Re: [PATCH v5 3/3] x86/tdx: Add Quote generation support

[Sathyanarayanan Kuppuswamy]

Hi Kai,

On 5/1/22 10:01 PM, Kai Huang wrote:
> > +
> > +static long tdx_get_quote(void __user *argp)
> > +{
> > +	struct tdx_quote_req quote_req;
> > +	long ret = 0;
> > +	int order;
> > +
> > +	/* Hold lock to serialize GetQuote requests */
> > +	mutex_lock(&quote_lock);
> > +
> > +	reinit_completion(&req_compl);
> > +
> > +	/* Copy GetQuote request struct from user buffer */
> > +	if (copy_from_user(&quote_req, argp, sizeof(struct tdx_quote_req))) {
> > +		ret = -EFAULT;
> > +		goto quote_failed;
> > +	}
> > +
> > +	/* Make sure the length & timeout is valid */
> > +	if (!quote_req.len || !quote_req.timeout) {
> > +		ret = -EINVAL;
> > +		goto quote_failed;
> > +	}
> > +
> > +	/* Get order for Quote buffer page allocation */
> > +	order = get_order(quote_req.len);
> > +
> > +	/*
> > +	 * Allocate buffer to get TD Quote from the VMM.
> > +	 * Size needs to be 4KB aligned (which is already
> > +	 * met in page allocation).
> > +	 */
> > +	tdquote = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
> > +	if (!tdquote) {
> > +		ret = -ENOMEM;
> > +		goto quote_failed;
> > +	}
> > +
> > +	/*
> > +	 * Since this buffer will be shared with the VMM via GetQuote
> > +	 * hypercall, decrypt it.
> > +	 */
> > +	ret = set_memory_decrypted((unsigned long)tdquote, 1UL << order);
> > +	if (ret)
> > +		goto quote_failed;
> > +
> > +	/* Copy TDREPORT from user buffer to kernel Quote buffer */
> > +	if (copy_from_user(tdquote, (void __user *)quote_req.buf, quote_req.len)) {
> > +		ret = -EFAULT;
> > +		goto quote_failed;
> > +	}
> > +
> > +	/* Submit GetQuote Request */
> > +	ret = tdx_get_quote_hypercall(tdquote, (1ULL << order) * PAGE_SIZE);
> > +	if (ret) {
> > +		pr_err("GetQuote hypercall failed, status:%lx\n", ret);
> > +		ret = -EIO;
> > +		goto quote_failed;
> > +	}
> > +
> > +	/* Wait for attestation completion */
> > +	ret = wait_for_completion_interruptible(&req_compl);
> > +	if (ret <= 0) {
> > +		ret = -EIO;
> > +		goto quote_failed;
> > +	}
> > +
> > +	/* Copy output data back to user buffer */
> > +	if (copy_to_user((void __user *)quote_req.buf, tdquote, quote_req.len))
> > +		ret = -EFAULT;
> > +
> > +quote_failed:
> > +	if (tdquote)
> > +		free_pages((unsigned long)tdquote, order);
>
> The buffer is freed w/o being converted back to private.  How can you prevent
> the buffer from being allocated by kernel and used as private pages again?

Yes. It needs set_memory_encrypted() call here. I will fix this in next
version.

> Also, the  buffer may be still used by VMM when timeout (IN_FLIGHT), how can
> this even work?

We will never reach here for IN_FLIGHT case. We will block in
wait_for_completion_interruptible() till the status changes to success
or failure.

> > +	tdquote = NULL;
> > +	mutex_unlock(&quote_lock);
> > +	return ret;
> > +}
> > +
> > +static void attestation_callback_handler(void)
> > +{
> > +	struct tdx_quote_hdr *quote_hdr;
> > +
> > +	quote_hdr = (struct tdx_quote_hdr *) tdquote;
> > +
> > +	/* Check for spurious callback IRQ case */
> > +	if (!tdquote || quote_hdr->status == GET_QUOTE_IN_FLIGHT)
> > +		return;
>
> I don't get the logic.  Please explain.

I am trying to handle spurious IRQ case here. If we receive a callback
IRQ from VMM before even we allocate tdquote or post the GetQuote
request, accessing quote_hdr->status will lead to NULL pointer
exception. So I have added check for valid quote buffer (tdquote !=
NULL)

Second condition (quote_hdr->status == GET_QUOTE_IN_FLIGHT)) makes
sure we don't mark the current quote request complete until the
Quote buffer status changes to GET_QUOTE_SUCCESS, GET_QUOTE_ERROR or
GET_QUOTE_SERVICE_UNAVAILABLE.

> > +
> > +	complete(&req_compl);
> > +}
> > +

--
Sathyanarayanan Kuppuswamy
Linux Kernel Developer