> > | process_one_work+0x578/0x92c
> > | worker_thread+0x804/0xcf8
> > | print_address_description+0x3c/0x4a8
> > | kthread+0x2a8/0x314
> > | ret_from_fork+0x10/0x18
> > | __kasan_report+0x100/0x124
> >
> > Signed-off-by: Sujit Kautkar <[email protected]>
> > ---
> > Changes in v3:
> > - Remove unecessary error check as per Matthias's comment
> >
> > Changes in v2:
> > - Fix typo in commit message
> >
> > drivers/rpmsg/rpmsg_char.c | 10 ++--------
> > 1 file changed, 2 insertions(+), 8 deletions(-)
> >
> > diff --git a/drivers/rpmsg/rpmsg_char.c b/drivers/rpmsg/rpmsg_char.c
> > index 876ce43df732b..a6a33155ca859 100644
> > --- a/drivers/rpmsg/rpmsg_char.c
> > +++ b/drivers/rpmsg/rpmsg_char.c
> > @@ -458,7 +458,7 @@ static void rpmsg_ctrldev_release_device(struct device *dev)
> >
> > ida_simple_remove(&rpmsg_ctrl_ida, dev->id);
> > ida_simple_remove(&rpmsg_minor_ida, MINOR(dev->devt));
> > - cdev_del(&ctrldev->cdev);
> > + cdev_device_del(&ctrldev->cdev, &ctrldev->dev);
>
> I am not able to find any other instance where cdev_device_del() is
> called from the device's release function itself, which tells me that
> this probably is not the right thing to do. Instead the appropriate way
> seem to put the cdev_device_del() in rpmsg_chrdev_remove().
Yes, it sounds reasonable to me to delete the char device when the control
device is removed.
> That said, we already do device_del() in rpmsg_chrdev_remove() so if the
> warning is trying to tell us that ctrldev->dev is not deleted I think we
> have an unbalanced put_device()?
My understanding is that the situation is analogous to this one:
commit 1413ef638abae4ab5621901cf4d8ef08a4a48ba6
Author: Kevin Hao <[email protected]>
Date: Fri Oct 11 23:00:14 2019 +0800
i2c: dev: Fix the race between the release of i2c_dev and cdev
The struct cdev is embedded in the struct i2c_dev. In the current code,
we would free the i2c_dev struct directly in put_i2c_dev(), but the
cdev is manged by a kobject, and the release of it is not predictable.
So it is very possible that the i2c_dev is freed before the cdev is
entirely released. We can easily get the following call trace with
CONFIG_DEBUG_KOBJECT_RELEASE and CONFIG_DEBUG_OBJECTS_TIMERS enabled.
ODEBUG: free active (active state 0) object type: timer_list hint: delayed_work_timer_fn+0x0/0x38
WARNING: CPU: 19 PID: 1 at lib/debugobjects.c:325 debug_print_object+0xb0/0xf0
...
This is a common issue when using cdev embedded in a struct.
Fortunately, we already have a mechanism to solve this kind of issue.
Please see commit 233ed09d7fda ("chardev: add helper function to
egister char devs with a struct device") for more detail.
In this patch, we choose to embed the struct device into the i2c_dev,
and use the API provided by the commit 233ed09d7fda to make sure that
the release of i2c_dev and cdev are in sequence.