From Sipser,
Language A is mapping reducible to language B, written $A \leq_m B$, if there is a computable function $f : \Sigma^* \rightarrow \Sigma^*$, where for every $w$, $w ∈ A \Leftrightarrow f(w) \in B$. The function $f$ is called the reduction from A to B.
A function is injective if every element in the co-domain is mapped to by at most one element of the domain.
I'm confused on how $w \notin A$ are handled. It seems that for $w \in A$ every element must map to a unique $b \in B$. Because $w \in A \Leftarrow f(w) \in B $.
New question then, is there any use to having an injective reduction? I've noticed that many proofs of NP-hardness seem to have reductions that are injective (at least on $w \in A$). If reductions are not injective ~~ Why don't we simply define our reduction function $R$ as follows, check if $w \in A$, If $w \in A$ then map to a constant $b \in B$. Else, map to a constant $c \notin B$?