media: v4l2-subdev: Fix NULL dereferences racing with unbind
v3 cover media: v4l2-subdev: Fix NULL dereferences racing with unbind
Hi,
v4l2_device_unregister_subdev() clears sd->v4l2_dev and, through media_gobj_destroy(), sd->entity.graph_obj.mdev before it unregisters the sub-device node, and drivers that call media_device_unregister() first, as vimc does, clear the latter even earlier. subdev_open() dereferences both pointers, and the EXT_CTRLS ioctls the first one, so a driver unbind racing with an open or an ioctl on the node oopses. syzbot hits the open() case by unbinding vimc.
Patch 2/3 of v2, now patch 1, checked the two pointers in subdev_open(). That only narrows the race: the sub-device can be unregistered right after the check. A patch along the same lines was also posted in [1]. This version takes a different approach: it stops reading the sub-device's registration state in these file operations and uses the node's own vdev->v4l2_dev instead, which unregistration never clears. subdev_open() and the EXT_CTRLS ioctls no longer read a pointer that the V4L2 core or the driver core clears when a driver is unbound.
What this does not fix is the lifetime of the sub-device and of the media device when a driver goes away with file handles open, which is the known limitation Sakari mentioned on v2 [2]. The commit messages say so.
Changes since v2:
- Patch 1 (was 2/3): use vdev->v4l2_dev->mdev instead of checking sd->v4l2_dev and sd->entity.graph_obj.mdev, and fail with -ENODEV if the media device's driver has been unbound meanwhile.
- Patch 2 is new: the same dereference of sd->v4l2_dev in the EXT_CTRLS ioctls, found by reading the code.
- Dropped the two ipu6 patches of v2 (1/3 and 3/3), as I said I would in [3].
- Rebased on media next.
Testing:
- Built with W=1 (allmodconfig) at each patch of the series on media next 8e26d4c20ed2.
- Run in QEMU/KVM (x86_64, 4 vCPUs, KASAN generic inline, vimc built in) on media next 2dcdfb625c3b, which has the same v4l2-subdev.c, one boot per 60 s run. 16 threads either open and close vimc's sub-device nodes, or loop on VIDIOC_G_EXT_CTRLS on handles they keep open across unbinds, while another thread unbinds and rebinds vimc through sysfs every 1 or 30 ms. Each oops kills the thread that hits it; the test re-creates up to 100 per run, so the counts are capped:
kernel open() runs EXT_CTRLS runs
media next 116 + 116 oops -
+ patch 1 0 + 0 116 + 116 oops
+ patches 1 and 2 0 + 0 0 + 0 The oopses are KASAN null-ptr-derefs in subdev_open() and subdev_do_ioctl() respectively. The EXT_CTRLS runs are not done on plain media next, since they open the nodes too and would oops in subdev_open() as well. The open() runs with patch 1 and all the runs with both patches showed no oops, KASAN report or warning at all:
about 8 million successful opens, 1.1 million opens refused with -ENODEV while vimc was being unbound (the test does not tell v4l2_open()'s own check from the new one), and, with both patches, 880 million EXT_CTRLS ioctls.
- Run on the IPU6 tablet where I first hit the open() oops (Intel N100, Chuwi Hi10 X1), with two GalaxyCore sensors, GC5035 and GC8034, whose drivers I have not posted yet. Kernel: media next 9cfc1aca0781 plus those drivers and their ipu-bridge and int3472 changes, with KASAN (generic inline), lockdep and UBSAN. 4 processes either open and close every sub-device node, or loop on VIDIOC_G_EXT_CTRLS on the sensor's node, keeping it open across unbinds, while the sensor's I2C driver is unbound and rebound 300 times through sysfs. One run per sensor and test:
kernel open() runs EXT_CTRLS runs
media next oops at the 4th - unbind of the first
+ patches 1 and 2 0 + 0 0 + 0
On media next the oops is the KASAN null-ptr-deref in subdev_open(). The test stops at the first oops, so the other runs were not done there, and the tablet gives no before and after for patch 2. With both patches: 2.9 million successful opens, 1048 refused with -ENODEV (again from either check), 81 million EXT_CTRLS ioctls, and v4l2-compliance and a 10-frame capture from each sensor still pass after the runs. One warning in the open() run on GC5035:
DEBUG_LOCKS_WARN_ON(lock->magic != lock), from __v4l2_subdev_state_alloc() in subdev_open(). The GC5035 driver uses its control handler's mutex as the state lock, and the open took it after the unbind had destroyed it in v4l2_ctrl_handler_free(): that is the lifetime limitation above, which these patches do not touch.
- syzbot has not tested the series: its report has no public reproducer.
- I can post the test program if that helps.
[1] https://lore.kernel.org/linux-media/[email protected]/
[2] https://lore.kernel.org/linux-media/[email protected]/
[3] https://lore.kernel.org/linux-media/[email protected]/
Nicola Fiorillo (2):
media: v4l2-subdev: Fix NULL pointer dereference in subdev_open()
media: v4l2-subdev: Fix NULL pointer dereference in the EXT_CTRLS
ioctls
drivers/media/v4l2-core/v4l2-subdev.c | 18 +++++++++++++----- 1 file changed, 13 insertions(+), 5 deletions(-)
base-commit: 8e26d4c20ed218db6c3123480a4a4e2545680566
Signature · 1 line
2.47.3
PATCH v3 1/2 media: v4l2-subdev: Fix NULL pointer dereference in subdev_open()
Unbinding a driver while something opens a /dev/v4l-subdevN node oopses the kernel:
BUG: kernel NULL pointer dereference, address: 0000000000000008 RIP: 0010:subdev_open+0x8a/0x190 [videodev] Call Trace: v4l2_open+0xa9/0x100 [videodev] chrdev_open+0xb2/0x230 do_dentry_open+0x14c/0x440 vfs_open+0x2e/0xe0 path_openat+0x82e/0x12d0 do_filp_open+0xc4/0x170 do_sys_openat2+0xae/0xe0 __x64_sys_openat+0x55/0xa0
v4l2_device_unregister_subdev() clears sd->v4l2_dev, then unregisters the media entity, which clears sd->entity.graph_obj.mdev, and only then unregisters the device node. Drivers that call media_device_unregister() before v4l2_device_unregister(), as vimc does, clear sd->entity.graph_obj.mdev even earlier, while the sub-device nodes are still registered. subdev_open() dereferences both pointers. v4l2_open() has checked that the node is registered, but it drops videodev_lock before calling fops->open(), so the whole unregistration can run in between. Reordering v4l2_device_unregister_subdev() would not help in the second case, and checking the two pointers in subdev_open() would only narrow the window.
Neither pointer is needed in subdev_open(). Both record that the sub-device is registered, and are cleared on purpose when it goes away, while open() runs on behalf of the device node. The node has its own pointer to the same v4l2_device, vdev->v4l2_dev, which is set when the node is registered and never cleared, and the V4L2 core already relies on it for as long as the node exists: v4l2_release() dereferences it on every close. The sub-device's entity is registered with vdev->v4l2_dev->mdev, so that is also the media device to take the module reference through.
That reference goes through mdev->dev->driver, which the driver core clears once the media device's own driver has been unbound, as when syzbot unbinds vimc. Read it once with READ_ONCE(), as dev_driver_string() does, and fail the open with -ENODEV if it is gone:
unbinding does not unload the module, so the pointer read is either still valid or NULL.
With this, subdev_open() no longer reads a pointer that the V4L2 core or the driver core clears when a driver is unbound. It can still run on a sub-device that has just been unregistered, exactly like a file handle opened an instant earlier. The lifetime of the sub-device and of the media device when a driver goes away with file handles open, and module unloading, are a separate, known limitation that this does not address.
Reproduced on a CHUWI Hi10 X1 (Alder Lake-N, IPU6) running 6.12.86, at cycle 7 of a loop unbinding and rebinding a sensor while four processes opened every /dev/v4l-subdev*. syzbot hit the second dereference on the same line by unbinding vimc, and so does a test in QEMU with KASAN that opens vimc's sub-device nodes while vimc is unbound and rebound; with this patch the same test shows no oops, KASAN report or warning.
--- drivers/media/v4l2-core/v4l2-subdev.c | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-)
drivers/media/v4l2-core/v4l2-subdev.c +10 -2
@@ -96,6 +96,7 @@ static int subdev_open(struct file *file) {struct video_device *vdev =video_devdata (file);struct v4l2_subdev *sd =vdev_to_v4l2_subdev (vdev); +struct media_device *mdev = vdev->v4l2_dev->mdev;struct v4l2_subdev_fh *subdev_fh;int ret;
@@ -112,10 +113,17 @@ static int subdev_open(struct file *file)v4l2_fh_init (&subdev_fh->vfh, vdev);v4l2_fh_add (&subdev_fh->vfh, file); -if (sd->v4l2_dev->mdev && sd->entity.graph_obj.mdev->dev) { +if (mdev && mdev->dev) { +struct device_driver *drv =READ_ONCE (mdev->dev->driver);struct module *owner; - owner = sd->entity.graph_obj.mdev->dev->driver->owner; +/* The media device's driver has been unbound meanwhile. */ +if (!drv) { + ret = -ENODEV ; + goto err; + } + + owner = drv->owner;if (!try_module_get (owner)) { ret = -EBUSY ; goto err;
Signature · 1 line
2.47.3
Re: [PATCH v3 1/2] media: v4l2-subdev: Fix NULL pointer dereference in subdev_open()
Quoted 44 lines · Nicola Fiorillo
Unbinding a driver while something opens a
/dev/v4l-subdevNnode oopses the kernel:BUG: kernel NULL pointer dereference, address: 0000000000000008 RIP: 0010:subdev_open+0x8a/0x190 [videodev] Call Trace: v4l2_open+0xa9/0x100 [videodev] chrdev_open+0xb2/0x230 do_dentry_open+0x14c/0x440 vfs_open+0x2e/0xe0 path_openat+0x82e/0x12d0 do_filp_open+0xc4/0x170 do_sys_openat2+0xae/0xe0 __x64_sys_openat+0x55/0xa0
v4l2_device_unregister_subdev()clearssd->v4l2_dev, then unregisters the media entity, which clearssd->entity.graph_obj.mdev, and only then unregisters the device node. Drivers that callmedia_device_unregister()beforev4l2_device_unregister(), as vimc does, clearsd->entity.graph_obj.mdeveven earlier, while the sub-device nodes are still registered.subdev_open()dereferences both pointers.v4l2_open()has checked that the node is registered, but it dropsvideodev_lockbefore callingfops->open(), so the whole unregistration can run in between. Reorderingv4l2_device_unregister_subdev()would not help in the second case, and checking the two pointers insubdev_open()would only narrow the window.Neither pointer is needed in
subdev_open(). Both record that the sub-device is registered, and are cleared on purpose when it goes away, whileopen()runs on behalf of the device node. The node has its own pointer to the samev4l2_device,vdev->v4l2_dev, which is set when the node is registered and never cleared, and the V4L2 core already relies on it for as long as the node exists:v4l2_release()dereferences it on every close. The sub-device's entity is registered withvdev->v4l2_dev->mdev, so that is also the media device to take the module reference through.That reference goes through
mdev->dev->driver, which the driver core clears once the media device's own driver has been unbound, as when syzbot unbinds vimc. Read it once withREAD_ONCE(), asdev_driver_string()does, and fail the open with-ENODEVif it is gone:
unbinding does not unload the module, so the pointer read is either still valid or NULL.
Quoted 8 lines
With this,
subdev_open()no longer reads a pointer that the V4L2 core or the driver core clears when a driver is unbound. It can still run on a sub-device that has just been unregistered, exactly like a file handle opened an instant earlier. The lifetime of the sub-device and of the media device when a driver goes away with file handles open, and module unloading, are a separate, known limitation that this does not address.Reproduced on a CHUWI Hi10 X1 (Alder Lake-N, IPU6) running 6.12.86, at
We don't develop or test patcheson 6.12.86.
Quoted 35 lines
cycle 7 of a loop unbinding and rebinding a sensor while four processes opened every
/dev/v4l-subdev*. syzbot hit the second dereference on the same line by unbinding vimc, and so does a test in QEMU with KASAN that opens vimc's sub-device nodes while vimc is unbound and rebound; with this patch the same test shows no oops, KASAN report or warning.Reported-by [email protected]Fixes 61f5db549dde ("[media] v4l: Make v4l2_subdev inherit from media_entity")Assisted-by LLMSigned-off-by Nicola Fiorillo <[email protected]>--- drivers/media/v4l2-core/v4l2-subdev.c | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-)drivers/media/v4l2-core/v4l2-subdev.c +3
-1@@ -96,6 +96,7 @@ static int subdev_open(struct file *file) {struct video_device *vdev =video_devdata (file);struct v4l2_subdev *sd =vdev_to_v4l2_subdev (vdev); +struct media_device *mdev = vdev->v4l2_dev->mdev;struct v4l2_subdev_fh *subdev_fh;int ret;@@ -112,10 +113,17 @@ static int subdev_open(struct file *file)v4l2_fh_init (&subdev_fh->vfh, vdev);v4l2_fh_add (&subdev_fh->vfh, file); -if (sd->v4l2_dev->mdev && sd->entity.graph_obj.mdev->dev) { +if (mdev && mdev->dev) { +struct device_driver *drv =READ_ONCE (mdev->dev->driver);
drivers/media/v4l2-core/v4l2-subdev.c static int subdev_open(struct file *file) · line 126
Quoted 10 lines
drivers/media/v4l2-core/v4l2-subdev.c +7
-1struct module *owner; - owner = sd->entity.graph_obj.mdev->dev->driver->owner; +/* The media device's driver has been unbound meanwhile. */ +if (!drv) { + ret = -ENODEV ; + goto err; + } + + owner = drv->owner;
All of this is a hack that may reduce a race window but it doesn't fix the problem.
Quoted 3 lines
drivers/media/v4l2-core/v4l2-subdev.c +0
-0if (!try_module_get (owner)) { ret = -EBUSY ; goto err;
Signature · 3 lines
Regards, Laurent Pinchart
Re: [PATCH v3 1/2] media: v4l2-subdev: Fix NULL pointer dereference in subdev_open()
Hi Laurent,
Thank you for looking at it.
On 6.12.86: agreed, that line does not belong in the commit message. The runs that matter are the ones on media next described in the cover letter, and those are what the commit message should have referred to.
On the fix: agreed as well, it does not fix the problem. subdev_open() now goes through vdev->v4l2_dev->mdev, which is only valid as long as the driver keeps the v4l2_device and the media_device alive. vimc does, through its v4l2_device release callback, but ipu6-isys embeds both in struct ipu6_isys, allocated with devm_kzalloc(), so they are freed when the driver is unbound.
I checked this on the IPU6 tablet (media next 9cfc1aca0781 plus the sensor drivers mentioned in the cover letter, without this series):
with the media device, a video node and a CSI-2 sub-device node of isys held open, I unbound isys, then issued one ioctl on each node and closed them. KASAN reports use-after-free in v4l2_ioctl() on the video node, then on close in v4l2_release(), v4l2_prio_close() and v4l2_device_release() on the struct ipu6_isys allocated in isys_probe() and freed by devres at unbind, and more in subdev_close() and __vb2_queue_free().
Patch 2/2 rests on the same vdev->v4l2_dev assumption, so I am withdrawing the whole series rather than keeping half of it. The underlying issue is the lifetime of the objects in the driver, and that needs a proper fix along the lines of what Hans did for em28xx, not another check in the file operations.
Thanks,
Nicola
Re: [PATCH v3 1/2] media: v4l2-subdev: Fix NULL pointer dereference in subdev_open()
Quoted 5 lines · Nicola Fiorillo
Hi Laurent,
Thank you for looking at it.
On 6.12.86: agreed, that line does not belong in the commit message.
Why was it there in the first place ? It gives me very little confidence on the quality of your contribution.
Quoted 25 lines
The runs that matter are the ones on media next described in the cover letter, and those are what the commit message should have referred to.
On the fix: agreed as well, it does not fix the problem.
subdev_open()now goes throughvdev->v4l2_dev->mdev, which is only valid as long as the driver keeps thev4l2_deviceand themedia_devicealive. vimc does, through itsv4l2_devicerelease callback, but ipu6-isys embeds both in structipu6_isys, allocated withdevm_kzalloc(), so they are freed when the driver is unbound.I checked this on the IPU6 tablet (media next
9cfc1aca0781plus the sensor drivers mentioned in the cover letter, without this series):
with the media device, a video node and a CSI-2 sub-device node of isys held open, I unbound isys, then issued one ioctl on each node and closed them. KASAN reports use-after-free inv4l2_ioctl()on the video node, then on close inv4l2_release(),v4l2_prio_close()andv4l2_device_release()on the structipu6_isysallocated inisys_probe()and freed by devres at unbind, and more insubdev_close()and__vb2_queue_free().Patch 2/2 rests on the same
vdev->v4l2_devassumption, so I am withdrawing the whole series rather than keeping half of it. The underlying issue is the lifetime of the objects in the driver, and that needs a proper fix along the lines of what Hans did for em28xx, not another check in the file operations.
Signature · 3 lines
Regards, Laurent Pinchart
Re: [PATCH v3 1/2] media: v4l2-subdev: Fix NULL pointer dereference in subdev_open()
Hi Laurent,
Quoted 5 lines · Laurent Pinchart
Quoted 1 line · Nicola Fiorillo
On 6.12.86: agreed, that line does not belong in the commit message.
Why was it there in the first place ? It gives me very little confidence on the quality of your contribution.
Sorry, that was my mistake. The sentence came from v1. I first hit the oops on the 6.12.86 kernel I use every day on the tablet. When I repeated the tests on media next, I updated the cover letter but not the commit message, and the sentence went unchanged into v2 and v3. I did not catch it when I reread the patch.
The GC5035 and GC8034 drivers I sent today were developed and tested only on media next (8e26d4c20), on the tablet.
English is not my first language: I write in Italian and translate with the help of an LLM.
Nicola
PATCH v3 2/2 media: v4l2-subdev: Fix NULL pointer dereference in the EXT_CTRLS ioctls
In reply to Nicola FiorilloVIDIOC_G_EXT_CTRLS, VIDIOC_S_EXT_CTRLS and VIDIOC_TRY_EXT_CTRLS on a sub-device node pass sd->v4l2_dev->mdev to the control framework. v4l2_device_unregister_subdev() clears sd->v4l2_dev before it unregisters the device node, and nothing serialises the video_is_registered() checks in v4l2_ioctl() and subdev_do_ioctl_lock() against it: sub-device nodes have no vdev->lock, and unregistration would not take it anyway. An EXT_CTRLS ioctl on a file handle opened before a driver is unbound can therefore dereference NULL:
KASAN: null-ptr-deref in range [0x0000000000000008-0x000000000000000f] RIP: 0010:subdev_do_ioctl+0x1926/0x2550 Call Trace: subdev_do_ioctl_lock+0x246/0x5c0 video_usercopy+0x47b/0xcc0 v4l2_ioctl+0x187/0x1f0 __x64_sys_ioctl+0x134/0x1c0
Use vdev->v4l2_dev->mdev instead, as the previous patch does in subdev_open(), and as the video node ioctls already do.
Found by reading the code, then reproduced in QEMU with KASAN: with the previous patch applied, threads looping on VIDIOC_G_EXT_CTRLS on vimc's sub-device nodes while vimc is unbound and rebound hit the oops above; with this patch the same test shows no oops, KASAN report or warning.
--- drivers/media/v4l2-core/v4l2-subdev.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/media/v4l2-core/v4l2-subdev.c +3 -3
@@ -775,19 +775,19 @@ static long subdev_do_ioctl(struct file *file, unsigned int cmd, void *arg,if (!vfh->ctrl_handler)return -ENOTTY ;return v4l2_g_ext_ctrls (vfh->ctrl_handler, - vdev, sd->v4l2_dev->mdev, arg); + vdev, vdev->v4l2_dev->mdev, arg);case VIDIOC_S_EXT_CTRLS :if (!vfh->ctrl_handler)return -ENOTTY ;return v4l2_s_ext_ctrls (vfh, vfh->ctrl_handler, - vdev, sd->v4l2_dev->mdev, arg); + vdev, vdev->v4l2_dev->mdev, arg);case VIDIOC_TRY_EXT_CTRLS :if (!vfh->ctrl_handler)return -ENOTTY ;return v4l2_try_ext_ctrls (vfh->ctrl_handler, - vdev, sd->v4l2_dev->mdev, arg); + vdev, vdev->v4l2_dev->mdev, arg);case VIDIOC_DQEVENT :if (!(sd->flags &V4L2_SUBDEV_FL_HAS_EVENTS ))
Signature · 1 line
2.47.3
Re: [PATCH v3 2/2] media: v4l2-subdev: Fix NULL pointer dereference in the EXT_CTRLS ioctls
Hi Nicola,
Quoted 8 lines · Nicola Fiorillo
VIDIOC_G_EXT_CTRLS,VIDIOC_S_EXT_CTRLSandVIDIOC_TRY_EXT_CTRLSon a sub-device node passsd->v4l2_dev->mdevto the control framework.v4l2_device_unregister_subdev()clearssd->v4l2_devbefore it unregisters the device node, and nothing serialises thevideo_is_registered()checks inv4l2_ioctl()andsubdev_do_ioctl_lock()against it: sub-device nodes have novdev->lock, and unregistration would not take it anyway. AnEXT_CTRLSioctl on a file handle opened before a driver is unbound can therefore dereference NULL:
Unregistering a sub-device node isn't doable safely currently. Addressing this properly requires much more than this patch does, and also I'm afraid this patch isn't part of properly addressing this either.
Signature · 3 lines
Kind regards, Sakari Ailus
Re: [PATCH v3 2/2] media: v4l2-subdev: Fix NULL pointer dereference in the EXT_CTRLS ioctls
Hi Sakari,
Quoted 3 lines · Sakari Ailus
Unregistering a sub-device node isn't doable safely currently. Addressing this properly requires much more than this patch does, and also I'm afraid this patch isn't part of properly addressing this either.
Understood. As I wrote to Laurent earlier today, I have withdrawn the whole series: both patches only narrow the window, and the real issue is the lifetime of the objects, not something more checks in the file operations can fix.
Thank you and Laurent for taking the time to review it.
Thanks,
Nicola