Social media platforms provide users with the freedom to express their opinions openly, which can significantly influence public attitudes toward sensitive issues such as drug use. Conversations surrounding drugs have the potential to shape perceptions and may even impact real-world behaviour. Consequently, automatically identifying user stance toward drug use is essential for monitoring emerging trends and supporting harm-reduction efforts. This paper introduces Rxtance, a curated dataset for stance detection on drug-related content collected from Twitter (X). The dataset comprises 3,202 manually annotated tweets covering all major categories of drugs, labelled into favouring, opposing, and neutral stances. To establish baseline performance, we implement several machine learning and deep learning models on the dataset. Our best-performing model uses BERT embeddings combined with an enhanced multi-head self-attention mechanism and a linear classification layer. Experimental results show that this architecture consistently outperforms all baselines, achieving the highest overall performance with an F1-score of 0.62. To the best of our knowledge, Rxtance is the first dataset to cover all major drug categories for stance detection, offering a valuable resource for automated content moderation, public-health surveillance, and social-wellbeing research.