Message-ID: <20120601144040.GA24774@build.ihdev.net>
Date:	Fri, 1 Jun 2012 10:40:40 -0400
From:	Sergey Lapin <slapin@...fans.org>
To:	Alan Cox <alan@...rguk.ukuu.org.uk>
Cc:	"David S. Miller" <davem@...emloft.net>,
	linux-kernel@...r.kernel.org, linux-x25@...r.kernel.org
Subject: Re: [PATCH 2/2] lapb-nl: Added driver

On Fri, Jun 01, 2012 at 02:52:01PM +0100, Alan Cox wrote:
> > +	  This driver allows UDP or NETLINK application to transfer data over
> > +	  serial port using LAPB for delivery guarantee. This allows use LAPB
> > +	  features without deploying full X.25 stack and use equipment, which
> > +	  implements LAPB, but not X.25 framing, using custom protocol on top
> > +	  of LAPB.
> 
> The netlink thing is a bit eccentric. I think I'd much rather see either
> it using raw sockets or an AF_LAPB or similar (AF_X25, SOCK_RAW ?)
Well, a protocol is very simple, I didn't want to make it into
whole socket thing...

> 
> > +	dev_info(&sl->netdev->dev, "rx\n");
> 
> Debug wants to go before submission or be turned down !
Sorry about that, forgot to change this (and the following) to dev_dbg
> 
> 
> > +int lapb_nl_ldisc_transmit(struct net_device *dev, u8 *data, size_t len)
> > +{
> > +	struct lapb_nl_ldisc *sl;
> > +	int actual, count;
> > +	BUG_ON(!dev);
> 
> How can that occur ?

This whole thing is ported from older (pre 2.6.30) kernels,
that's from these days artefacts (including various checks),
will remove. I even don't remember why these were added...
> 
> > +	sl = get_lapb_data(dev);
> > +	BUG_ON(!sl);
> > +	spin_lock(&sl->lock);
> > +	if (sl->tty == NULL) {
> > +		spin_unlock(&sl->lock);
> > +		dev_err(&dev->dev, "lapb: tbusy drop\n");
> 
> What stops this changing during the transmit call ? You probably want to
> grab a kref and drop it when the ldisc is closed. That would also negate
> this meaningless check
Ditto.
> 
> 
> > +		return -ENOTTY;
> > +	}
> > +	count = lapb_esc(data, sl->xbuff, len);
> > +
> > +	/* Order of next two lines is *very* important.^S */
> > +	set_bit(TTY_DO_WRITE_WAKEUP, &sl->tty->flags);
> > +	actual = sl->tty->ops->write(sl->tty, sl->xbuff, count);
> > +	dev_info(&sl->netdev->dev, "tx\n");
> 
> Again all the debug wants a tidy
> > +
> > +	print_hex_dump(KERN_INFO, "ldisc out:",
> > +			DUMP_PREFIX_OFFSET, 16, 2, sl->xbuff,
> > +			count, 0);
> > +	sl->xleft = count - actual;
> > +	sl->xhead = sl->xbuff + actual;
> > +	/* VSV */
> > +	clear_bit(SLF_OUTWAIT, &sl->flags);	/* reset outfill flag */
> 
> You seem to have copied this flag but not use it ?
Sorry, another old artefact, will remove.

> 
> 
> > +int lapb_nl_ldisc_start(struct net_device *dev)
> > +{
> > +	struct lapb_nl_ldisc *sl;
> > +	unsigned long len;
> > +	BUG_ON(!dev);
> > +	sl = get_lapb_data(dev);
> > +	BUG_ON(!sl);
> > +	if (sl->tty == NULL)
> > +		return -ENODEV;
> 
> Again the sl->tty locking needs sorting (I think this may well be true of
> slip and the others too)

Could you please explain a bit more on this?

> 
> 
> > +static int lapb_ldisc_open(struct tty_struct *tty)
> > +{
> > +	struct net_device *netdev;
> > +	struct lapb_nl_ldisc *sl;
> > +	int ret;
> > +	BUG_ON(!tty);
> 
> Can't happen

Will remove this one.

> 
> > +	sl = kzalloc(sizeof(struct lapb_nl_ldisc), GFP_KERNEL);
> 
> What if the allocation fails
> 
> > +	sl->tty = tty;
> 
> kref
> 
> > +	sl->netdev = netdev;
> > +	sl->magic = LAPB_LDISC_MAGIC;
> > +	tty->disc_data = sl;
> > +	tty->receive_room = 65536;
> > +	tty_driver_flush_buffer(tty);
> > +	tty_ldisc_flush(tty);
> 
> Why flush ?
Problems occured a long time ago, will remove and check.
> 
> 
> > +static int lapb_ldisc_ioctl(struct tty_struct *tty, struct file *file,
> > +			    unsigned int cmd, unsigned long arg)
> > +{
> > +	struct lapb_nl_ldisc *sl = tty->disc_data;
> > +	BUG_ON(!sl);
> > +	dev_dbg(&sl->netdev->dev, "LAPB ldisc ioctl %04x\n", cmd);
> > +
> > +	/* First make sure we're connected. */
> > +	if (!sl || sl->magic != LAPB_LDISC_MAGIC)
> > +		return -EINVAL;
> 
> What locks this check ?
> 
> [Its safe because the tty mid layer has an ldisc ref of its own as far as
> I can see but the check is still bogus and does nothing)

Well, as I see, this whole bunch of checks needs cleaning.

> 
> > +	if (!sl->netdev)
> > +		return -EBUSY;
> > +
> > +	switch (cmd) {
> > +	case SIOCGIFNAME:
> > +		if (copy_to_user((void __user *)arg,
> > +			(void *)(sl->netdev->name),
> > +			strlen(sl->netdev->name) + 1))
> > +				return -EFAULT;
> > +		return 0;
> > +
> > +	default:
> > +		return tty_mode_ioctl(tty, file, cmd, arg);
> > +	}
> > +}
> > +
> > +#ifdef CONFIG_COMPAT
> > +static long lapb_ldisc_compat_ioctl(struct tty_struct *tty, struct file *file,
> > +				    unsigned int cmd, unsigned long arg)
> > +{
> > +	switch (cmd) {
> > +	case SIOCGX25PARMS:
> > +	case SIOCSX25PARMS:
> > +		return lapb_ldisc_ioctl(tty, file, cmd,
> > +				       (unsigned long)compat_ptr(arg));
> 
> Which will in turn call the default so this seems surplus ?
Will remove COMPAT support or rewrite.

> 
> 
> 
> The big question to me is the API for it all, which looks marginally
> insane. Is there a URL to the userspace for this lot that might help
> folks work out where your API is actually sensible or if not what it
> ought to look like.
> 
> Alan
Well, API is quite simple, It is used by python code via specially crafted
C library, will clean it up proprietary code and put on github soon.

The whole thing is data packat transfer. Application is also notified when
packet is delivered. That's it, no fancy stuff. There is also UDP code, but it is
weird mess, and I think if it is to live or be removed. What do you think?

Thanks a lot for review,
S.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists