On Mon, May 02, 2022 at 12:49:04PM +0900, Tetsuo Handa wrote: > Since usb_register_dev() from imon_init_display() from imon_probe() holds > minor_rwsem while display_open() which holds driver_lock and ictx->lock is > called with minor_rwsem held from usb_open(), holding driver_lock or > ictx->lock when calling usb_register_dev() causes circular locking > dependency problem. > > Since usb_deregister_dev() from imon_disconnect() holds minor_rwsem while > display_open() which holds driver_lock is called with minor_rwsem held, > holding driver_lock when calling usb_deregister_dev() also causes circular > locking dependency problem. > > Sean Young explained that the problem is there are imon devices which have > two usb interfaces, even though it is one device. The probe and disconnect > function of both usb interfaces can run concurrently. > > Alan Stern responded that the driver and USB cores guarantee that when an > interface is probed, both the interface and its USB device are locked. > Ditto for when the disconnect callback gets run. So concurrent probing/ > disconnection of multiple interfaces on the same device is not possible. > > Therefore, we don't need locks for handling race between imon_probe() and > imon_disconnect(). But we still need to handle race between display_open() > /vfd_write()/lcd_write()/display_close() and imon_disconnect(), for > disconnect event can happen while file descriptors are in use. > > Since "struct file"->private_data is set by display_open(), vfd_write()/ > lcd_write()/display_close() can assume that "struct file"->private_data > is not NULL even after usb_set_intfdata(interface, NULL) was called. > > Replace insufficiently held driver_lock with refcount_t based management. > Add a boolean flag for recording whether imon_disconnect() was already > called. Use RCU for accessing this boolean flag and refcount_t. > > Since the boolean flag for imon_disconnect() is shared, disconnect event > on either intf0 or intf1 affects both interfaces. But I assume that this > change does not matter, for usually disconnect event would not happen > while interfaces are in use. > > Link: https://syzkaller.appspot.com/bug?extid=c558267ad910fc494497 > Reported-by: syzbot <syzbot+c558267ad910fc494497@xxxxxxxxxxxxxxxxxxxxxxxxx> > Signed-off-by: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx> > Tested-by: syzbot <syzbot+c558267ad910fc494497@xxxxxxxxxxxxxxxxxxxxxxxxx> > Cc: Sean Young <sean@xxxxxxxx> > Cc: Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> > --- > Changes in v2: > Defer free_imon_context() using refcount till display_close() is called. > > drivers/media/rc/imon.c | 99 +++++++++++++++++++---------------------- > 1 file changed, 47 insertions(+), 52 deletions(-) > > diff --git a/drivers/media/rc/imon.c b/drivers/media/rc/imon.c > index 54da6f60079b..9a4f24e294bc 100644 > --- a/drivers/media/rc/imon.c > +++ b/drivers/media/rc/imon.c > @@ -153,6 +153,24 @@ struct imon_context { > const struct imon_usb_dev_descr *dev_descr; > /* device description with key */ > /* table for front panels */ > + /* > + * Fields for deferring free_imon_context(). > + * > + * Since reference to "struct imon_context" is stored into > + * "struct file_operations"->private_data, we need to remember > + * how many file descriptors might access this "struct imon_context". > + */ > + refcount_t users; Are you sure this is going to work properly? How do you handle userspace passing around file descriptors to other processes? You really should not ever have to count this. > + /* > + * Use a flag for telling display_open()/vfd_write()/lcd_write() that > + * imon_disconnect() was already called. > + */ > + bool disconnected; > + /* > + * We need to wait for RCU grace period in order to allow > + * display_open() to safely check ->disconnected and increment ->users. > + */ > + struct rcu_head rcu; > }; > > #define TOUCH_TIMEOUT (HZ/30) > @@ -160,18 +178,18 @@ struct imon_context { > /* vfd character device file operations */ > static const struct file_operations vfd_fops = { > .owner = THIS_MODULE, > - .open = &display_open, > - .write = &vfd_write, > - .release = &display_close, > + .open = display_open, > + .write = vfd_write, > + .release = display_close, > .llseek = noop_llseek, > }; > > /* lcd character device file operations */ > static const struct file_operations lcd_fops = { > .owner = THIS_MODULE, > - .open = &display_open, > - .write = &lcd_write, > - .release = &display_close, > + .open = display_open, > + .write = lcd_write, > + .release = display_close, > .llseek = noop_llseek, > }; > > @@ -439,9 +457,6 @@ static struct usb_driver imon_driver = { > .id_table = imon_usb_id_table, > }; > > -/* to prevent races between open() and disconnect(), probing, etc */ > -static DEFINE_MUTEX(driver_lock); > - > /* Module bookkeeping bits */ > MODULE_AUTHOR(MOD_AUTHOR); > MODULE_DESCRIPTION(MOD_DESC); > @@ -481,9 +496,11 @@ static void free_imon_context(struct imon_context *ictx) > struct device *dev = ictx->dev; > > usb_free_urb(ictx->tx_urb); > + WARN_ON(ictx->dev_present_intf0); > usb_free_urb(ictx->rx_urb_intf0); > + WARN_ON(ictx->dev_present_intf1); > usb_free_urb(ictx->rx_urb_intf1); > - kfree(ictx); > + kfree_rcu(ictx, rcu); > > dev_dbg(dev, "%s: iMON context freed\n", __func__); > } > @@ -499,9 +516,6 @@ static int display_open(struct inode *inode, struct file *file) > int subminor; > int retval = 0; > > - /* prevent races with disconnect */ > - mutex_lock(&driver_lock); > - > subminor = iminor(inode); > interface = usb_find_interface(&imon_driver, subminor); > if (!interface) { > @@ -509,13 +523,16 @@ static int display_open(struct inode *inode, struct file *file) > retval = -ENODEV; > goto exit; > } > - ictx = usb_get_intfdata(interface); > > - if (!ictx) { > + rcu_read_lock(); > + ictx = usb_get_intfdata(interface); > + if (!ictx || ictx->disconnected || !refcount_inc_not_zero(&ictx->users)) { > + rcu_read_unlock(); > pr_err("no context found for minor %d\n", subminor); > retval = -ENODEV; > goto exit; > } > + rcu_read_unlock(); > > mutex_lock(&ictx->lock); > > @@ -533,8 +550,10 @@ static int display_open(struct inode *inode, struct file *file) > > mutex_unlock(&ictx->lock); > > + if (retval && refcount_dec_and_test(&ictx->users)) > + free_imon_context(ictx); > + > exit: > - mutex_unlock(&driver_lock); > return retval; > } > > @@ -544,16 +563,9 @@ static int display_open(struct inode *inode, struct file *file) > */ > static int display_close(struct inode *inode, struct file *file) > { > - struct imon_context *ictx = NULL; > + struct imon_context *ictx = file->private_data; > int retval = 0; > > - ictx = file->private_data; > - > - if (!ictx) { > - pr_err("no context for device\n"); > - return -ENODEV; > - } > - > mutex_lock(&ictx->lock); > > if (!ictx->display_supported) { > @@ -568,6 +580,8 @@ static int display_close(struct inode *inode, struct file *file) > } > > mutex_unlock(&ictx->lock); > + if (refcount_dec_and_test(&ictx->users)) > + free_imon_context(ictx); Why not just put a kref into your larger structure? I think trying to count users of open/close is never going to work, just allow the normal open/close logic to work properly and track your data structure based on reference counts like it should be doing already. thanks, greg k-h
