[PATCH] HID: steam: fix deadlock with input devices.

[PATCH] HID: steam: fix deadlock with input devices.

[Rodrigo Rivas Costa]

When using this driver with the wireless dongle and some usermode
program that monitors every input device (acpid, for example), while
another usermode client opens and closes the low-level device
repeadedly, the system eventually deadlocks.

The reason is that steam_input_register_device() must not be called with
the mutex held, because the input subsystem has its own synchronization
that clashes with this one: it is possible that steam_input_open() is
called before input_register_device() returns, and since
steam_input_open() needs to lock the mutex, it deadlocks.

However we must hold the mutex when calling any function that sends
commands to the controller. If not, random commands end up falling fail.

Reported-by: Simon Gene Gottlieb <simon@gottliebtfreitag.de>
Signed-off-by: Rodrigo Rivas Costa <rodrigorivascosta@gmail.com>
Tested-by: Simon Gene Gottlieb <simon@gottliebtfreitag.de>
---
 drivers/hid/hid-steam.c | 26 +++++++++++++++++++-------
 1 file changed, 19 insertions(+), 7 deletions(-)

diff --git a/drivers/hid/hid-steam.c b/drivers/hid/hid-steam.c
index 8141cadfca0e..8dae0f9b819e 100644
--- a/drivers/hid/hid-steam.c
+++ b/drivers/hid/hid-steam.c
@@ -499,6 +499,7 @@ static void steam_battery_unregister(struct steam_device *steam)
 static int steam_register(struct steam_device *steam)
 {
 	int ret;
+	bool client_opened;
 
 	/*
 	 * This function can be called several times in a row with the
@@ -511,9 +512,11 @@ static int steam_register(struct steam_device *steam)
 		 * Unlikely, but getting the serial could fail, and it is not so
 		 * important, so make up a serial number and go on.
 		 */
+		mutex_lock(&steam->mutex);
 		if (steam_get_serial(steam) < 0)
 			strlcpy(steam->serial_no, "XXXXXXXXXX",
 					sizeof(steam->serial_no));
+		mutex_unlock(&steam->mutex);
 
 		hid_info(steam->hdev, "Steam Controller '%s' connected",
 				steam->serial_no);
@@ -528,13 +531,15 @@ static int steam_register(struct steam_device *steam)
 	}
 
 	mutex_lock(&steam->mutex);
-	if (!steam->client_opened) {
+	client_opened = steam->client_opened;
+	if (!client_opened)
 		steam_set_lizard_mode(steam, lizard_mode);
+	mutex_unlock(&steam->mutex);
+
+	if (!client_opened)
 		ret = steam_input_register(steam);
-	} else {
+	else
 		ret = 0;
-	}
-	mutex_unlock(&steam->mutex);
 
 	return ret;
 }
@@ -630,14 +635,21 @@ static void steam_client_ll_close(struct hid_device *hdev)
 {
 	struct steam_device *steam = hdev->driver_data;
 
+	unsigned long flags;
+	bool connected;
+
+	spin_lock_irqsave(&steam->lock, flags);
+	connected = steam->connected;
+	spin_unlock_irqrestore(&steam->lock, flags);
+
 	mutex_lock(&steam->mutex);
 	steam->client_opened = false;
+	if (connected)
+		steam_set_lizard_mode(steam, lizard_mode);
 	mutex_unlock(&steam->mutex);
 
-	if (steam->connected) {
-		steam_set_lizard_mode(steam, lizard_mode);
+	if (connected)
 		steam_input_register(steam);
-	}
 }
 
 static int steam_client_ll_raw_request(struct hid_device *hdev,
-- 
2.21.0

Re: [PATCH] HID: steam: fix deadlock with input devices.

[Jiri Kosina]

On Fri, 15 Mar 2019, Rodrigo Rivas Costa wrote:

> When using this driver with the wireless dongle and some usermode
> program that monitors every input device (acpid, for example), while
> another usermode client opens and closes the low-level device
> repeadedly, the system eventually deadlocks.
> 
> The reason is that steam_input_register_device() must not be called with
> the mutex held, because the input subsystem has its own synchronization
> that clashes with this one: it is possible that steam_input_open() is
> called before input_register_device() returns, and since
> steam_input_open() needs to lock the mutex, it deadlocks.
> 
> However we must hold the mutex when calling any function that sends
> commands to the controller. If not, random commands end up falling fail.
> 
> Reported-by: Simon Gene Gottlieb <simon@gottliebtfreitag.de>
> Signed-off-by: Rodrigo Rivas Costa <rodrigorivascosta@gmail.com>
> Tested-by: Simon Gene Gottlieb <simon@gottliebtfreitag.de>

Applied to for-5.1/upstream-fixes, thanks.

-- 
Jiri Kosina
SUSE Labs