[PATCH 4.4 010/114] HID: uhid: Fix worker destroying device without any protection

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



From: Jann Horn <jannh@xxxxxxxxxx>

commit 4ea5763fb79ed89b3bdad455ebf3f33416a81624 upstream.

uhid has to run hid_add_device() from workqueue context while allowing
parallel use of the userspace API (which is protected with ->devlock).
But hid_add_device() can fail. Currently, that is handled by immediately
destroying the associated HID device, without using ->devlock - but if
there are concurrent requests from userspace, that's wrong and leads to
NULL dereferences and/or memory corruption (via use-after-free).

Fix it by leaving the HID device as-is in the worker. We can clean it up
later, either in the UHID_DESTROY command handler or in the ->release()
handler.

Cc: stable@xxxxxxxxxxxxxxx
Fixes: 67f8ecc550b5 ("HID: uhid: fix timeout when probe races with IO")
Signed-off-by: Jann Horn <jannh@xxxxxxxxxx>
Signed-off-by: Jiri Kosina <jkosina@xxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
---
 drivers/hid/uhid.c |   29 +++++++++++++++++++++++++----
 1 file changed, 25 insertions(+), 4 deletions(-)

--- a/drivers/hid/uhid.c
+++ b/drivers/hid/uhid.c
@@ -33,11 +33,22 @@
 
 struct uhid_device {
 	struct mutex devlock;
+
+	/* This flag tracks whether the HID device is usable for commands from
+	 * userspace. The flag is already set before hid_add_device(), which
+	 * runs in workqueue context, to allow hid_add_device() to communicate
+	 * with userspace.
+	 * However, if hid_add_device() fails, the flag is cleared without
+	 * holding devlock.
+	 * We guarantee that if @running changes from true to false while you're
+	 * holding @devlock, it's still fine to access @hid.
+	 */
 	bool running;
 
 	__u8 *rd_data;
 	uint rd_size;
 
+	/* When this is NULL, userspace may use UHID_CREATE/UHID_CREATE2. */
 	struct hid_device *hid;
 	struct uhid_event input_buf;
 
@@ -68,9 +79,18 @@ static void uhid_device_add_worker(struc
 	if (ret) {
 		hid_err(uhid->hid, "Cannot register HID device: error %d\n", ret);
 
-		hid_destroy_device(uhid->hid);
-		uhid->hid = NULL;
+		/* We used to call hid_destroy_device() here, but that's really
+		 * messy to get right because we have to coordinate with
+		 * concurrent writes from userspace that might be in the middle
+		 * of using uhid->hid.
+		 * Just leave uhid->hid as-is for now, and clean it up when
+		 * userspace tries to close or reinitialize the uhid instance.
+		 *
+		 * However, we do have to clear the ->running flag and do a
+		 * wakeup to make sure userspace knows that the device is gone.
+		 */
 		uhid->running = false;
+		wake_up_interruptible(&uhid->report_wait);
 	}
 }
 
@@ -479,7 +499,7 @@ static int uhid_dev_create2(struct uhid_
 	void *rd_data;
 	int ret;
 
-	if (uhid->running)
+	if (uhid->hid)
 		return -EALREADY;
 
 	rd_size = ev->u.create2.rd_size;
@@ -560,7 +580,7 @@ static int uhid_dev_create(struct uhid_d
 
 static int uhid_dev_destroy(struct uhid_device *uhid)
 {
-	if (!uhid->running)
+	if (!uhid->hid)
 		return -EINVAL;
 
 	uhid->running = false;
@@ -569,6 +589,7 @@ static int uhid_dev_destroy(struct uhid_
 	cancel_work_sync(&uhid->worker);
 
 	hid_destroy_device(uhid->hid);
+	uhid->hid = NULL;
 	kfree(uhid->rd_data);
 
 	return 0;





[Index of Archives]     [Linux Kernel]     [Kernel Development Newbies]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite Hiking]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux