Skip to contentSkip to conversation

usb: gadget: goku_udc: Fix error path

RESEND/PATCH New Rahul Ruikar 3 participants 7 messages Oct 3, 2010 – Oct 5, 2010 USB

lore
Share & tools
lore
#0 lore raw

call put_device() when device_register() fails.

---
 drivers/usb/gadget/goku_udc.c |   12 ++++++++++--
 drivers/usb/gadget/goku_udc.h |    3 ++-
 2 files changed, 12 insertions(+), 3 deletions(-)
drivers/usb/gadget/goku_udc.c +10 -2
index 1088d08..bb70125 100644
@@ -1744,7 +1744,10 @@ static void goku_remove(struct pci_dev *pdev)
 				pci_resource_len (pdev, 0));
 	if (dev->enabled)
 		pci_disable_device(pdev);
-	device_unregister(&dev->gadget.dev);
+	if (dev->reg_status == 2)
+		device_unregister(&dev->gadget.dev);
+	else if (dev->reg_status == 1)
+		put_device(&dev->gadget.dev);
 
 	pci_set_drvdata(pdev, NULL);
 	dev->regs = NULL;
@@ -1847,8 +1850,13 @@ static int goku_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	/* done */
 	the_controller = dev;
 	retval = device_register(&dev->gadget.dev);
-	if (retval == 0)
+	if (retval != 0) {
+		dev->reg_status = 2;
+		goto done;
+	} else {
+		dev->reg_status = 1;
 		return 0;
+	}
 
 done:
 	if (dev)
drivers/usb/gadget/goku_udc.h +2 -1
index 566cb23..64f2bfb 100644
@@ -251,7 +251,8 @@ struct goku_udc {
 					got_region:1,
 					req_config:1,
 					configured:1,
-					enabled:1;
+					enabled:1,
+					reg_status:2;
 
 	/* pci state used to access those endpoints */
 	struct pci_dev			*pdev;
Signature · 1 line
1.7.2.3
#1 lore raw
Quoted 2 lines · Rahul Ruikar

call put_device() when device_register() fails.

Sorry I didn't realize what you were trying to do here. This is not correct at all.

The right thing is to fix device_register() to call put_device() itself. It's a bit involved, because all the callers will need to be audited but someone is working on this I think.

regards,
dan carpenter

#2 lore raw

Dan,

Following things I tried to do with this change.
there can be 3 cases where one need to handle device_register()
1) reaching error path but device_register() is never called..( ie, error occurred before calling device_register() function call, in this case "dev->reg_status" will have value "0" and in error handler device_unregister() or put_device() will not be called.

2) error occurred at device_register() ( ie. it fails and calls error handler, this way "dev->reg_status" will have value "2" and in error handler, based on this value "put_device() will be called.

3) error occured after device_register() success. ( ie, error occured due to some other function, ) in this case "dev->reg_status" will have value "1" and in error handler, based on this value "device_unregister" will be called.

but anyways someone has now proposed to change it to new usb interface, so this patch is of no use now.

Thanks
Rahul Ruikar

Quoted 15 lines · Dan Carpenter
Quoted 2 lines · Rahul Ruikar

call put_device() when device_register() fails.

Sorry I didn't realize what you were trying to do here.  This is not correct at all.

The right thing is to fix device_register() to call put_device() itself. It's a bit involved, because all the callers will need to be audited but someone is working on this I think.

regards,
dan carpenter

#3 lore raw
Quoted 9 lines · Rahul Ruikar

Dan,

Following things I tried to do with this change.
there can be 3 cases where one need to handle device_register()
1) reaching error path but device_register() is never called..( ie, error occurred before calling device_register() function call, in this case "dev->reg_status" will have value "0" and in error handler device_unregister() or put_device() will not be called.

Good point. This is a bug in the original code.

Quoted 4 lines

2) error occurred at device_register() ( ie. it fails and calls error handler, this way "dev->reg_status" will have value "2" and in error handler, based on this value "put_device() will be called.

Yeah. But device_register() sucks. It shouldn't require such byzantine error handling. Someone is working on this.

regards,
dan carpenter

Quoted 11 lines

3) error occured after device_register() success. ( ie, error occured due to some other function, ) in this case "dev->reg_status" will have value "1" and in error handler, based on this value "device_unregister" will be called.

but anyways someone has now proposed to change it to new usb interface, so this patch is of no use now.

Thanks
Rahul Ruikar

#4 lore raw
In reply to Dan Carpenter
Quoted 10 lines · Dan Carpenter
Quoted 2 lines · Rahul Ruikar

call put_device() when device_register() fails.

Sorry I didn't realize what you were trying to do here. This is not correct at all.

The right thing is to fix device_register() to call put_device() itself. It's a bit involved, because all the callers will need to be audited but someone is working on this I think.

No, no one is working on this, and no, it's not possible to make this type of change to the driver core as discussed in the past due to the way struct device can be embedded within another structure.

thanks,

greg k-h

#5 lore raw
Quoted 8 lines · Greg KH
Quoted 7 lines · Dan Carpenter
Quoted 2 lines · Rahul Ruikar

call put_device() when device_register() fails.

Sorry I didn't realize what you were trying to do here. This is not correct at all.

Quoted 7 lines
Quoted 3 lines

The right thing is to fix device_register() to call put_device() itself. It's a bit involved, because all the callers will need to be audited but someone is working on this I think.

No, no one is working on this, and no, it's not possible to make this type of change to the driver core as discussed in the past due to the way struct device can be embedded within another structure.

Sorry I guess I lost track of the discussion. We could write a helper macro which added the extra put device on failure? Isn't that the more common case anyway? To be honest, I'm still not clear on when the extra put_device() is appropriate and when it's not. :/ I'll look through the archive.

Anyway, I would prefer a slightly different patch. The logic is easier to follow if we keep the put_device() right next to the device_register(). Also if we keep it in one function and we add a helper macro later then coccinelle can rewrite it automatically.

I'll send my patch.

regards,
dan carpenter

#6 lore raw

PATCH 1/1 usb: gadget: goku_udc: Fix error path

In reply to Greg KH

This is based on an initial patch by Rahul Ruikar.

The goku_remove() function can be called before device_register() so it can call device_unregister() improperly. Also if the call to device_register() fails we need to call put_device().

As I was changing the error handling in goku_probe(), I noticed that the label was "done" but actually if the function succeeds we return earlier. I renamed the error path to "err" instead of "done."

drivers/usb/gadget/goku_udc.c +15 -11
index 1088d08..7232018 100644
@@ -1744,7 +1744,8 @@ static void goku_remove(struct pci_dev *pdev)
 				pci_resource_len (pdev, 0));
 	if (dev->enabled)
 		pci_disable_device(pdev);
-	device_unregister(&dev->gadget.dev);
+	if (dev->registered)
+		device_unregister(&dev->gadget.dev);
 
 	pci_set_drvdata(pdev, NULL);
 	dev->regs = NULL;
@@ -1774,7 +1775,7 @@ static int goku_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (!pdev->irq) {
 		printk(KERN_ERR "Check PCI %s IRQ setup!\n", pci_name(pdev));
 		retval = -ENODEV;
-		goto done;
+		goto err;
 	}
 
 	/* alloc, and start init */
@@ -1782,7 +1783,7 @@ static int goku_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (dev == NULL){
 		pr_debug("enomem %s\n", pci_name(pdev));
 		retval = -ENOMEM;
-		goto done;
+		goto err;
 	}
 
 	spin_lock_init(&dev->lock);
@@ -1800,7 +1801,7 @@ static int goku_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	retval = pci_enable_device(pdev);
 	if (retval < 0) {
 		DBG(dev, "can't enable, %d\n", retval);
-		goto done;
+		goto err;
 	}
 	dev->enabled = 1;
 
@@ -1809,7 +1810,7 @@ static int goku_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (!request_mem_region(resource, len, driver_name)) {
 		DBG(dev, "controller already in use\n");
 		retval = -EBUSY;
-		goto done;
+		goto err;
 	}
 	dev->got_region = 1;
 
@@ -1817,7 +1818,7 @@ static int goku_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (base == NULL) {
 		DBG(dev, "can't map memory\n");
 		retval = -EFAULT;
-		goto done;
+		goto err;
 	}
 	dev->regs = (struct goku_udc_regs __iomem *) base;
 
@@ -1833,7 +1834,7 @@ static int goku_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 			driver_name, dev) != 0) {
 		DBG(dev, "request interrupt %d failed\n", pdev->irq);
 		retval = -EBUSY;
-		goto done;
+		goto err;
 	}
 	dev->got_irq = 1;
 	if (use_dma)
@@ -1844,13 +1845,16 @@ static int goku_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	create_proc_read_entry(proc_node_name, 0, NULL, udc_proc_read, dev);
 #endif
 
-	/* done */
 	the_controller = dev;
 	retval = device_register(&dev->gadget.dev);
-	if (retval == 0)
-		return 0;
+	if (retval) {
+		put_device(&dev->gadget.dev);
+		goto err;
+	}
+	dev->registered = 1;
+	return 0;
 
-done:
+err:
 	if (dev)
 		goku_remove (pdev);
 	return retval;