On Sat, Feb 20, 2021 at 12:08:32AM +0900, Tetsuo Handa wrote:

syzbot is reporting an ERR_PTR(-EINTR) pointer dereference at
vhci_shutdown_connection() [1], for kthread_create() became killable due
to commit 786235eeba0e1e85 ("kthread: make kthread_create() killable").

When SIGKILLed while attach_store() is calling kthread_get_run(),
ERR_PTR(-EINTR) is stored into vdev->ud.tcp_{rx,tx}, and then
kthread_stop_put() is called on vdev->ud.tcp_{rx,tx} from
vhci_shutdown_connection() because vdev->ud.tcp_{rx,tx} != NULL.

Prior to commit 9720b4bc76a83807 ("staging/usbip: convert to kthread"),
"current" pointer is assigned to vdev->ud.tcp_{rx,tx} by usbip_thread()
kernel thread, and hence vdev->ud.tcp_{rx,tx} != NULL means a valid task
pointer. However, this patch does not make kthread_get_run() return NULL
when kthread_create() failed, for this patch removes kthread_get_run() in
order to fix the other bug described below.

syzbot is also reporting a NULL pointer dereference at sock_sendmsg() [2],
for lack of serialization between attach_store() and event_handler()
causes vhci_shutdown_connection() to observe vdev->ud.tcp_tx == NULL while
vdev->ud.tcp_socket != NULL. Please read the reference link for details of
this race window.

Therefore, this patch does the following things in order to fix reported
bugs and other possible bugs.

(1) Handle kthread_create() failure (which fixes [1]) by grouping socket
    lookup, kthread_create() and get_task_struct() into
    usbip_prepare_threads() function.

(2) Serialize usbip_sockfd_store(), detach_store(), attach_store() and
    ud->eh_ops.{shutdown,reset,unusable}() operations using
    usbip_event_mutex mutex (which fixes [2]). Introducing such large
    mutex should be safe because ud->tcp_{tx,rx} must not wait for
    event_handler() to flush because event_handler() is processed by a
    singlethreaded workqueue.

(3) Add SOCK_STREAM check into usbip_prepare_threads(), for current code
    is not verifying that a file descriptor passed is actually a stream
    socket. If the file descriptor passed was a SOCK_DGRAM socket,
    sock_recvmsg() can't detect end of stream.

(4) Don't perform ud->tcp_socket = NULL in vhci_device_reset().
    Since ud->tcp_{tx,rx} depend on ud->tcp_socket != NULL whereas
    ud->tcp_socket and ud->tcp_{tx,rx} are assigned at the same time,
    it is never safe to reset ud->tcp_socket from vhci_device_reset()
    without calling kthread_stop_put() from vhci_shutdown_connection().

(5) usbip_sockfd_store() must perform

      if ({sdev,udc}->ud.status != SDEV_ST_AVAILABLE) {
        /* misc assignments for attach operation */
        {sdev,udc}->ud.status = SDEV_ST_USED;
      }

    atomically, or multiple ud->tcp_{tx,rx} are created (which will later
    cause a crash like [2]) and refcount on ud->tcp_socket is leaked when
    usbip_sockfd_store() is concurrently called.

[1] https://syzkaller.appspot.com/bug?extid=a93fba6d384346a761e3
[2] https://syzkaller.appspot.com/bug?extid=95ce4b142579611ef0a9

Reported-and-tested-by: syzbot <redacted>
Reported-by: syzbot <redacted>
Reported-by: syzbot <redacted>
References: https://lkml.kernel.org/r/676d4518-0faa-9fab-15db-0db8d216d7fb@i-love.sakura.ne.jp
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Fixes: 9720b4bc76a83807 ("staging/usbip: convert to kthread")
---
 drivers/usb/usbip/stub_dev.c     | 56 ++++++++++++++++++--------------
 drivers/usb/usbip/usbip_common.c | 55 +++++++++++++++++++++++++++++++
 drivers/usb/usbip/usbip_common.h | 25 +++++++-------
 drivers/usb/usbip/usbip_event.c  | 15 +++++++++
 drivers/usb/usbip/vhci_hcd.c     |  6 ----
 drivers/usb/usbip/vhci_sysfs.c   | 50 ++++++++++++++++++++--------
 drivers/usb/usbip/vudc_sysfs.c   | 50 ++++++++++++++++------------
 7 files changed, 181 insertions(+), 76 deletions(-)

What changed from v1?  Why isn't that info below the --- line?

Please do a v3 with that fixed up.

thanks,

greg k-h