Skip to contentSkip to conversation

BAD PATCH: USB: remove use of the bus rwsem, as it doesn't really protect anything.

DISCUSSION Alan Cox 2 participants 2 messages Apr 27, 2007 – May 2, 2007 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

Alan Cox

author
#1 lore raw

Unless there is something I'm missing most of these patches seem totally unsafe.

drivers/usb/core/devices.c static char *usb_dump_interface_descriptor(char *start, char *end, · line 251

Quoted 10 lines
drivers/usb/core/devices.c +0 -1
@@ -246,7 +246,6 @@ static char *usb_dump_interface_descriptor(char *start, char *end,
 
 	if (start > end)
 		return start;
-	down_read(&usb_bus_type.subsys.rwsem);
 	if (iface) {
 		driver_name = (iface->dev.driver
 				? iface->dev.driver->name

iface->dev.driver can now become NULL during the evaluation above

drivers/usb/core/devices.c static int releaseintf(struct dev_state *ps, unsigned int ifnum) · line 448

Quoted 11 lines
drivers/usb/core/devices.c +0 -2
@@ -444,8 +441,6 @@ static int releaseintf(struct dev_state *ps, unsigned int ifnum)
 	if (ifnum >= 8*sizeof(ps->ifclaimed))
 		return err;
 	dev = ps->dev;
-	/* lock against other changes to driver bindings */
-	down_write(&usb_bus_type.subsys.rwsem);
 	intf = usb_ifnum_to_if(dev, ifnum);
 	if (!intf)
 		err = -ENOENT;
@@ -453,7 +448,6 @@ static int releaseintf(struct dev_state *ps, unsigned int ifnum)
 		usb_driver_release_interface(&usbfs_driver, intf);

Which takes iface->dev.driver to NULL

Quoted 7 lines
drivers/usb/core/devices.c +0 -1
 		err = 0;
 	}
-	up_write(&usb_bus_type.subsys.rwsem);
 	return err;
 }
 

drivers/usb/core/devices.c static int proc_getdriver(struct dev_state *ps, void __user *arg) · line 812

Quoted 7 lines
drivers/usb/core/devices.c +0 -1
@@ -813,7 +807,6 @@ static int proc_getdriver(struct dev_state *ps, void __user *arg)
 
 	if (copy_from_user(&gd, arg, sizeof(gd)))
 		return -EFAULT;
-	down_read(&usb_bus_type.subsys.rwsem);
 	intf = usb_ifnum_to_if(ps->dev, gd.interface);
 	if (!intf || !intf->dev.driver)
 		ret = -ENODATA;

Ditto ...

#2 lore raw

On Fri, Apr 27, 2007 at 10:49:32PM +0100, Alan Cox wrote:

(please CC: me on patches that I submitted, I missed this...)

Quoted 2 lines

Unless there is something I'm missing most of these patches seem totally unsafe.

Hm, no, they were using a lock that was never being used by the core, thereby protecting nothing.

Quoted 4 lines
Quoted 4 lines
drivers/usb/core/devices.c +0 -0
@@ -246,7 +246,6 @@ static char *usb_dump_interface_descriptor(char *start, char *end,
 
Quoted 8 lines

drivers/usb/core/devices.c static char *usb_dump_interface_descriptor(char *start, char *end, · line 251

Quoted 6 lines
drivers/usb/core/devices.c +0 -1
 	if (start > end)
 		return start;
-	down_read(&usb_bus_type.subsys.rwsem);
 	if (iface) {
 		driver_name = (iface->dev.driver
 				? iface->dev.driver->name

iface->dev.driver can now become NULL during the evaluation above

All of the USB stuff is already covered by a different lock, see the linux-usb-devel list for when I originally proposed this patch (also
CC:ed to verify I didn't miss anything...)

Quoted 22 lines

drivers/usb/core/devices.c static int releaseintf(struct dev_state *ps, unsigned int ifnum) · line 448

Quoted 11 lines
drivers/usb/core/devices.c +0 -2
@@ -444,8 +441,6 @@ static int releaseintf(struct dev_state *ps, unsigned int ifnum)
 	if (ifnum >= 8*sizeof(ps->ifclaimed))
 		return err;
 	dev = ps->dev;
-	/* lock against other changes to driver bindings */
-	down_write(&usb_bus_type.subsys.rwsem);
 	intf = usb_ifnum_to_if(dev, ifnum);
 	if (!intf)
 		err = -ENOENT;
@@ -453,7 +448,6 @@ static int releaseintf(struct dev_state *ps, unsigned int ifnum)
 		usb_driver_release_interface(&usbfs_driver, intf);

Which takes iface->dev.driver to NULL

Quoted 8 lines
drivers/usb/core/devices.c +0 -1
 		err = 0;
 	}
-	up_write(&usb_bus_type.subsys.rwsem);
 	return err;
 }
 
@@ -813,7 +807,6 @@ static int proc_getdriver(struct dev_state *ps, void __user *arg)
 
Quoted 8 lines

drivers/usb/core/devices.c static int proc_getdriver(struct dev_state *ps, void __user *arg) · line 812

Quoted 6 lines
drivers/usb/core/devices.c +0 -1
 	if (copy_from_user(&gd, arg, sizeof(gd)))
 		return -EFAULT;
-	down_read(&usb_bus_type.subsys.rwsem);
 	intf = usb_ifnum_to_if(ps->dev, gd.interface);
 	if (!intf || !intf->dev.driver)
 		ret = -ENODATA;

Ditto ...

thanks,

greg k-h