Skip to contentSkip to conversation

fix lock imbalances in /drivers/media/video/cafe_ccic.c

PATCH In review iceberg 4 participants 5 messages Sep 10, 2009 – Sep 11, 2009 Media (V4L/DVB)

lore
Share & tools
lore

iceberg

author
#0 lore raw

PATCH 1/1 fix lock imbalances in /drivers/media/video/cafe_ccic.c

In ./drivers/media/video/cafe_ccic.c, in function cafe_pci_probe:
Mutex must be unlocked before exit
1. On paths starting with mutex lock in line 1912, then continuing in lines:
1929, 1936 (goto unreg) and 1940 (goto iounmap) .
2. On path starting in line 1971 mutex lock, and then continuing in line 1978
(goto out_smbus) mutex.

Fix lock imbalances in function cafe_pci_probe.
Found by Linux Driver Verification project.

---

(quoted diff) +0 -0

b/./b/drivers/media/video/cafe_ccic.c
index c4d181d..2987433 100644

./b/drivers/media/video/cafe_ccic.c renamed from ./a/drivers/media/video/cafe_ccic.c +10 -3
@@ -1925,19 +1925,24 @@ static int cafe_pci_probe(struct pci_dev *pdev,
         * Get set up on the PCI bus.
         */
        ret = pci_enable_device(pdev);
-       if (ret)
+       if (ret) {
+               mutex_unlock(&cam->s_mutex);
                goto out_unreg;
+       }
        pci_set_master(pdev);
 
        ret = -EIO;
        cam->regs = pci_iomap(pdev, 0, 0);
        if (! cam->regs) {
                printk(KERN_ERR "Unable to ioremap cafe-ccic regs\n");
+               mutex_unlock(&cam->s_mutex);
                goto out_unreg;
        }
        ret = request_irq(pdev->irq, cafe_irq, IRQF_SHARED, "cafe-ccic", cam);
-       if (ret)
+       if (ret) {
+               mutex_unlock(&cam->s_mutex);
                goto out_iounmap;
+       }
        /*
         * Initialize the controller and leave it powered up.  It will
         * stay that way until the sensor driver shows up.
@@ -1974,8 +1979,10 @@ static int cafe_pci_probe(struct pci_dev *pdev,
 /*     cam->vdev.debug = V4L2_DEBUG_IOCTL_ARG;*/
        cam->vdev.v4l2_dev = &cam->v4l2_dev;
        ret = video_register_device(&cam->vdev, VFL_TYPE_GRABBER, -1);
-       if (ret)
+       if (ret) {
+               mutex_unlock(&cam->s_mutex);
                goto out_smbus;
+       }
        video_set_drvdata(&cam->vdev, cam);
 
        /*
#1 lore raw
Quoted 9 lines · iceberg

In ./drivers/media/video/cafe_ccic.c, in function cafe_pci_probe:
Mutex must be unlocked before exit
1. On paths starting with mutex lock in line 1912, then continuing in lines:
1929, 1936 (goto unreg) and 1940 (goto iounmap) .
2. On path starting in line 1971 mutex lock, and then continuing in line 1978
(goto out_smbus) mutex.

Fix lock imbalances in function cafe_pci_probe.
Found by Linux Driver Verification project.

Could you run this through checkpatch and fix any issues you find? It looks like you need to use tabs for indentation in a couple places.

Daniel

#2 lore raw
In reply to iceberg
Quoted 6 lines · On Thu, 10 Sep 2009 18:37:34 +0000 iceberg <[email protected]> wrote:

In ./drivers/media/video/cafe_ccic.c, in function cafe_pci_probe:
Mutex must be unlocked before exit
1. On paths starting with mutex lock in line 1912, then continuing in lines:
1929, 1936 (goto unreg) and 1940 (goto iounmap) .
2. On path starting in line 1971 mutex lock, and then continuing in line 1978
(goto out_smbus) mutex.

That's a definite bug, but I hate all those unlocks in the error branches. As it happens, we don't really need the mutex until the device has been exposed to the rest of the kernel, so I propose the following as a better patch.

Thanks for pointing this out,

jon

---
Fix a mutex leak

Certain error exits from cafe_pci_probe() can leave the camera mutex locked. For much of the time, we didn't need the mutex anyway; take it out and add an unlock in the path where it is needed.

---
 drivers/media/video/cafe_ccic.c |    3 +--
 1 files changed, 1 insertions(+), 2 deletions(-)
drivers/media/video/cafe_ccic.c +1 -2
index c4d181d..0f62b5e 100644
@@ -1909,7 +1909,6 @@ static int cafe_pci_probe(struct pci_dev *pdev,
 		goto out_free;
 
 	mutex_init(&cam->s_mutex);
-	mutex_lock(&cam->s_mutex);
 	spin_lock_init(&cam->dev_lock);
 	cam->state = S_NOTREADY;
 	cafe_set_config_needed(cam, 1);
@@ -1949,7 +1948,6 @@ static int cafe_pci_probe(struct pci_dev *pdev,
 	 * because the sensor could attach in this call chain, leading to
 	 * unsightly deadlocks.
 	 */
-	mutex_unlock(&cam->s_mutex);  /* attach can deadlock */
 	ret = cafe_smbus_setup(cam);
 	if (ret)
 		goto out_freeirq;
@@ -1991,6 +1989,7 @@ static int cafe_pci_probe(struct pci_dev *pdev,
 	return 0;
 
 out_smbus:
+	mutex_unlock(&cam->s_mutex);
 	cafe_smbus_shutdown(cam);
 out_freeirq:
 	cafe_ctlr_power_down(cam);
Signature · 1 line
1.6.2.5
#3 lore raw
Quoted 10 lines · Jonathan Corbet
Quoted 6 lines · On Thu, 10 Sep 2009 18:37:34 +0000 iceberg <[email protected]> wrote:

In ./drivers/media/video/cafe_ccic.c, in function cafe_pci_probe:
Mutex must be unlocked before exit
1. On paths starting with mutex lock in line 1912, then continuing in lines:
1929, 1936 (goto unreg) and 1940 (goto iounmap) .
2. On path starting in line 1971 mutex lock, and then continuing in line 1978
(goto out_smbus) mutex.

Quoted 6 lines

That's a definite bug, but I hate all those unlocks in the error branches. As it happens, we don't really need the mutex until the device has been exposed to the rest of the kernel, so I propose the following as a better patch.

Thanks for pointing this out,

Actually, for something like this, I would put the mutex_unlock in the error path, and just add a local variable to tell that it is locked.

	int is_locked = 0;

[...]

	mutex_lock(&cam->s_mutex);
	is_locked = 1;

[...]

	mutex_unlock(&cam->s_mutex);
	is_locked = 0;

[...]

out_iounmap:

	pci_iounmap(pdev, cam->regs);
	if (is_locked)
		mutex_unlock(&cam->s_mutex);

out_free:

[...]

Or something similar. I hate the multiple unlocks too.

-- Steve

Quoted 52 lines

jon

---
Fix a mutex leak

Certain error exits from cafe_pci_probe() can leave the camera mutex locked. For much of the time, we didn't need the mutex anyway; take it out and add an unlock in the path where it is needed.

---
 drivers/media/video/cafe_ccic.c |    3 +--
 1 files changed, 1 insertions(+), 2 deletions(-)
drivers/media/video/cafe_ccic.c +1 -2
index c4d181d..0f62b5e 100644
@@ -1909,7 +1909,6 @@ static int cafe_pci_probe(struct pci_dev *pdev,
 		goto out_free;
 
 	mutex_init(&cam->s_mutex);
-	mutex_lock(&cam->s_mutex);
 	spin_lock_init(&cam->dev_lock);
 	cam->state = S_NOTREADY;
 	cafe_set_config_needed(cam, 1);
@@ -1949,7 +1948,6 @@ static int cafe_pci_probe(struct pci_dev *pdev,
 	 * because the sensor could attach in this call chain, leading to
 	 * unsightly deadlocks.
 	 */
-	mutex_unlock(&cam->s_mutex);  /* attach can deadlock */
 	ret = cafe_smbus_setup(cam);
 	if (ret)
 		goto out_freeirq;
@@ -1991,6 +1989,7 @@ static int cafe_pci_probe(struct pci_dev *pdev,
 	return 0;
 
 out_smbus:
+	mutex_unlock(&cam->s_mutex);
 	cafe_smbus_shutdown(cam);
 out_freeirq:
 	cafe_ctlr_power_down(cam);
Signature · 7 lines
1.6.2.5

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
author
#4 lore raw
In reply to Jonathan Corbet

On Thursday 10 September 2009 15:30:03 you wrote:

	Incorrect patch. See path starting with "if (cam->sensor == null) {" in line 

1960. In this case we goto out_smbs and try mutex_unlock on unlocking mutex.

Quoted 60 lines

On Thu, 10 Sep 2009 18:37:34 +0000

Quoted 6 lines · iceberg

In ./drivers/media/video/cafe_ccic.c, in function cafe_pci_probe:
Mutex must be unlocked before exit
1. On paths starting with mutex lock in line 1912, then continuing in
lines: 1929, 1936 (goto unreg) and 1940 (goto iounmap) .
2. On path starting in line 1971 mutex lock, and then continuing in line
1978 (goto out_smbus) mutex.

That's a definite bug, but I hate all those unlocks in the error branches. As it happens, we don't really need the mutex until the device has been exposed to the rest of the kernel, so I propose the following as a better patch.

Thanks for pointing this out,

jon

---
Fix a mutex leak

Certain error exits from cafe_pci_probe() can leave the camera mutex locked. For much of the time, we didn't need the mutex anyway; take it out and add an unlock in the path where it is needed.

---
 drivers/media/video/cafe_ccic.c |    3 +--
 1 files changed, 1 insertions(+), 2 deletions(-)
(quoted diff) +0 -0

b/drivers/media/video/cafe_ccic.c index c4d181d..0f62b5e 100644

drivers/media/video/cafe_ccic.c +1 -2
@@ -1909,7 +1909,6 @@ static int cafe_pci_probe(struct pci_dev *pdev,
 		goto out_free;
 
 	mutex_init(&cam->s_mutex);
-	mutex_lock(&cam->s_mutex);
 	spin_lock_init(&cam->dev_lock);
 	cam->state = S_NOTREADY;
 	cafe_set_config_needed(cam, 1);
@@ -1949,7 +1948,6 @@ static int cafe_pci_probe(struct pci_dev *pdev,
 	 * because the sensor could attach in this call chain, leading to
 	 * unsightly deadlocks.
 	 */
-	mutex_unlock(&cam->s_mutex);  /* attach can deadlock */
 	ret = cafe_smbus_setup(cam);
 	if (ret)
 		goto out_freeirq;
@@ -1991,6 +1989,7 @@ static int cafe_pci_probe(struct pci_dev *pdev,
 	return 0;
 
 out_smbus:
+	mutex_unlock(&cam->s_mutex);
 	cafe_smbus_shutdown(cam);
 out_freeirq:
 	cafe_ctlr_power_down(cam);