@@ -58,6 +58,9 @@ MODULE_DEVICE_TABLE (usb, wdm_ids);
#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
@@ -151,7 +154,7 @@ static void wdm_out_callback(struct urb *urb)
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)
@@ -393,6 +396,9 @@ static ssize_t wdm_write
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;
@@ -424,6 +430,7 @@ static ssize_t wdm_write
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;
@@ -583,28 +590,58 @@ static ssize_t wdm_read
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. */
+
+ /*
+ * 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);
- 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));
-
- /* 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;
+
+ 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);
+}
- return usb_translate_errors(desc->werr);
+/*
+ * 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)
@@ -729,6 +766,7 @@ static const struct file_operations wdm_fops = {
.owner = THIS_MODULE,
.read = wdm_read,
.write = wdm_write,
+ .fsync = wdm_fsync,
.open = wdm_open,
.flush = wdm_flush,
.release = wdm_release,
Based on https://lkml.kernel.org/r/20200923092136.14824-1-oneukum@suse.com from Oliver Neukum <oneukum@suse.com>: 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/27b7545e-8f41-10b8-7c02-e35a08eb1611@i-love.sakura.ne.jp [3] https://lkml.kernel.org/r/79ba410f-e0ef-2465-b94f-6b9a4a82adf5@i-love.sakura.ne.jp [4] https://lkml.kernel.org/r/20200530011040.GB12419@rowland.harvard.edu [5] https://lkml.kernel.org/r/c85331fc-874c-6e46-a77f-0ef1dc075308@i-love.sakura.ne.jp Reported-by: syzbot <syzbot+854768b99f19e89d7f81@syzkaller.appspotmail.com> Signed-off-by: Oliver Neukum <oneukum@suse.com> Debugged-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Alan Stern <stern@rowland.harvard.edu> --- drivers/usb/class/cdc-wdm.c | 72 ++++++++++++++++++++++++++++--------- 1 file changed, 55 insertions(+), 17 deletions(-)