At 2022-09-16 23:36:47, "Alan Stern" <stern@xxxxxxxxxxxxxxxxxxx> wrote: >On Fri, Sep 16, 2022 at 11:20:23PM +0800, Liang He wrote: >> >> >> At 2022-09-16 23:04:02, "Alan Stern" <stern@xxxxxxxxxxxxxxxxxxx> wrote: >> >On Fri, Sep 16, 2022 at 03:35:52PM +0800, Liang He wrote: >> >> In usb_console_setup(), if we goto error_get_interface and the >> >> usb_serial_put() may finally call kfree(serial). However, the next >> >> line will call 'mutex_unlock(&serial->disc_mutex)' which can cause >> >> a potential UAF bug. >> > >> >Why not just move the mutex_unlock() call up one line, before the >> >usb_serial_put()? >> > >> >> Fixes: 7bd032dc2793 ("USB serial: update the console driver") >> >> Signed-off-by: Liang He <windhl@xxxxxxx> >> >> --- >> >> >> >> I don't know if the refcount can be zero here, so if it cannot be zero, >> >> this code is safe and please ignore my patch. >> >> >> >> drivers/usb/serial/console.c | 5 +++-- >> >> 1 file changed, 3 insertions(+), 2 deletions(-) >> >> >> >> diff --git a/drivers/usb/serial/console.c b/drivers/usb/serial/console.c >> >> index b97aa40ca4d1..21ac2dd6baca 100644 >> >> --- a/drivers/usb/serial/console.c >> >> +++ b/drivers/usb/serial/console.c >> >> @@ -62,6 +62,7 @@ static int usb_console_setup(struct console *co, char *options) >> >> int cflag = CREAD | HUPCL | CLOCAL; >> >> char *s; >> >> struct usb_serial *serial; >> >> + struct mutex *s_mutex; >> >> struct usb_serial_port *port; >> >> int retval; >> >> struct tty_struct *tty = NULL; >> >> @@ -116,7 +117,7 @@ static int usb_console_setup(struct console *co, char *options) >> >> return -ENODEV; >> >> } >> >> serial = port->serial; >> >> - >> >> + s_mutex = &serial->disc_mutex; >> >> retval = usb_autopm_get_interface(serial->interface); >> >> if (retval) >> >> goto error_get_interface; >> >> @@ -190,7 +191,7 @@ static int usb_console_setup(struct console *co, char *options) >> >> usb_autopm_put_interface(serial->interface); >> >> error_get_interface: >> >> usb_serial_put(serial); >> >> - mutex_unlock(&serial->disc_mutex); >> >> + mutex_unlock(s_mutex); >> > >> >If the old code was unsafe then so is this, because s_mutex points to a >> >mutex that is embedded within the serial structure. If the structure >> >was deallocated by usb_serial_put() then so was the mutex. >> > >> >Alan Stern >> > >> >> return retval; >> >> } >> >> >> >> -- >> >> 2.25.1 >> >> >> >> Hi, Alan Stern, >> >> Thanks for your review and this patch is indeed wrong! >> >> But I am not sure if we can safely move the usb_serial_put() >> out of mutex_unlock(). >> >> If it is safe, I can give a new version of patch very soon. >> >> Can you help me confirm if it is safe? > >I cannot. You need to ask Johan (the USB-serial maintainer). > >Alan Stern Still thanks! And from a recent similar commit, I think we can move mutex_unlock above the usb_serial_put(): https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?h=v6.0-rc5&id=6c53b45c71b4920b5e62f0ea8079a1da382b9434 Johan, please confirm if this can be accepted. Thanks, Liang