From: Rasmus Villemoes <linux@rasmusvillemoes.dk>
To: Sudip Mukherjee <sudipm.mukherjee@gmail.com>
Cc: Mike Snitzer <snitzer@redhat.com>, Alasdair Kergon <agk@redhat.com>,
        Shaohua Li <shli@kernel.org>, linux-kernel@vger.kernel.org,
        dm-devel@redhat.com, linux-raid@vger.kernel.org
Subject: Re: dm: ioctl: use kvfree
Organization: D03
References: <1460387677-19948-1-git-send-email-sudipm.mukherjee@gmail.com>
	<20160411151714.GB25938@redhat.com> <570BCC51.50004@gmail.com>
Date: Mon, 11 Apr 2016 21:52:18 +0200
In-Reply-To: <570BCC51.50004@gmail.com> (Sudip Mukherjee's message of "Mon, 11
	Apr 2016 21:39:53 +0530")
Message-ID: <87wpo3q5xp.fsf@rasmusvillemoes.dk>
User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/24.3 (gnu/linux)
MIME-Version: 1.0
Content-Type: text/plain
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1262
Lines: 34

On Mon, Apr 11 2016, Sudip Mukherjee <sudipm.mukherjee@gmail.com> wrote:

> On Monday 11 April 2016 08:47 PM, Mike Snitzer wrote:
>> On Mon, Apr 11 2016 at 11:14am -0400,
>> Sudip Mukherjee <sudipm.mukherjee@gmail.com> wrote:
>>
>>> We can use kvfree() instead of calling kfree() and vfree() based on
>>> if-else and param_flags. kvfree() will check the type of address and
>>> will call the respective function to free it.
>>> Additionally we can also remove the use of DM_PARAMS_KMALLOC and
>>> DM_PARAMS_VMALLOC.
>>>
>>> Signed-off-by: Sudip Mukherjee <sudip.mukherjee@codethink.co.uk>
>>
>> Have you actually tested htis?  Last time I looked to do this it exposed
>> crashes.  I don't have time to dig into this again right now but this is
>> _not_ as simple as this patch implies.
>>
>
> No, it was just build tested. Is it possible to test it in qemu or kvm?
>

No need to test it, just read copy_params() and its caller,
ctl_ioctl(). The latter passes a stack buffer as param_kernel, and
copy_params() does

        if (ioctl_flags & IOCTL_FLAGS_NO_PARAMS) {
                dmi = param_kernel;

with dmi later returned via *param. So it is entirely possible that
free_params ends up calling neither kfree or vfree, since there's
nothing to free.

Rasmus