答复: [PATCH][v2] tty: fix race between flush_to_ldisc and tty_open

[Li,Rongqing]

> -----邮件原件-----
> 发件人: Kohli, Gaurav [mailto:gko...@codeaurora.org]
> 发送时间: 2019年1月11日 21:57
> 收件人: Li,Rongqing <lirongq...@baidu.com>; linux-kernel@vger.kernel.org;
> jsl...@suse.com; gre...@linuxfoundation.org; a...@linux.intel.com
> 主题: Re: [PATCH][v2] tty: fix race between flush_to_ldisc and tty_open
> 
> Hi,
> 
> it don't seems to be good idea to put lock on one function and unlock in some
> other function. If in future some one has to call tty_init_dev, how he can 
> track
> the unlocking as well of ldisc lock.
> 
> Regards
> Gaurav
> 

This similar condition has existed for a long time, tty_unlock(tty) must be 
called in
some other functions who call tty_init_dev, since tty_init_dev hold tty_lock, 
and does
not release it

so I think it is user responsibility to fully understand tty_init_dev; and I 
can add some 
comments for tty_init_dev if this patch can be acceptable;


or a workaround like below:
diff --git a/drivers/tty/tty_buffer.c b/drivers/tty/tty_buffer.c
index a42a028a9d4e..2f5ad256b6ad 100644
--- a/drivers/tty/tty_buffer.c
+++ b/drivers/tty/tty_buffer.c
@@ -425,7 +425,7 @@ static void flush_to_ldisc(struct work_struct *work)
        struct tty_ldisc *disc;
 
        tty = port->itty;
-       if (tty == NULL)
+       if (tty == NULL || tty->driver_data == NULL)
                return;
 
        disc = tty_ldisc_ref(tty);

thanks

-RongQing