linux-kernel - Re: tty_lock held during transmit wait in close: still unresolved

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Fri, 27 May 2011 13:38:54 +0200
From:	Arnd Bergmann <arnd@...db.de>
To:	Andreas Bombe <aeb@...ian.org>
Cc:	Alan Cox <alan@...rguk.ukuu.org.uk>, linux-kernel@...r.kernel.org,
	Greg KH <greg@...ah.com>
Subject: Re: tty_lock held during transmit wait in close: still unresolved

On Friday 27 May 2011, Andreas Bombe wrote:
> > At any point you can show the code sleeps you can drop and retake the
> > tty mutex either side of it, so you should be able to do that in the
> > close timeout case. You may need to think about the order of locking with
> > the port mutex but I suspect you can drop and retake both there.
> 
> …basically emulating the BKL semantics? Sounds more doable. I'll look
> into it.

If I understand it correctly, the problem is the msleep_interruptible()
in __uart_wait_until_sent(), right?

Note that this function may be called with or without the port mutex
held, depending on whether the caller is uart_close or uart_wait_until_sent.
The tricky part here will be making sure that you hold neither the
port mutex nor the tty_mutex while sleeping, and to always retake them
in the correct order (tty_mutex before port mutex).

My mistake here must have been that I assumed the timeout was relatively
short to not hurt when holding a mutex, since we already hold the port
mutex. I expected the wait time to be a fraction of a second as in the time
that it takes to send a few remaining characters, which would be acceptable,
unlike the 30 second sleep that you are seeing.

> Of course that means it has to be done individually in all drivers.

Right. Fortunately, we have now reduced the number of drivers a bit, by
moving some of them to staging or completely out of the kernel.

Some drivers call their wait_until_sent function directly from their
close function, some call it through tty_wait_until_sent, and some
actually do both.

Further, some of the drivers have a rather ugly part in them where we take
tty_lock() conditionally in wait_until_sent(), depending on whether the
current thread already holds it (i.e. when coming from ->close, not when
coming from ioctl).

	Arnd
--
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/