Re: [PATCH] kdesu broken

[OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>]

Alan Cox <alan@lxorguk.ukuu.org.uk> writes:

> commit 70325fd1d4341896c17b6f1f1965370b5258d0b1
> Author: Alan Cox <alan@linux.intel.com>
> Date:   Mon Jul 27 14:18:52 2009 +0100
>
>     pty: ensure writes hit the reader before close
>     
>     Implement TTY_EOF/EOFPENDING flags so that we can propogate the close of the
>     pty through the buffering correctly. The new flag state is locked but the
>     tty buffer lock as it relates to buffers, and also because the buffer
>     lock is already held in the hot path.
>     
>     Signed-off-by: Alan Cox <alan@linux.intel.com>

I also tested this patch, and it seems to work in some case. However, in
some case, n_tty_read() didn't return -EIO for read() of master.

> +	spin_lock_irqsave(&pair->buf.lock, flags);
> +	pair->packet = 0;
> +	/* Indicate that the other end is now closed, set the
> +	   ENDPENDING marker so that the true end can be processed by

This seems typo s/ENDPENDING/EOFPENDING/


> @@ -47,13 +50,22 @@ static void pty_close(struct tty_struct *tty, struct file *filp)

[...]

> +	set_bit(TTY_EOFPENDING, &pair->flags);
> +	set_bit(TTY_OTHER_CLOSED, &pair->flags);
> +	spin_unlock_irqrestore(&pair->buf.lock, flags);

	tty_flip_buffer_push() or something?

> +	wake_up_interruptible(&pair->read_wait);
> +	wake_up_interruptible(&pair->write_wait);

It seems, this sets TTY_EOFPENDING, but if flush_to_ldisc() was done
already here, anybody doesn't set TTY_EOF for master.

Thanks.
-- 
OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>