media: uvcvideo: Fix race conditions

The uvcvideo code has no lock protection against USB disconnects while video operations are ongoing. This has resulted in random error reports, typically pointing to a crash in usb_ifnum_to_if(), called from usb_hcd_alloc_bandwidth(). A typical traceback is as follows.

usb 1-4: USB disconnect, device number 3

BUG: unable to handle kernel NULL pointer dereference at 0000000000000000
PGD 0 P4D 0
Oops: 0000 [#1] PREEMPT SMP PTI
CPU: 0 PID: 5633 Comm: V4L2CaptureThre Not tainted 4.19.113-08536-g5d29ca36db06 #1
Hardware name: GOOGLE Edgar, BIOS Google_Edgar.7287.167.156 03/25/2019
RIP: 0010:usb_ifnum_to_if+0x29/0x40
Code: <...>
RSP: 0018:ffffa46f42a47a80 EFLAGS: 00010246
RAX: 0000000000000000 RBX: 0000000000000000 RCX: ffff904a396c9000
RDX: ffff904a39641320 RSI: 0000000000000001 RDI: 0000000000000000
RBP: ffffa46f42a47a80 R08: 0000000000000002 R09: 0000000000000000
R10: 0000000000009975 R11: 0000000000000009 R12: 0000000000000000
R13: ffff904a396b3800 R14: ffff904a39e88000 R15: 0000000000000000
FS: 00007f396448e700(0000) GS:ffff904a3ba00000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 000000016cb46000 CR4: 00000000001006f0
Call Trace:
 usb_hcd_alloc_bandwidth+0x1ee/0x30f
 usb_set_interface+0x1a3/0x2b7
 uvc_video_start_transfer+0x29b/0x4b8 [uvcvideo]
 uvc_video_start_streaming+0x91/0xdd [uvcvideo]
 uvc_start_streaming+0x28/0x5d [uvcvideo]
 vb2_start_streaming+0x61/0x143 [videobuf2_common]
 vb2_core_streamon+0xf7/0x10f [videobuf2_common]
 uvc_queue_streamon+0x2e/0x41 [uvcvideo]
 uvc_ioctl_streamon+0x42/0x5c [uvcvideo]
 __video_do_ioctl+0x33d/0x42a
 video_usercopy+0x34e/0x5ff
 ? video_ioctl2+0x16/0x16
 v4l2_ioctl+0x46/0x53
 do_vfs_ioctl+0x50a/0x76f
 ksys_ioctl+0x58/0x83
 __x64_sys_ioctl+0x1a/0x1e
 do_syscall_64+0x54/0xde

While this is problem rarely observed in the field, it is relatively easy to reproduce by adding msleep() calls into the code.

I don't presume to claim that I found every issue, but this patch series should fix at least the major problems.

The patch series was tested exensively on a Chromebook running chromeos-4.19 and on a Linux system running a v5.8.y based kernel.

----------------------------------------------------------------
Guenter Roeck (5):
media: uvcvideo: Cancel async worker earlier
media: uvcvideo: Lock video streams and queues while unregistering
media: uvcvideo: Release stream queue when unregistering video device
media: uvcvideo: Protect uvc queue file operations against disconnect
media: uvcvideo: In uvc_v4l2_open, check if video device is registered

 drivers/media/usb/uvc/uvc_ctrl.c   | 11 ++++++----
 drivers/media/usb/uvc/uvc_driver.c | 12 ++++++++++
 drivers/media/usb/uvc/uvc_queue.c  | 32 +++++++++++++++++++++++++--
 drivers/media/usb/uvc/uvc_v4l2.c   | 45 ++++++++++++++++++++++++++++++++++++--
 drivers/media/usb/uvc/uvcvideo.h   |  1 +
 5 files changed, 93 insertions(+), 8 deletions(-)

So far the asynchronous control worker was canceled only in uvc_ctrl_cleanup_device. This is much later than the call to uvc_disconnect. However, after the call to uvc_disconnect returns, there must be no more USB activity. This can result in all kinds of problems in the USB code. One observed example:

URB ffff993e83d0bc00 submitted while active

WARNING: CPU: 0 PID: 4046 at drivers/usb/core/urb.c:364 usb_submit_urb+0x4ba/0x55e
Modules linked in: <...>
CPU: 0 PID: 4046 Comm: kworker/0:35 Not tainted 4.19.139 #18
Hardware name: Google Phaser/Phaser, BIOS Google_Phaser.10952.0.0 08/09/2018
Workqueue: events uvc_ctrl_status_event_work [uvcvideo]
RIP: 0010:usb_submit_urb+0x4ba/0x55e
Code: <...>
RSP: 0018:ffffb08d471ebde8 EFLAGS: 00010246
RAX: a6da85d923ea5d00 RBX: ffff993e71985928 RCX: 0000000000000000
RDX: ffff993f37a1de90 RSI: ffff993f37a153d0 RDI: ffff993f37a153d0
RBP: ffffb08d471ebe28 R08: 000000000000003b R09: 001424bf85822e96
R10: 0000001000000000 R11: ffffffff975a4398 R12: ffff993e83d0b000
R13: ffff993e83d0bc00 R14: 0000000000000000 R15: 00000000fffffff0
FS:  0000000000000000(0000) GS:ffff993f37a00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000ec9c0000 CR3: 000000025b160000 CR4: 0000000000340ef0
Call Trace:
 uvc_ctrl_status_event_work+0xd6/0x107 [uvcvideo]
 process_one_work+0x19b/0x4c5
 worker_thread+0x10d/0x286
 kthread+0x138/0x140
 ? process_one_work+0x4c5/0x4c5
 ? kthread_associate_blkcg+0xc1/0xc1
 ret_from_fork+0x1f/0x40

Introduce new function uvc_ctrl_stop_device() to cancel the worker and call it from uvc_unregister_video() to solve the problem.

---
 drivers/media/usb/uvc/uvc_ctrl.c   | 11 +++++++----
 drivers/media/usb/uvc/uvc_driver.c |  1 +
 drivers/media/usb/uvc/uvcvideo.h   |  1 +
 3 files changed, 9 insertions(+), 4 deletions(-)

index e399b9fad757..130c56e0063d 100644
@@ -2340,14 +2340,17 @@ static void uvc_ctrl_cleanup_mappings(struct uvc_device *dev,
 	}
 }
 
-void uvc_ctrl_cleanup_device(struct uvc_device *dev)
+void uvc_ctrl_stop_device(struct uvc_device *dev)
 {
-	struct uvc_entity *entity;
-	unsigned int i;
-
 	/* Can be uninitialized if we are aborting on probe error. */
 	if (dev->async_ctrl.work.func)
 		cancel_work_sync(&dev->async_ctrl.work);
+}
+
+void uvc_ctrl_cleanup_device(struct uvc_device *dev)
+{
+	struct uvc_entity *entity;
+	unsigned int i;
 
 	/* Free controls and control mappings for all entities. */
 	list_for_each_entry(entity, &dev->entities, list) {

index 431d86e1c94b..bfba67a69185 100644
@@ -1952,6 +1952,7 @@ static void uvc_unregister_video(struct uvc_device *dev)
 	}
 
 	uvc_status_unregister(dev);
+	uvc_ctrl_stop_device(dev);
 
 	if (dev->vdev.dev)
 		v4l2_device_unregister(&dev->vdev);

index 6ab972c643e3..543afcd9fd26 100644
@@ -830,6 +830,7 @@ int uvc_query_v4l2_menu(struct uvc_video_chain *chain,
 int uvc_ctrl_add_mapping(struct uvc_video_chain *chain,
 			 const struct uvc_control_mapping *mapping);
 int uvc_ctrl_init_device(struct uvc_device *dev);
+void uvc_ctrl_stop_device(struct uvc_device *dev);
 void uvc_ctrl_cleanup_device(struct uvc_device *dev);
 int uvc_ctrl_restore_values(struct uvc_device *dev);
 bool uvc_ctrl_status_event(struct urb *urb, struct uvc_video_chain *chain,

2.17.1

The call to uvc_disconnect() is not protected by any mutex. This means it can and will be called while other accesses to the video device are in progress. This can cause all kinds of race conditions, including crashes such as the following.

usb 1-4: USB disconnect, device number 3

BUG: unable to handle kernel NULL pointer dereference at 0000000000000000
PGD 0 P4D 0
Oops: 0000 [#1] PREEMPT SMP PTI
CPU: 0 PID: 5633 Comm: V4L2CaptureThre Not tainted 4.19.113-08536-g5d29ca36db06 #1
Hardware name: GOOGLE Edgar, BIOS Google_Edgar.7287.167.156 03/25/2019
RIP: 0010:usb_ifnum_to_if+0x29/0x40
Code: <...>
RSP: 0018:ffffa46f42a47a80 EFLAGS: 00010246
RAX: 0000000000000000 RBX: 0000000000000000 RCX: ffff904a396c9000
RDX: ffff904a39641320 RSI: 0000000000000001 RDI: 0000000000000000
RBP: ffffa46f42a47a80 R08: 0000000000000002 R09: 0000000000000000
R10: 0000000000009975 R11: 0000000000000009 R12: 0000000000000000
R13: ffff904a396b3800 R14: ffff904a39e88000 R15: 0000000000000000
FS: 00007f396448e700(0000) GS:ffff904a3ba00000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 000000016cb46000 CR4: 00000000001006f0
Call Trace:
 usb_hcd_alloc_bandwidth+0x1ee/0x30f
 usb_set_interface+0x1a3/0x2b7
 uvc_video_start_transfer+0x29b/0x4b8 [uvcvideo]
 uvc_video_start_streaming+0x91/0xdd [uvcvideo]
 uvc_start_streaming+0x28/0x5d [uvcvideo]
 vb2_start_streaming+0x61/0x143 [videobuf2_common]
 vb2_core_streamon+0xf7/0x10f [videobuf2_common]
 uvc_queue_streamon+0x2e/0x41 [uvcvideo]
 uvc_ioctl_streamon+0x42/0x5c [uvcvideo]
 __video_do_ioctl+0x33d/0x42a
 video_usercopy+0x34e/0x5ff
 ? video_ioctl2+0x16/0x16
 v4l2_ioctl+0x46/0x53
 do_vfs_ioctl+0x50a/0x76f
 ksys_ioctl+0x58/0x83
 __x64_sys_ioctl+0x1a/0x1e
 do_syscall_64+0x54/0xde

usb_set_interface() should not be called after the USB device has been unregistered. However, in the above case the disconnect happened after v4l2_ioctl() was called, but before the call to usb_ifnum_to_if().

Acquire various mutexes in uvc_unregister_video() to fix the majority (maybe all) of the observed race conditions.

The uvc_device lock prevents races against suspend and resume calls and the poll function.

The uvc_streaming lock prevents races against stream related functions; for the most part, those are ioctls. This lock also requires other functions using this lock to check if a video device is still registered after acquiring it. For example, it was observed that the video device was already unregistered by the time the stream lock was acquired in uvc_ioctl_streamon().

The uvc_queue lock prevents races against queue functions, Most of those are already protected by the uvc_streaming lock, but some are called directly. This is done as added protection; an actual race was not (yet) observed.

---
 drivers/media/usb/uvc/uvc_driver.c |  9 +++++++
 drivers/media/usb/uvc/uvc_v4l2.c   | 39 +++++++++++++++++++++++++++++-
 2 files changed, 47 insertions(+), 1 deletion(-)

index bfba67a69185..a5808305f1e3 100644
@@ -1941,14 +1941,22 @@ static void uvc_unregister_video(struct uvc_device *dev)
 {
 	struct uvc_streaming *stream;
 
+	mutex_lock(&dev->lock);
+
 	list_for_each_entry(stream, &dev->streams, list) {
 		if (!video_is_registered(&stream->vdev))
 			continue;
 
+		mutex_lock(&stream->mutex);
+		mutex_lock(&stream->queue.mutex);
+
 		video_unregister_device(&stream->vdev);
 		video_unregister_device(&stream->meta.vdev);
 
 		uvc_debugfs_cleanup_stream(stream);
+
+		mutex_unlock(&stream->queue.mutex);
+		mutex_unlock(&stream->mutex);
 	}
 
 	uvc_status_unregister(dev);

@@ -1960,6 +1968,7 @@ static void uvc_unregister_video(struct uvc_device *dev)
 	if (media_devnode_is_registered(dev->mdev.devnode))
 		media_device_unregister(&dev->mdev);
 #endif
+	mutex_unlock(&dev->lock);
 }
 
 int uvc_register_video_device(struct uvc_device *dev,

index 0335e69b70ab..7e5e583dae5e 100644
@@ -237,6 +237,12 @@ static int uvc_v4l2_try_format(struct uvc_streaming *stream,
 	 * the Windows driver).
 	 */
 	mutex_lock(&stream->mutex);
+	if (!video_is_registered(&stream->vdev)) {
+		mutex_unlock(&stream->mutex);
+		ret = -ENODEV;
+		goto done;
+	}
+
 	if (stream->dev->quirks & UVC_QUIRK_PROBE_EXTRAFIELDS)
 		probe->dwMaxVideoFrameSize =
 			stream->ctrl.dwMaxVideoFrameSize;

@@ -274,6 +280,12 @@ static int uvc_v4l2_get_format(struct uvc_streaming *stream,
 		return -EINVAL;
 
 	mutex_lock(&stream->mutex);
+
+	if (!video_is_registered(&stream->vdev)) {
+		ret = -ENODEV;
+		goto done;
+	}
+
 	format = stream->cur_format;
 	frame = stream->cur_frame;
 

@@ -312,6 +324,11 @@ static int uvc_v4l2_set_format(struct uvc_streaming *stream,
 
 	mutex_lock(&stream->mutex);
 
+	if (!video_is_registered(&stream->vdev)) {
+		ret = -ENODEV;
+		goto done;
+	}
+
 	if (uvc_queue_allocated(&stream->queue)) {
 		ret = -EBUSY;
 		goto done;

@@ -387,6 +404,11 @@ static int uvc_v4l2_set_streamparm(struct uvc_streaming *stream,
 
 	mutex_lock(&stream->mutex);
 
+	if (!video_is_registered(&stream->vdev)) {
+		mutex_unlock(&stream->mutex);
+		return -ENODEV;
+	}
+
 	if (uvc_queue_streaming(&stream->queue)) {
 		mutex_unlock(&stream->mutex);
 		return -EBUSY;

@@ -713,6 +735,10 @@ static int uvc_ioctl_reqbufs(struct file *file, void *fh,
 		return ret;
 
 	mutex_lock(&stream->mutex);
+	if (!video_is_registered(&stream->vdev)) {
+		mutex_unlock(&stream->mutex);
+		return -ENODEV;
+	}
 	ret = uvc_request_buffers(&stream->queue, rb);
 	mutex_unlock(&stream->mutex);
 	if (ret < 0)

@@ -797,7 +823,12 @@ static int uvc_ioctl_streamon(struct file *file, void *fh,
 		return -EBUSY;
 
 	mutex_lock(&stream->mutex);
+	if (!video_is_registered(&stream->vdev)) {
+		ret = -ENODEV;
+		goto unlock;
+	}
 	ret = uvc_queue_streamon(&stream->queue, type);
+unlock:
 	mutex_unlock(&stream->mutex);
 
 	return ret;

@@ -808,15 +839,21 @@ static int uvc_ioctl_streamoff(struct file *file, void *fh,
 {
 	struct uvc_fh *handle = fh;
 	struct uvc_streaming *stream = handle->stream;
+	int ret = 0;
 
 	if (!uvc_has_privileges(handle))
 		return -EBUSY;
 
 	mutex_lock(&stream->mutex);
+	if (!video_is_registered(&stream->vdev)) {
+		ret = -ENODEV;
+		goto unlock;
+	}
 	uvc_queue_streamoff(&stream->queue, type);
+unlock:
 	mutex_unlock(&stream->mutex);
 
-	return 0;
+	return ret;
 }
 
 static int uvc_ioctl_enum_input(struct file *file, void *fh,

2.17.1

The following traceback was observed when stress testing the uvcvideo driver.

config->interface[0] is NULL

WARNING: CPU: 0 PID: 2757 at drivers/usb/core/usb.c:285 usb_ifnum_to_if+0x81/0x85
^^^ added log message and WARN() to prevent crash
Modules linked in: <...>
CPU: 0 PID: 2757 Comm: inotify_reader Not tainted 4.19.139 #20
Hardware name: Google Phaser/Phaser, BIOS Google_Phaser.10952.0.0 08/09/2018
RIP: 0010:usb_ifnum_to_if+0x81/0x85
Code: <...>
RSP: 0018:ffff9ee20141fa58 EFLAGS: 00010246
RAX: 438a0e5a525f1800 RBX: 0000000000000000 RCX: 0000000000000000
RDX: ffff975477a1de90 RSI: ffff975477a153d0 RDI: ffff975477a153d0
RBP: ffff9ee20141fa70 R08: 000000000000002c R09: 002daec189138d78
R10: 0000001000000000 R11: ffffffffa7da42e6 R12: ffff975459594800
R13: 0000000000000001 R14: 0000000000000001 R15: ffff975465376400
FS:  00007ddebffd6700(0000) GS:ffff975477a00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000012c83000 CR3: 00000001bbaf8000 CR4: 0000000000340ef0
Call Trace:
 usb_set_interface+0x3b/0x2f3
 uvc_video_stop_streaming+0x38/0x81 [uvcvideo]
 uvc_stop_streaming+0x21/0x49 [uvcvideo]
 __vb2_queue_cancel+0x33/0x249 [videobuf2_common]
 vb2_core_queue_release+0x1c/0x45 [videobuf2_common]
 uvc_queue_release+0x26/0x32 [uvcvideo]
 uvc_v4l2_release+0x41/0xdd [uvcvideo]
 v4l2_release+0x99/0xed
 __fput+0xf0/0x1ea
 task_work_run+0x7f/0xa7
 do_exit+0x1d1/0x6eb
 do_group_exit+0x9d/0xac
 get_signal+0x135/0x482
 do_signal+0x4a/0x63c
 exit_to_usermode_loop+0x86/0xa5
 do_syscall_64+0x171/0x269
 ? __do_page_fault+0x272/0x474
 entry_SYSCALL_64_after_hwframe+0x44/0xa9
RIP: 0033:0x7ddec156dc26
Code: Bad RIP value.
RSP: 002b:00007ddebffd5c10 EFLAGS: 00000293 ORIG_RAX: 0000000000000017
RAX: fffffffffffffdfe RBX: 000000000000000a RCX: 00007ddec156dc26
RDX: 0000000000000000 RSI: 00007ddebffd5d28 RDI: 000000000000000a
RBP: 00007ddebffd5c50 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000293 R12: 00007ddebffd5d28
R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000

When this was observed, a USB disconnect was in progress, uvc_disconnect() had returned, but usb_disable_device() was still running. Drivers are not supposed to call any USB functions after the driver disconnect function has been called. The uvcvideo driver does not follow that convention and tries to write to the disconnected USB interface anyway. This results in a variety of race conditions.

To solve this specific problem, release the uvc queue from uvc_unregister_video() after the associated video devices have been unregistered. Since the function already holds the uvc queue mutex, bypass uvc_queue_release() and call vb2_queue_release() directly.

---
 drivers/media/usb/uvc/uvc_driver.c | 2 ++
 1 file changed, 2 insertions(+)

index a5808305f1e3..27b72806b9b9 100644
@@ -1955,6 +1955,8 @@ static void uvc_unregister_video(struct uvc_device *dev)
 
 		uvc_debugfs_cleanup_stream(stream);
 
+		vb2_queue_release(&stream->queue.queue);
+
 		mutex_unlock(&stream->queue.mutex);
 		mutex_unlock(&stream->mutex);
 	}

2.17.1

uvc queue file operations have no mutex protection against USB disconnect. This is questionable at least for the poll operation, which has to wait for the uvc queue mutex. By the time that mutex has been acquired, is it possible that the video device has been unregistered.

Protect all file operations by using the queue mutex to avoid possible race conditions. After acquiring the mutex, check if the video device is still registered, and bail out if not.

---
 drivers/media/usb/uvc/uvc_queue.c | 32 +++++++++++++++++++++++++++++--
 1 file changed, 30 insertions(+), 2 deletions(-)

index cd60c6c1749e..c7964177a99d 100644
@@ -358,24 +358,52 @@ int uvc_queue_streamoff(struct uvc_video_queue *queue, enum v4l2_buf_type type)
 
 int uvc_queue_mmap(struct uvc_video_queue *queue, struct vm_area_struct *vma)
 {
-	return vb2_mmap(&queue->queue, vma);
+	struct uvc_streaming *stream = uvc_queue_to_stream(queue);
+	int ret;
+
+	mutex_lock(&queue->mutex);
+	if (!video_is_registered(&stream->vdev)) {
+		ret = -ENODEV;
+		goto unlock;
+	}
+	ret = vb2_mmap(&queue->queue, vma);
+unlock:
+	mutex_unlock(&queue->mutex);
+	return ret;
 }
 
 #ifndef CONFIG_MMU
 unsigned long uvc_queue_get_unmapped_area(struct uvc_video_queue *queue,
 		unsigned long pgoff)
 {
-	return vb2_get_unmapped_area(&queue->queue, 0, 0, pgoff, 0);
+	struct uvc_streaming *stream = uvc_queue_to_stream(queue);
+	unsigned long ret;
+
+	mutex_lock(&queue->mutex);
+	if (!video_is_registered(&stream->vdev)) {
+		ret = -ENODEV;
+		goto unlock;
+	}
+	ret = vb2_get_unmapped_area(&queue->queue, 0, 0, pgoff, 0);
+unlock:
+	mutex_unlock(&queue->mutex);
+	return ret;
 }
 #endif
 
 __poll_t uvc_queue_poll(struct uvc_video_queue *queue, struct file *file,
 			    poll_table *wait)
 {
+	struct uvc_streaming *stream = uvc_queue_to_stream(queue);
 	__poll_t ret;
 
 	mutex_lock(&queue->mutex);
+	if (!video_is_registered(&stream->vdev)) {
+		ret = -ENODEV;
+		goto unlock;
+	}
 	ret = vb2_poll(&queue->queue, file, wait);
+unlock:
 	mutex_unlock(&queue->mutex);
 
 	return ret;

2.17.1

Looks like this has to return EPOLLERR. I'll make that change in v2.

Guenter

uvc_v4l2_open() acquires the uvc device mutex. After doing so, it does not check if the video device is still registered. This may result in race conditions and can result in an attempt to submit an urb to a disconnected USB interface (from uvc_v4l2_open).

The problem was observed after adding a call to msleep() just before acquiring the mutex and disconnecting the camera during the sleep.

Check if the video device is still registered after acquiring the mutex to avoid the problem. In the release function, only call uvc_status_stop() if the video device is still registered. If the video device has been unregistered, the urb associated with uvc status has already been killed in uvc_status_unregister(). Trying to kill it again won't do any good and might have unexpected side effects.

---
 drivers/media/usb/uvc/uvc_v4l2.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

index 7e5e583dae5e..1f7d454557b9 100644
@@ -548,6 +548,10 @@ static int uvc_v4l2_open(struct file *file)
 	}
 
 	mutex_lock(&stream->dev->lock);
+	if (!video_is_registered(&stream->vdev)) {
+		mutex_unlock(&stream->dev->lock);
+		return -ENODEV;
+	}
 	if (stream->dev->users == 0) {
 		ret = uvc_status_start(stream->dev, GFP_KERNEL);
 		if (ret < 0) {

@@ -590,7 +594,7 @@ static int uvc_v4l2_release(struct file *file)
 	file->private_data = NULL;
 
 	mutex_lock(&stream->dev->lock);
-	if (--stream->dev->users == 0)
+	if (--stream->dev->users == 0 && video_is_registered(&stream->vdev))
 		uvc_status_stop(stream->dev);
 	mutex_unlock(&stream->dev->lock);
 

2.17.1

Hi Guenter,

Thank you for the patches.

I'll review each patch individually, but I think 2/5, 4/5 and 5/5 should be handled in the V4L2 core, not the uvcvideo driver. Otherwise we would have to replicate that logic in all drivers, while I think it can easily be implemented in a generic fashion as previously discussed.

Regards,

Laurent Pinchart

Hi Laurent,

The problem is that the v4l2 core already does support locking. There is a global lock, in struct video_device, a queue lock in struct v4l2_m2m_ctx, and another queue lock in struct vb2_queue. However, all of those have to be initialized from the driver. The uvcvideo driver uses its own locks and does not set the lock pointers in the various generic structures. I was able to figure out how to use the uvcvideo specific locks in the uvcvideo driver, but all my attempts to initialize and use the generic locks failed.

It may well be that the generic code isn't entirely clean - for example I am not sure if the lock protection in v4l2_open() is complete since it doesn't handle disconnects after checking if the video device is still registered (and I don't really see the point of the second video_is_registered() call in v4l2_open). However, that may just be a lack of understanding on my side on how the code is supposed to work. Maybe the actual device open function is expected to have its own protection against underlying hardware removal and video device unregistration while opening the device.

[ Regarding the second call to video_is_registered() in v4l2_open():
Add msleep(5000) between it and the call to the driver open function, disconnect the device during the sleep, and it will happily call the device open function on a non-registered video device. That is what patch 5/5 tries to fix or the uvcvideo driver.
The same problem applies to other file operations in v4l2-dev.c: They all check if the video device is registered before calling the device specific code, but I don't really see the point of doing that because there is no protection against unregistration after the check was made and before/while the device specific code is running.
Patch 4/5 tries to fix this for the uvcvideo driver.
If that is a bug in the v4l2 code, I'll be happy to work on a fix, but the only generic fix I could think of would be to utilize the lock in struct video_device ... but that lock isn't initialized by the uvcvideo driver.
]

Either case, I don't think my understanding of the interaction between v4l2 and uvcvideo is good enough to make more invasive changes. I _think_ any generic improvement should start with refactoring the uvcvideo code to use the v4l2 locking mechanism. However, from the exchange here, my understanding is that this locking mechanism is not used on purpose. That means we'll have a uvcvideo specific locking mechanism, period, and I don't think it is even possible to solve the problem without utilizing this locking mechanism.

Of course, it may as well be that I am completely off track and clueless. After all, the first time I looked into this code was about two weeks ago. So please bear with me if I talk nonsense.

Thanks,
Guenter

Hi Guenter,

It would be rather impolite to claim you're clueless, given that you managed to write this patch series only two weeks after first looking into the problem :-)

I'll try to prototype what I envision would be a good solution in the V4L2 core. If stars align, I may even try to push it one level up, to the chardev layer. Would you then be able to test it ?

Regards,

Laurent Pinchart

[ ... ]

Sure, I'll be happy to do that.

I ordered a couple of non-UVC webcams (pwc and gspca) from eBay for comparison. Both of those use the v4l2 locking mechanism, so we should be able to see the difference.

Thanks,
Guenter

Turns out gspca webcams (or at least the one I got - Logitech QuickCam for Notebooks Deluxe) don't have a problem. As mentioned before, the gspca driver uses the locking mechanism provided by the v4l2/vb2 code. Unlike uvcvideo, its open function doesn't trigger sending send usb packets. Its usb disconnect function acquires the lock, and the rest of the code holds that lock where needed. I''ll keep trying, but so far I can't get it to do anything wrong.

I am still waiting for the pwc webcam, but that also uses the v4l2/vb2 locking mechanism, and it seems unlikely that I'll get it to fail. I'll try anyway.

Guenter