Message-ID: <497DEC67.8030709@steeleye.com>
Date: Mon, 26 Jan 2009 12:01:27 -0500
From: Paul Clements <paul.clements@steeleye.com>
User-Agent: Swiftdove 2.0.0.9 (X11/20071116)
MIME-Version: 1.0
To: Pavel Machek <pavel@suse.cz>
CC: kernel list <linux-kernel@vger.kernel.org>, Andrew Morton <akpm@osdl.org>
Subject: Re: nbd: add locking to nbd_ioctl
References: <20090116115512.GA10771@elf.ucw.cz> <4970A696.9070307@steeleye.com> <20090116153603.GD2022@elf.ucw.cz> <4970B59A.9090807@steeleye.com> <20090119095459.GA11187@atrey.karlin.mff.cuni.cz> <497494BB.3080800@steeleye.com> <20090126164959.GB4145@atrey.karlin.mff.cuni.cz>
In-Reply-To: <20090126164959.GB4145@atrey.karlin.mff.cuni.cz>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1790
Lines: 38

Pavel Machek wrote:
>> Pavel Machek wrote:
>>>> Pavel Machek wrote:
>>>>> On Fri 2009-01-16 10:24:06, Paul Clements wrote:
>>>> lo->sock is only modified under tx_lock (except for SET_SOCK, where the 
>>>> device is being initialized, in which case it's impossible for any other 
>>>> thread to be accessing the device)
>>> Well, unless the user is evil or confused? :-).
>> Even in that case, you're just going to get EBUSY. Nothing bad will 
>> happen. SET_SOCK checks for lo->file, so it cannot be called on an 
>> active nbd device.
>>
>>
>>>> As for other fields, I assume you're talking about blksize, et al. 
>>>> Taking tx_lock doesn't prevent you from screwing yourself if you modify 
>>>> those while the device is active. You'd need to disallow those ioctls 
>>>> when the device is active (check lo->file). Again, this is only going to 
>>>> happen if you really misuse the ioctls.
>>> Ok, I'll take a look at the missing checks. I'd really like to make
>>> this "stable" -- no amount of misuse should crash the kernel.
>> Just to summarize, I don't think we need to hold tx_lock around the 
>> entirety of nbd_ioctl. We do need one extra tx_lock around xmit_timeout 
>> and we do need to check for lo->file and return EBUSY in all of the 
>> SET_*SIZE* ioctls.
> 
> I could do that but it would be a bit too complex, and still rely on
> big kernel lock. Would you agree to patch that added tx_lock around
> all of it, and moved ioctl to unlocked ioctl?

OK, I can buy the complexity argument. Your patch sounds fine to me.

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