2022-01-10 17:12:40

by Matthias Kaehlcke

[permalink] [raw]
Subject: [PATCH v5 1/2] rpmsg: char: Fix race between the release of rpmsg_ctrldev and cdev

From: Sujit Kautkar <[email protected]>

struct rpmsg_ctrldev contains a struct cdev. The current code frees
the rpmsg_ctrldev struct in rpmsg_ctrldev_release_device(), but the
cdev is a managed object, therefore its release is not predictable
and the rpmsg_ctrldev could be freed before the cdev is entirely
released, as in the backtrace below.

[ 93.625603] ODEBUG: free active (active state 0) object type: timer_list hint: delayed_work_timer_fn+0x0/0x7c
[ 93.636115] WARNING: CPU: 0 PID: 12 at lib/debugobjects.c:488 debug_print_object+0x13c/0x1b0
[ 93.644799] Modules linked in: veth xt_cgroup xt_MASQUERADE rfcomm algif_hash algif_skcipher af_alg uinput ip6table_nat fuse uvcvideo videobuf2_vmalloc venus_enc venus_dec videobuf2_dma_contig hci_uart btandroid btqca snd_soc_rt5682_i2c bluetooth qcom_spmi_temp_alarm snd_soc_rt5682v
[ 93.715175] CPU: 0 PID: 12 Comm: kworker/0:1 Tainted: G B 5.4.163-lockdep #26
[ 93.723855] Hardware name: Google Lazor (rev3 - 8) with LTE (DT)
[ 93.730055] Workqueue: events kobject_delayed_cleanup
[ 93.735271] pstate: 60c00009 (nZCv daif +PAN +UAO)
[ 93.740216] pc : debug_print_object+0x13c/0x1b0
[ 93.744890] lr : debug_print_object+0x13c/0x1b0
[ 93.749555] sp : ffffffacf5bc7940
[ 93.752978] x29: ffffffacf5bc7940 x28: dfffffd000000000
[ 93.758448] x27: ffffffacdb11a800 x26: dfffffd000000000
[ 93.763916] x25: ffffffd0734f856c x24: dfffffd000000000
[ 93.769389] x23: 0000000000000000 x22: ffffffd0733c35b0
[ 93.774860] x21: ffffffd0751994a0 x20: ffffffd075ec27c0
[ 93.780338] x19: ffffffd075199100 x18: 00000000000276e0
[ 93.785814] x17: 0000000000000000 x16: dfffffd000000000
[ 93.791291] x15: ffffffffffffffff x14: 6e6968207473696c
[ 93.796768] x13: 0000000000000000 x12: ffffffd075e2b000
[ 93.802244] x11: 0000000000000001 x10: 0000000000000000
[ 93.807723] x9 : d13400dff1921900 x8 : d13400dff1921900
[ 93.813200] x7 : 0000000000000000 x6 : 0000000000000000
[ 93.818676] x5 : 0000000000000080 x4 : 0000000000000000
[ 93.824152] x3 : ffffffd0732a0fa4 x2 : 0000000000000001
[ 93.829628] x1 : ffffffacf5bc7580 x0 : 0000000000000061
[ 93.835104] Call trace:
[ 93.837644] debug_print_object+0x13c/0x1b0
[ 93.841963] __debug_check_no_obj_freed+0x25c/0x3c0
[ 93.846987] debug_check_no_obj_freed+0x18/0x20
[ 93.851669] slab_free_freelist_hook+0xbc/0x1e4
[ 93.856346] kfree+0xfc/0x2f4
[ 93.859416] rpmsg_ctrldev_release_device+0x78/0xb8
[ 93.864445] device_release+0x84/0x168
[ 93.868310] kobject_cleanup+0x12c/0x298
[ 93.872356] kobject_delayed_cleanup+0x10/0x18
[ 93.876948] process_one_work+0x578/0x92c
[ 93.881086] worker_thread+0x804/0xcf8
[ 93.884963] kthread+0x2a8/0x314
[ 93.888303] ret_from_fork+0x10/0x18

The cdev_device_add/del() API was created to address this issue
(see commit 233ed09d7fda), use it instead of cdev add/del().

Fixes: c0cdc19f84a4 ("rpmsg: Driver for user space endpoint interface")
Signed-off-by: Sujit Kautkar <[email protected]>
Signed-off-by: Matthias Kaehlcke <[email protected]>
Reviewed-by: Mathieu Poirier <[email protected]>
Reviewed-by: Bjorn Andersson <[email protected]>
---

Changes in v5:
- added 'Fixes' tag
- added 'Reviewed-by' tags from Mathieu and Bjorn

Changes in v4:
- call cdev_device_del() from rpmsg_chrdev_remove() instead of
rpmsg_ctrldev_release_device()
- updated subject (was: "rpmsg: glink: Update cdev add/del API in
rpmsg_ctrldev_release_device()")
- updated commit message
- replaced backtrace in commit message with one that doesn't have
a dump_backtrace() call

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 | 11 ++---------
1 file changed, 2 insertions(+), 9 deletions(-)

diff --git a/drivers/rpmsg/rpmsg_char.c b/drivers/rpmsg/rpmsg_char.c
index 2bebc9b2d163..ba85f5d11960 100644
--- a/drivers/rpmsg/rpmsg_char.c
+++ b/drivers/rpmsg/rpmsg_char.c
@@ -461,7 +461,6 @@ 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);
kfree(ctrldev);
}

@@ -496,19 +495,13 @@ 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;
@@ -534,7 +527,7 @@ static void rpmsg_chrdev_remove(struct rpmsg_device *rpdev)
if (ret)
dev_warn(&rpdev->dev, "failed to nuke endpoints: %d\n", ret);

- device_del(&ctrldev->dev);
+ cdev_device_del(&ctrldev->cdev, &ctrldev->dev);
put_device(&ctrldev->dev);
}

--
2.34.1.575.g55b058a8bb-goog



2022-01-10 17:12:42

by Matthias Kaehlcke

[permalink] [raw]
Subject: [PATCH v5 2/2] rpmsg: char: Fix race between the release of rpmsg_eptdev and cdev

struct rpmsg_eptdev contains a struct cdev. The current code frees
the rpmsg_eptdev struct in rpmsg_eptdev_destroy(), but the cdev is
a managed object, therefore its release is not predictable and the
rpmsg_eptdev could be freed before the cdev is entirely released.

The cdev_device_add/del() API was created to address this issue
(see commit 233ed09d7fda), use it instead of cdev add/del().

Fixes: c0cdc19f84a4 ("rpmsg: Driver for user space endpoint interface")
Signed-off-by: Matthias Kaehlcke <[email protected]>
---

Changes in v5:
- patch added to the series

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 ba85f5d11960..5bc1e6017587 100644
--- a/drivers/rpmsg/rpmsg_char.c
+++ b/drivers/rpmsg/rpmsg_char.c
@@ -92,7 +92,7 @@ static int rpmsg_eptdev_destroy(struct device *dev, void *data)
/* wake up any blocked readers */
wake_up_interruptible(&eptdev->readq);

- device_del(&eptdev->dev);
+ cdev_device_del(&eptdev->cdev, &eptdev->dev);
put_device(&eptdev->dev);

return 0;
@@ -380,19 +380,13 @@ static int rpmsg_eptdev_create(struct rpmsg_ctrldev *ctrldev,
dev->id = ret;
dev_set_name(dev, "rpmsg%d", ret);

- ret = cdev_add(&eptdev->cdev, dev->devt, 1);
+ ret = cdev_device_add(&eptdev->cdev, &eptdev->dev);
if (ret)
goto free_ept_ida;

/* We can now rely on the release function for cleanup */
dev->release = rpmsg_eptdev_release_device;

- ret = device_add(dev);
- if (ret) {
- dev_err(dev, "device_add failed: %d\n", ret);
- put_device(dev);
- }
-
return ret;

free_ept_ida:
--
2.34.1.575.g55b058a8bb-goog


2022-01-10 18:09:53

by Mathieu Poirier

[permalink] [raw]
Subject: Re: [PATCH v5 2/2] rpmsg: char: Fix race between the release of rpmsg_eptdev and cdev

On Mon, Jan 10, 2022 at 09:12:32AM -0800, Matthias Kaehlcke wrote:
> struct rpmsg_eptdev contains a struct cdev. The current code frees
> the rpmsg_eptdev struct in rpmsg_eptdev_destroy(), but the cdev is
> a managed object, therefore its release is not predictable and the
> rpmsg_eptdev could be freed before the cdev is entirely released.
>
> The cdev_device_add/del() API was created to address this issue
> (see commit 233ed09d7fda), use it instead of cdev add/del().
>
> Fixes: c0cdc19f84a4 ("rpmsg: Driver for user space endpoint interface")
> Signed-off-by: Matthias Kaehlcke <[email protected]>
> ---
>
> Changes in v5:
> - patch added to the series
>
> 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 ba85f5d11960..5bc1e6017587 100644
> --- a/drivers/rpmsg/rpmsg_char.c
> +++ b/drivers/rpmsg/rpmsg_char.c
> @@ -92,7 +92,7 @@ static int rpmsg_eptdev_destroy(struct device *dev, void *data)
> /* wake up any blocked readers */
> wake_up_interruptible(&eptdev->readq);
>
> - device_del(&eptdev->dev);
> + cdev_device_del(&eptdev->cdev, &eptdev->dev);
> put_device(&eptdev->dev);
>
> return 0;
> @@ -380,19 +380,13 @@ static int rpmsg_eptdev_create(struct rpmsg_ctrldev *ctrldev,
> dev->id = ret;
> dev_set_name(dev, "rpmsg%d", ret);
>
> - ret = cdev_add(&eptdev->cdev, dev->devt, 1);
> + ret = cdev_device_add(&eptdev->cdev, &eptdev->dev);
> if (ret)
> goto free_ept_ida;
>
> /* We can now rely on the release function for cleanup */
> dev->release = rpmsg_eptdev_release_device;
>
> - ret = device_add(dev);
> - if (ret) {
> - dev_err(dev, "device_add failed: %d\n", ret);
> - put_device(dev);
> - }
> -
> return ret;
>

Reviewed-by: Mathieu Poirier <[email protected]>

> free_ept_ida:
> --
> 2.34.1.575.g55b058a8bb-goog
>

2022-01-10 18:18:53

by Bjorn Andersson

[permalink] [raw]
Subject: Re: [PATCH v5 2/2] rpmsg: char: Fix race between the release of rpmsg_eptdev and cdev

On Mon 10 Jan 09:12 PST 2022, Matthias Kaehlcke wrote:

> struct rpmsg_eptdev contains a struct cdev. The current code frees
> the rpmsg_eptdev struct in rpmsg_eptdev_destroy(), but the cdev is
> a managed object, therefore its release is not predictable and the
> rpmsg_eptdev could be freed before the cdev is entirely released.
>
> The cdev_device_add/del() API was created to address this issue
> (see commit 233ed09d7fda), use it instead of cdev add/del().
>
> Fixes: c0cdc19f84a4 ("rpmsg: Driver for user space endpoint interface")
> Signed-off-by: Matthias Kaehlcke <[email protected]>
> ---
>
> Changes in v5:
> - patch added to the series
>
> 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 ba85f5d11960..5bc1e6017587 100644
> --- a/drivers/rpmsg/rpmsg_char.c
> +++ b/drivers/rpmsg/rpmsg_char.c

There's a cdev_del() in rpmsg_eptdev_release_device() that I think needs
to go as well, in line with patch 1.

Regards,
Bjorn

> @@ -92,7 +92,7 @@ static int rpmsg_eptdev_destroy(struct device *dev, void *data)
> /* wake up any blocked readers */
> wake_up_interruptible(&eptdev->readq);
>
> - device_del(&eptdev->dev);
> + cdev_device_del(&eptdev->cdev, &eptdev->dev);
> put_device(&eptdev->dev);
>
> return 0;
> @@ -380,19 +380,13 @@ static int rpmsg_eptdev_create(struct rpmsg_ctrldev *ctrldev,
> dev->id = ret;
> dev_set_name(dev, "rpmsg%d", ret);
>
> - ret = cdev_add(&eptdev->cdev, dev->devt, 1);
> + ret = cdev_device_add(&eptdev->cdev, &eptdev->dev);
> if (ret)
> goto free_ept_ida;
>
> /* We can now rely on the release function for cleanup */
> dev->release = rpmsg_eptdev_release_device;
>
> - ret = device_add(dev);
> - if (ret) {
> - dev_err(dev, "device_add failed: %d\n", ret);
> - put_device(dev);
> - }
> -
> return ret;
>
> free_ept_ida:
> --
> 2.34.1.575.g55b058a8bb-goog
>

2022-01-10 18:37:23

by Matthias Kaehlcke

[permalink] [raw]
Subject: Re: [PATCH v5 2/2] rpmsg: char: Fix race between the release of rpmsg_eptdev and cdev

On Mon, Jan 10, 2022 at 10:19:32AM -0800, Bjorn Andersson wrote:
> On Mon 10 Jan 09:12 PST 2022, Matthias Kaehlcke wrote:
>
> > struct rpmsg_eptdev contains a struct cdev. The current code frees
> > the rpmsg_eptdev struct in rpmsg_eptdev_destroy(), but the cdev is
> > a managed object, therefore its release is not predictable and the
> > rpmsg_eptdev could be freed before the cdev is entirely released.
> >
> > The cdev_device_add/del() API was created to address this issue
> > (see commit 233ed09d7fda), use it instead of cdev add/del().
> >
> > Fixes: c0cdc19f84a4 ("rpmsg: Driver for user space endpoint interface")
> > Signed-off-by: Matthias Kaehlcke <[email protected]>
> > ---
> >
> > Changes in v5:
> > - patch added to the series
> >
> > 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 ba85f5d11960..5bc1e6017587 100644
> > --- a/drivers/rpmsg/rpmsg_char.c
> > +++ b/drivers/rpmsg/rpmsg_char.c
>
> There's a cdev_del() in rpmsg_eptdev_release_device() that I think needs
> to go as well, in line with patch 1.

Good catch, thanks, will fix.