linux-kernel - Re: [PATCH] Xen: ARM: Zero reserved fields of xatp before making hypervisor call

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 for Android: free password hash cracker in your pocket

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

Message-ID: <5858BB54.5090106@codeaurora.org>
Date:   Mon, 19 Dec 2016 23:02:12 -0600
From:   Jiandi An <anjiandi@...eaurora.org>
To:     Stefano Stabellini <sstabellini@...nel.org>
CC:     Juergen Gross <jgross@...e.com>, boris.ostrovsky@...cle.com,
        xen-devel@...ts.xenproject.org, linux-kernel@...r.kernel.org,
        julien.grall@....com, shankerd@...eaurora.org,
        shannon.zhao@...aro.org
Subject: Re: [PATCH] Xen: ARM: Zero reserved fields of xatp before making
 hypervisor call

On 12/19/16 12:49, Stefano Stabellini wrote:
> On Mon, 19 Dec 2016, Juergen Gross wrote:
>> On 19/12/16 03:56, Jiandi An wrote:
>>> Ensure all reserved fields of xatp are zero before making hypervisor
>>> call to XEN in xen_map_device_mmio().  xenmem_add_to_physmap_one() in
>>> XEN fails the mapping request if extra.res reserved field in xatp is
>>> not zero for XENMAPSPACE_dev_mmio request.
>>>
>>> Signed-off-by: Jiandi An <anjiandi@...eaurora.org>
>>> ---
>>>  drivers/xen/arm-device.c | 3 +++
>>>  1 file changed, 3 insertions(+)
>>>
>>> diff --git a/drivers/xen/arm-device.c b/drivers/xen/arm-device.c
>>> index 778acf8..208273b 100644
>>> --- a/drivers/xen/arm-device.c
>>> +++ b/drivers/xen/arm-device.c
>>> @@ -87,6 +87,9 @@ static int xen_map_device_mmio(const struct resource *resources,
>>>  			idxs[j] = XEN_PFN_DOWN(r->start) + j;
>>>  		}
>>>  
>>> +		/* Ensure reserved fields are set to zero */
>>> +		memset(&xatp, 0, sizeof(xatp));
>>> +
>>>  		xatp.domid = DOMID_SELF;
>>>  		xatp.size = nr;
>>>  		xatp.space = XENMAPSPACE_dev_mmio;
>>
>> Instead of setting xatp to 0 in each iteration I'd prefer a static
>> initialization of .domid and .space:
>>
>> 	struct xen_add_to_physmap_range xatp = {
>> 		.domid = DOMID_SELF,
>> 		.space = XENMAPSPACE_dev_mmio
>> 	};
> 
> +1
> 

Hi Juergen,

Thanks for you comments.  xatp is passed to XEN via the hypervisor call in each loop.
XEN touches xatp and returns it back.  For example XEN returns error of underlying mapping call in the err[] array in xatp. (The err[] is not checked after the hypervisor call returns and it's a bug to be addressed in a separate patch)  XEN could theoretically corrupt xatp when it's returned.  And the loop would go on to the next iteration passing in whatever that's in xatp returned by the previous hypervisor call.  Harder to debug in my opinion if xatp get corrupted by XEN somehow when a bug is introduced in XEN.  At first I put the memset of xatp at the beginning outside of the loop.  But I thought it's better to initialize xatp that's passed in each time a hypervisor call is made so we know exactly we set going into the hypervisor call.

-- 
Jiandi An
Qualcomm Datacenter Technologies, Inc. as an affiliate of Qualcomm Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a Linux Foundation Collaborative Project.