How does a random forest reduce the risk of overfitting compared to a single decision tree?

Question $3$ options:

By using the entire dataset for each tree.

By using only a subset of features at each split in the trees.

By using linear regression within each tree.

By pruning the trees to a fixed depth.