Re: [PATCH v2 2/3] xen/privcmd: Add IOCTL_PRIVCMD_DM_OP

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

On 02/10/2017 11:28 AM, Paul Durrant wrote:
>> -----Original Message-----
>> From: Boris Ostrovsky [mailto:boris.ostrovsky@oracle.com]
>> Sent: 10 February 2017 16:18
>> To: Paul Durrant <Paul.Durrant@citrix.com>; xen-devel@lists.xenproject.org;
>> linux-kernel@vger.kernel.org
>> Cc: Juergen Gross <jgross@suse.com>
>> Subject: Re: [PATCH v2 2/3] xen/privcmd: Add IOCTL_PRIVCMD_DM_OP
>>
>> On 02/10/2017 09:24 AM, Paul Durrant wrote:
>>> +static long privcmd_ioctl_dm_op(void __user *udata)
>>> +{
>>> +	struct privcmd_dm_op kdata;
>>> +	struct privcmd_dm_op_buf *kbufs;
>>> +	unsigned int nr_pages = 0;
>>> +	struct page **pages = NULL;
>>> +	struct xen_dm_op_buf *xbufs = NULL;
>>> +	unsigned int i;
>>> +	long rc;
>>> +
>>> +	if (copy_from_user(&kdata, udata, sizeof(kdata)))
>>> +		return -EFAULT;
>>> +
>>> +	if (kdata.num == 0)
>>> +		return 0;
>>> +
>>> +	/*
>>> +	 * Set a tolerable upper limit on the number of buffers
>>> +	 * without being overly restrictive, since we can't easily
>>> +	 * predict what future dm_ops may require.
>>> +	 */
>> I think this deserves its own macro since it really has nothing to do
>> with page size, has it? Especially since you are referencing it again
>> below too.
>>
>>
>>> +	if (kdata.num * sizeof(*kbufs) > PAGE_SIZE)
>>> +		return -E2BIG;
>>> +
>>> +	kbufs = kcalloc(kdata.num, sizeof(*kbufs), GFP_KERNEL);
>>> +	if (!kbufs)
>>> +		return -ENOMEM;
>>> +
>>> +	if (copy_from_user(kbufs, kdata.ubufs,
>>> +			   sizeof(*kbufs) * kdata.num)) {
>>> +		rc = -EFAULT;
>>> +		goto out;
>>> +	}
>>> +
>>> +	for (i = 0; i < kdata.num; i++) {
>>> +		if (!access_ok(VERIFY_WRITE, kbufs[i].uptr,
>>> +			       kbufs[i].size)) {
>>> +			rc = -EFAULT;
>>> +			goto out;
>>> +		}
>>> +
>>> +		nr_pages += DIV_ROUND_UP(
>>> +			offset_in_page(kbufs[i].uptr) + kbufs[i].size,
>>> +			PAGE_SIZE);
>>> +	}
>>> +
>>> +	/*
>>> +	 * Again, set a tolerable upper limit on the number of pages
>>> +	 * needed to lock all the buffers without being overly
>>> +	 * restrictive, since we can't easily predict the size of
>>> +	 * buffers future dm_ops may use.
>>> +	 */
>> OTOH, these two cases describe different types of copying (the first one
>> is for buffer descriptors and the second is for buffers themselves). And
>> so should they be limited by the same value?
>>
> I think there needs to be some limit and limiting the allocation to a page was the best I came up with. Can you think of a better one?

How about something like (with rather arbitrary values)

#define PRIVCMD_DMOP_MAX_NUM_BUFFERS       16
#define PRIVCMD_DMOP_MAX_TOT_BUFFER_SZ     4096

and make them part of the interface (i.e. put them into privcmd.h)?

-boris