Re: [PATCH v7 1/3] Drivers: hv: vmbus: Add vmbus_requestor data structure for VMBus hardening

[Andrea Parri <parri.andrea@gmail.com>]

> > @@ -300,6 +303,22 @@ int hv_ringbuffer_write(struct vmbus_channel *channel,
> >  						     kv_list[i].iov_len);
> >  	}
> > 
> > +	/*
> > +	 * Allocate the request ID after the data has been copied into the
> > +	 * ring buffer.  Once this request ID is allocated, the completion
> > +	 * path could find the data and free it.
> > +	 */
> > +
> > +	if (desc->flags == VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED) {
> > +		rqst_id = vmbus_next_request_id(&channel->requestor, requestid);
> > +		if (rqst_id == VMBUS_RQST_ERROR) {
> > +			pr_err("No request id available\n");
> > +			return -EAGAIN;
> > +		}
> > +	}
> > +	desc = hv_get_ring_buffer(outring_info) + old_write;
> > +	desc->trans_id = (rqst_id == VMBUS_NO_RQSTOR) ? requestid : rqst_id;
> > +
> 
> This is a nit, but the above would be clearer to me if written like this:
> 
> 	flags = desc->flags;
> 	if (flags == VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED) {
> 		rqst_id = vmbus_next_request_id(&channel->requestor, requestid);
> 		if (rqst_id == VMBUS_RQST_ERROR) {
> 			pr_err("No request id available\n");
> 			return -EAGAIN;
> 		}
> 	} else {
> 		rqst_id = requestid;
> 	}
> 	desc = hv_get_ring_buffer(outring_info) + old_write;
> 	desc->trans_id = rqst_id;
> 
> The value of the flags field controls what will be used as the value for the
> rqst_id.  Having another test to see which value will be used as the trans_id
> somehow feels a bit redundant.  And then rqst_id doesn't have to be initialized.

Agreed, will apply in the next version.


> 
> >  	/* Set previous packet start */
> >  	prev_indices = hv_get_ring_bufferindices(outring_info);
> > 
> > @@ -319,8 +338,13 @@ int hv_ringbuffer_write(struct vmbus_channel *channel,
> > 
> >  	hv_signal_on_write(old_write, channel);
> > 
> > -	if (channel->rescind)
> > +	if (channel->rescind) {
> > +		if (rqst_id != VMBUS_NO_RQSTOR) {
> 
> Of course, with my proposed change, the above test would also have to be for
> the value of the flags field, which actually makes the code a bit more consistent.

Yes, indeed.  Thank you for the review and the suggestion.

  Andrea