Re: [PATCH] kvm_s390_vm_start_migration: check dirty_bitmap before using it as target for memset()

[David Hildenbrand]

On 09.09.19 18:17, Igor Mammedov wrote:
> On Mon, 9 Sep 2019 17:47:37 +0200
> David Hildenbrand <david@xxxxxxxxxx> wrote:
> 
>> On 09.09.19 16:55, Igor Mammedov wrote:
>>> If userspace doesn't set KVM_MEM_LOG_DIRTY_PAGES on memslot before calling
>>> kvm_s390_vm_start_migration(), kernel will oops with:
>>>   
>>
>> We usually have the subject in a "KVM: s390: ..." format. Like
>>
>> "KVM: s390: check dirty_bitmap before using it as target for memset()"
>>
>>>   Unable to handle kernel pointer dereference in virtual kernel address space
>>>   Failing address: 0000000000000000 TEID: 0000000000000483
>>>   Fault in home space mode while using kernel ASCE.
>>>   AS:0000000002a2000b R2:00000001bff8c00b R3:00000001bff88007 S:00000001bff91000 P:000000000000003d
>>>   Oops: 0004 ilc:2 [#1] SMP
>>>   ...
>>>   Call Trace:
>>>   ([<001fffff804ec552>] kvm_s390_vm_set_attr+0x347a/0x3828 [kvm])
>>>    [<001fffff804ecfc0>] kvm_arch_vm_ioctl+0x6c0/0x1998 [kvm]
>>>    [<001fffff804b67e4>] kvm_vm_ioctl+0x51c/0x11a8 [kvm]
>>>    [<00000000008ba572>] do_vfs_ioctl+0x1d2/0xe58
>>>    [<00000000008bb284>] ksys_ioctl+0x8c/0xb8
>>>    [<00000000008bb2e2>] sys_ioctl+0x32/0x40
>>>    [<000000000175552c>] system_call+0x2b8/0x2d8
>>>   INFO: lockdep is turned off.
>>>   Last Breaking-Event-Address:
>>>    [<0000000000dbaf60>] __memset+0xc/0xa0
>>>
>>> due to ms->dirty_bitmap being NULL, which migh crash the host.  
>>
>> s/migh/might/
>>
>>>
>>> Make sure that ms->dirty_bitmap is set before using it or
>>> print a warning and return -ENIVAL otherwise.
>>>
>>> Signed-off-by: Igor Mammedov <imammedo@xxxxxxxxxx>
>>> ---
>>>
>>> PS:
>>>   keeping it private for now as issue might DoS host,
>>>   I'll leave it upto maintainers to decide if it should be handled as security
>>>   bug (I'm not sure what process for handling such bugs should be used).
>>>
>>>
>>>  arch/s390/kvm/kvm-s390.c | 4 ++++
>>>  1 file changed, 4 insertions(+)
>>>
>>> diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
>>> index f329dcb3f44c..dfba51c9d60c 100644
>>> --- a/arch/s390/kvm/kvm-s390.c
>>> +++ b/arch/s390/kvm/kvm-s390.c
>>> @@ -1018,6 +1018,10 @@ static int kvm_s390_vm_start_migration(struct kvm *kvm)
>>>  	/* mark all the pages in active slots as dirty */
>>>  	for (slotnr = 0; slotnr < slots->used_slots; slotnr++) {
>>>  		ms = slots->memslots + slotnr;
>>> +		if (!ms->dirty_bitmap) {
>>> +			WARN(1, "ms->dirty_bitmap == NULL\n");
>>> +			return -EINVAL;
>>> +		}  
>>
>> if (WARN_ON_ONCE(!ms->dirty_bitmap))
>> 	return -EINVAL;
>>
>> But I wonder if the WARN is really needed. (or simply a wrong usage of the interface)
> I added WARN because of there is no any visible sign that something
> went wrong in userspace, this way at least we would have a trace of
> invalid API usage.
> 
> But I can drop it if you prefer.

Yeah, I understand the rational, but WARN's should usually not be
triggered from user space (and -EINVAL is some telling user space
already that it is doing something wrong). If you care about a
notification maybe use a pr_warn_ratelimited() instead. Otherwise, I
suggest updating the documentation in which cases one could get -EINVAL.

Thanks!

-- 

Thanks,

David / dhildenb