You only have 3 observations. You have two linearly independent predictors (ie one is not a linear transformation of the other). You are perfectly predicting the observations. This will happen anytime you only have 3 observations and try to use two predictors. If you have 4 observations and try to use 3 predictors (as long as the predictors aren't linear combinations of each other) you will be able to perfectly predict the responses.

Take for example if you have two observations and a single predictor. It should be pretty obviously that you can perfectly fit a line through the two points. If you don't see this draw a few scatterplots using two observations.

What it boils down to is if you have n observations and try to use n-1 predictors that aren't linear combinations of each other then you will always be able to perfectly predict the response.