The custom ring-buffer implementation was merged without any locking or explicit memory barriers, but a spinlock was later added by commit 9d33efd9a791 ("USB: ldusb bugfix").
The lock did not cover the update of the tail index once the entry had been processed, something which could lead to memory corruption on weakly ordered architectures or due to compiler optimisations.
Specifically, a completion handler running on another CPU might observe the incremented tail index and update the entry before ld_usb_read() is done with it.
Fixes: 2824bd250f0b ("[PATCH] USB: add ldusb driver") Fixes: 9d33efd9a791 ("USB: ldusb bugfix") Cc: stable stable@vger.kernel.org # 2.6.13 Signed-off-by: Johan Hovold johan@kernel.org --- drivers/usb/misc/ldusb.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/usb/misc/ldusb.c b/drivers/usb/misc/ldusb.c index 15b5f06fb0b3..c3e764909fd0 100644 --- a/drivers/usb/misc/ldusb.c +++ b/drivers/usb/misc/ldusb.c @@ -495,11 +495,11 @@ static ssize_t ld_usb_read(struct file *file, char __user *buffer, size_t count, retval = -EFAULT; goto unlock_exit; } - dev->ring_tail = (dev->ring_tail+1) % ring_buffer_size; - retval = bytes_to_read;
spin_lock_irq(&dev->rbsl); + dev->ring_tail = (dev->ring_tail + 1) % ring_buffer_size; + if (dev->buffer_overflow) { dev->buffer_overflow = 0; spin_unlock_irq(&dev->rbsl);