From: Linyu Yuan <quic_linyyuan@xxxxxxxxxxx> commit 0698f0209d8032e8869525aeb68f65ee7fde12ad upstream. In ffs_epfile_io(), when read/write data in blocking mode, it will wait the completion in interruptible mode, if task receive a signal, it will terminate the wait, at same time, if function unbind occurs, ffs_func_unbind() will kfree all eps, ffs_epfile_io() still try to dequeue request by dereferencing ep which may become invalid. Fix it by add ep spinlock and will not dereference ep if it is not valid. Cc: <stable@xxxxxxxxxxxxxxx> # 5.15 Reported-by: Michael Wu <michael@xxxxxxxxxxxxxxxxx> Tested-by: Michael Wu <michael@xxxxxxxxxxxxxxxxx> Reviewed-by: John Keeping <john@xxxxxxxxxxxx> Signed-off-by: Linyu Yuan <quic_linyyuan@xxxxxxxxxxx> Link: https://lore.kernel.org/r/1654863478-26228-3-git-send-email-quic_linyyuan@xxxxxxxxxxx Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> --- drivers/usb/gadget/function/f_fs.c | 6 ++++++ 1 file changed, 6 insertions(+) --- a/drivers/usb/gadget/function/f_fs.c +++ b/drivers/usb/gadget/function/f_fs.c @@ -1080,6 +1080,11 @@ static ssize_t ffs_epfile_io(struct file spin_unlock_irq(&epfile->ffs->eps_lock); if (wait_for_completion_interruptible(&io_data->done)) { + spin_lock_irq(&epfile->ffs->eps_lock); + if (epfile->ep != ep) { + ret = -ESHUTDOWN; + goto error_lock; + } /* * To avoid race condition with ffs_epfile_io_complete, * dequeue the request first then check @@ -1087,6 +1092,7 @@ static ssize_t ffs_epfile_io(struct file * condition with req->complete callback. */ usb_ep_dequeue(ep->ep, req); + spin_unlock_irq(&epfile->ffs->eps_lock); wait_for_completion(&io_data->done); interrupted = io_data->status < 0; }
