Traceability codes are combinatorial objects introduced by Chor, Fiat and Naor in 1994 to be used to trace the origin of digital content in traitor tracing schemes. Let $F$ be an alphabet set of size $q$ and $n$ be a positive integer. A $t$-traceability code is a code $\mathscr{C}\subseteq F^n$ which can be used to catch at least one colluder from a collusion of at most $t$ traitors. It has been shown that $t$-traceability codes do not exist for $q\le t$. When $q>t^2$, $t$-traceability codes with positive code rate can be constructed from error correcting codes with large minimum distance. Therefore, Barg and Kabatiansky asked in 2004 that whether there exist $t$-traceability codes with positive code rate for $t+1\le q\le t^2$. In 2010, Blackburn, Etzion and Ng gave an affirmative answer to this question for $q\ge t^2-\lceil t/2\rceil+1$, using the probabilistic methods. However, they did not see how their probabilistic methods can be used to answer this question for the remaining values of $q$. They even suspected that there may be a `Plotkin bound' of traceability codes that forbids the existence of such codes. In this paper, we give a complete answer to Barg-Kabatiansky's question (in the affirmative). Surprisingly, our construction is deterministic.