tty: Fix timeout on pty set ldisc
authorPeter Hurley <peter@hurleysoftware.com>
Wed, 5 Nov 2014 17:13:08 +0000 (12:13 -0500)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 6 Nov 2014 02:50:43 +0000 (18:50 -0800)
When changing the ldisc on one end of a pty pair, there may be
waiting readers/writers on the other end which may not exit from
the ldisc i/o loop, preventing tty_ldisc_lock_pair_timeout() from
acquiring the other side's ldisc lock.

Only acquire this side's ldisc lock; although this will no longer
prevent the other side from writing new input, that input will not
be processed until after the ldisc change completes. This has no
effect on normal ttys; new input from the driver was never disabled.

Remove tty_ldisc_enable_pair().

Reviewed-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Peter Hurley <peter@hurleysoftware.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/tty/tty_ldisc.c

index 1dbe27824220266e1dd23b71bab7f6edadae32d1..6368dd95e13740ffd52f4a54e65ed11179ca41b0 100644 (file)
@@ -393,16 +393,6 @@ static void __lockfunc tty_ldisc_unlock_pair(struct tty_struct *tty,
                __tty_ldisc_unlock(tty2);
 }
 
-static void __lockfunc tty_ldisc_enable_pair(struct tty_struct *tty,
-                                            struct tty_struct *tty2)
-{
-       clear_bit(TTY_LDISC_HALTED, &tty->flags);
-       if (tty2)
-               clear_bit(TTY_LDISC_HALTED, &tty2->flags);
-
-       tty_ldisc_unlock_pair(tty, tty2);
-}
-
 /**
  *     tty_ldisc_flush -       flush line discipline queue
  *     @tty: tty
@@ -535,14 +525,13 @@ int tty_set_ldisc(struct tty_struct *tty, int ldisc)
 {
        int retval;
        struct tty_ldisc *old_ldisc, *new_ldisc;
-       struct tty_struct *o_tty = tty->link;
 
        new_ldisc = tty_ldisc_get(tty, ldisc);
        if (IS_ERR(new_ldisc))
                return PTR_ERR(new_ldisc);
 
        tty_lock(tty);
-       retval = tty_ldisc_lock_pair_timeout(tty, o_tty, 5 * HZ);
+       retval = tty_ldisc_lock(tty, 5 * HZ);
        if (retval) {
                tty_ldisc_put(new_ldisc);
                tty_unlock(tty);
@@ -554,7 +543,7 @@ int tty_set_ldisc(struct tty_struct *tty, int ldisc)
         */
 
        if (tty->ldisc->ops->num == ldisc) {
-               tty_ldisc_enable_pair(tty, o_tty);
+               tty_ldisc_unlock(tty);
                tty_ldisc_put(new_ldisc);
                tty_unlock(tty);
                return 0;
@@ -565,7 +554,7 @@ int tty_set_ldisc(struct tty_struct *tty, int ldisc)
        if (test_bit(TTY_HUPPED, &tty->flags)) {
                /* We were raced by the hangup method. It will have stomped
                   the ldisc data and closed the ldisc down */
-               tty_ldisc_enable_pair(tty, o_tty);
+               tty_ldisc_unlock(tty);
                tty_ldisc_put(new_ldisc);
                tty_unlock(tty);
                return -EIO;
@@ -599,13 +588,11 @@ int tty_set_ldisc(struct tty_struct *tty, int ldisc)
        /*
         *      Allow ldisc referencing to occur again
         */
-       tty_ldisc_enable_pair(tty, o_tty);
+       tty_ldisc_unlock(tty);
 
        /* Restart the work queue in case no characters kick it off. Safe if
           already running */
        schedule_work(&tty->port->buf.work);
-       if (o_tty)
-               schedule_work(&o_tty->port->buf.work);
 
        tty_unlock(tty);
        return retval;