[PATCH] Bluetooth: fix oops in rfcomm tty code

From: Vegard Nossum
Date: Sun Jul 13 2008 - 13:02:11 EST

Soeren Sonnenburg reported:
> this oops happened after a couple of s2ram cycles so it might be very
> well crap. However I somehow triggered it by /etc/init.d/bluetooth
> stop/start's which also call hid2hci maybe even a connection was about
> to be established at that time. As I remember having seen a problem like
> this before I thought I report it (even though I have a madwifi tainted
> kernel).
> kobject_add_internal failed for rfcomm0 with -EEXIST, don't try to register things with the same name in the same directory.

It turns out that the following sequence of actions will reproduce the

1. Create a new rfcomm device (using RFCOMMCREATEDEV ioctl)
2. (Try to) open the device
3. Release the rfcomm device (using RFCOMMRELEASEDEV ioctl)

At this point, the "rfcomm?" tty is still in use, but the device is gone
from the internal rfcomm list, so the device id can be reused.

4. Create a new rfcomm device with the same device id as before

And now kobject will complain that the tty already exists.

(See http://lkml.org/lkml/2008/7/13/89 for a reproducible test-case.)

This patch attempts to correct this by only removing the device from the
internal rfcomm list of devices at the final unregister, so that the id
won't get reused until the device has been completely destructed.

This should be safe as the RFCOMM_TTY_RELEASED bit will be set for the
device and prevent the device from being reopened after it has been

We also fix a race (which would lead to the same oops) by including the
tty setup/teardown code in the rfcomm_dev_lock, the lock protecting the
list of devices.

Thanks to Dave Young for additional suggestions.

Reported-by: Soeren Sonnenburg <kernel@xxxxxx>
Cc: Marcel Holtmann <marcel@xxxxxxxxxxxx>
Cc: Maxim Krasnyansky <maxk@xxxxxxxxxxxx>
Cc: David Woodhouse <dwmw2@xxxxxxxxxxxxx>
Cc: Dave Young <hidave.darkstar@xxxxxxxxx>
Signed-off-by: Vegard Nossum <vegard.nossum@xxxxxxxxx>
net/bluetooth/rfcomm/tty.c | 27 ++++++++++++++-------------
1 files changed, 14 insertions(+), 13 deletions(-)

diff --git a/net/bluetooth/rfcomm/tty.c b/net/bluetooth/rfcomm/tty.c
index c919187..1709ccf 100644
--- a/net/bluetooth/rfcomm/tty.c
+++ b/net/bluetooth/rfcomm/tty.c
@@ -96,9 +96,11 @@ static void rfcomm_dev_destruct(struct rfcomm_dev *dev)
BT_DBG("dev %p dlc %p", dev, dlc);

/* Refcount should only hit zero when called from rfcomm_dev_del()
- which will have taken us off the list. Everything else are
- refcounting bugs. */
- BUG_ON(!list_empty(&dev->list));
+ which will have set the RFCOMM_TTY_RELEASED bit. Everything else
+ are refcounting bugs. */
+ BUG_ON(!test_bit(RFCOMM_TTY_RELEASED, &dev->flags));
+ write_lock_bh(&rfcomm_dev_lock);

/* Detach DLC if it's owned by this dev */
@@ -108,8 +110,11 @@ static void rfcomm_dev_destruct(struct rfcomm_dev *dev)


+ list_del_init(&dev->list);
tty_unregister_device(rfcomm_tty_driver, dev->id);

+ write_unlock_bh(&rfcomm_dev_lock);

/* It's safe to call module_put() here because socket still
@@ -127,10 +132,8 @@ static inline void rfcomm_dev_put(struct rfcomm_dev *dev)
/* The reason this isn't actually a race, as you no
doubt have a little voice screaming at you in your
head, is that the refcount should never actually
- reach zero unless the device has already been taken
- off the list, in rfcomm_dev_del(). And if that's not
- true, we'll hit the BUG() in rfcomm_dev_destruct()
- anyway. */
+ reach zero unless we've already set the
+ RFCOMM_TTY_RELEASED bit in rfcomm_dev_del(). */
if (atomic_dec_and_test(&dev->refcnt))
@@ -278,9 +281,8 @@ static int rfcomm_dev_add(struct rfcomm_dev_req *req, struct rfcomm_dlc *dlc)

- write_unlock_bh(&rfcomm_dev_lock);
if (err < 0) {
+ write_unlock_bh(&rfcomm_dev_lock);
return err;
@@ -290,6 +292,7 @@ out:
if (IS_ERR(dev->tty_dev)) {
err = PTR_ERR(dev->tty_dev);
+ write_unlock_bh(&rfcomm_dev_lock);
return err;
@@ -302,6 +305,8 @@ out:
if (device_create_file(dev->tty_dev, &dev_attr_channel) < 0)
BT_ERR("Failed to create channel attribute");

+ write_unlock_bh(&rfcomm_dev_lock);
return dev->id;

@@ -314,10 +319,6 @@ static void rfcomm_dev_del(struct rfcomm_dev *dev)
set_bit(RFCOMM_TTY_RELEASED, &dev->flags);

- write_lock_bh(&rfcomm_dev_lock);
- list_del_init(&dev->list);
- write_unlock_bh(&rfcomm_dev_lock);


To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/