USB: cdc-wdm: Make wdm_flush() interruptible and add wdm_fsync().

syzbot is reporting hung task at wdm_flush() [1], for there is a circular
dependency that wdm_flush() from flip_close() for /dev/cdc-wdm0 forever
waits for /dev/raw-gadget to be closed while close() for /dev/raw-gadget
cannot be called unless close() for /dev/cdc-wdm0 completes.

Tetsuo Handa considered that such circular dependency is an usage error [2]
which corresponds to an unresponding broken hardware [3]. But Alan Stern
responded that we should be prepared for such hardware [4]. Therefore,
this patch changes wdm_flush() to use wait_event_interruptible_timeout()
which gives up after 30 seconds, for hardware that remains silent must be
ignored. The 30 seconds are coming out of thin air.

Changing wait_event() to wait_event_interruptible_timeout() makes error
reporting from close() syscall less reliable. To compensate it, this patch
also implements wdm_fsync() which does not use timeout. Those who want to
be very sure that data has gone out to the device are now advised to call
fsync(), with a caveat that fsync() can return -EINVAL when running on
older kernels which do not implement wdm_fsync().

This patch also fixes three more problems (listed below) found during
exhaustive discussion and testing.

Since multiple threads can concurrently call wdm_write()/wdm_flush(),
we need to use wake_up_all() whenever clearing WDM_IN_USE in order to
make sure that all waiters are woken up. Also, error reporting needs
to use fetch-and-clear approach in order not to report same error for
multiple times.

Since wdm_flush() checks WDM_DISCONNECTING, wdm_write() should as well
check WDM_DISCONNECTING.

In wdm_flush(), since locks are not held, it is not safe to dereference
desc->intf after checking that WDM_DISCONNECTING is not set [5]. Thus,
remove dev_err() from wdm_flush().

[1] https://syzkaller.appspot.com/bug?id=e7b761593b23eb50855b9ea31e3be5472b711186
[2] https://lkml.kernel.org/r/[e-mail]
[3] https://lkml.kernel.org/r/[e-mail]
[4] https://lkml.kernel.org/r/[e-mail]
[5] https://lkml.kernel.org/r/[e-mail]

Reported-by: syzbot <[e-mail]>
Cc: stable <[e-mail]>
Co-developed-by: Tetsuo Handa <[e-mail]>
Signed-off-by: Tetsuo Handa <[e-mail]>
Signed-off-by: Oliver Neukum <[e-mail]>
Cc: Alan Stern <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]
Signed-off-by: Greg Kroah-Hartman <[e-mail]>

drivers/usb/class/cdc-wdm.c

···
 
 #define WDM_MAX			16
 
+/* we cannot wait forever at flush() */
+#define WDM_FLUSH_TIMEOUT	(30 * HZ)
+
 /* CDC-WMC r1.1 requires wMaxCommand to be "at least 256 decimal (0x100)" */
 #define WDM_DEFAULT_BUFSIZE	256
 
···
 	kfree(desc->outbuf);
 	desc->outbuf = NULL;
 	clear_bit(WDM_IN_USE, &desc->flags);
-	wake_up(&desc->wait);
+	wake_up_all(&desc->wait);
 }
 
 static void wdm_in_callback(struct urb *urb)
···
 	if (test_bit(WDM_RESETTING, &desc->flags))
 		r = -EIO;
 
+	if (test_bit(WDM_DISCONNECTING, &desc->flags))
+		r = -ENODEV;
+
 	if (r < 0) {
 		rv = r;
 		goto out_free_mem_pm;
···
 	if (rv < 0) {
 		desc->outbuf = NULL;
 		clear_bit(WDM_IN_USE, &desc->flags);
+		wake_up_all(&desc->wait); /* for wdm_wait_for_response() */
 		dev_err(&desc->intf->dev, "Tx URB error: %d\n", rv);
 		rv = usb_translate_errors(rv);
 		goto out_free_mem_pm;
···
 	return rv;
 }
 
-static int wdm_flush(struct file *file, fl_owner_t id)
+static int wdm_wait_for_response(struct file *file, long timeout)
 {
 	struct wdm_device *desc = file->private_data;
+	long rv; /* Use long here because (int) MAX_SCHEDULE_TIMEOUT < 0. */
 
-	wait_event(desc->wait,
-			/*
-			 * needs both flags. We cannot do with one
-			 * because resetting it would cause a race
-			 * with write() yet we need to signal
-			 * a disconnect
-			 */
-			!test_bit(WDM_IN_USE, &desc->flags) ||
-			test_bit(WDM_DISCONNECTING, &desc->flags));
+	/*
+	 * Needs both flags. We cannot do with one because resetting it would
+	 * cause a race with write() yet we need to signal a disconnect.
+	 */
+	rv = wait_event_interruptible_timeout(desc->wait,
+			      !test_bit(WDM_IN_USE, &desc->flags) ||
+			      test_bit(WDM_DISCONNECTING, &desc->flags),
+			      timeout);
 
-	/* cannot dereference desc->intf if WDM_DISCONNECTING */
+	/*
+	 * To report the correct error. This is best effort.
+	 * We are inevitably racing with the hardware.
+	 */
 	if (test_bit(WDM_DISCONNECTING, &desc->flags))
 		return -ENODEV;
-	if (desc->werr < 0)
-		dev_err(&desc->intf->dev, "Error in flush path: %d\n",
-			desc->werr);
+	if (!rv)
+		return -EIO;
+	if (rv < 0)
+		return -EINTR;
 
-	return usb_translate_errors(desc->werr);
+	spin_lock_irq(&desc->iuspin);
+	rv = desc->werr;
+	desc->werr = 0;
+	spin_unlock_irq(&desc->iuspin);
+
+	return usb_translate_errors(rv);
+
+}
+
+/*
+ * You need to send a signal when you react to malicious or defective hardware.
+ * Also, don't abort when fsync() returned -EINVAL, for older kernels which do
+ * not implement wdm_flush() will return -EINVAL.
+ */
+static int wdm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
+{
+	return wdm_wait_for_response(file, MAX_SCHEDULE_TIMEOUT);
+}
+
+/*
+ * Same with wdm_fsync(), except it uses finite timeout in order to react to
+ * malicious or defective hardware which ceased communication after close() was
+ * implicitly called due to process termination.
+ */
+static int wdm_flush(struct file *file, fl_owner_t id)
+{
+	return wdm_wait_for_response(file, WDM_FLUSH_TIMEOUT);
 }
 
 static __poll_t wdm_poll(struct file *file, struct poll_table_struct *wait)
···
 	.owner =	THIS_MODULE,
 	.read =		wdm_read,
 	.write =	wdm_write,
+	.fsync =	wdm_fsync,
 	.open =		wdm_open,
 	.flush =	wdm_flush,
 	.release =	wdm_release,