<p>Face Recognition (FR) systems are vulnerable to both physical (e.g., printed photos) and digital attacks (e.g., DeepFake). However, most researchers address them separately, requiring multiple models and increasing computational load. The primary reasons for the lack of an integrated model are twofold: (1) the absence of a dataset that includes both physical and digital attacks, where the same ID contains the real face as well as all types of attacks; (2) the significant intra-class variance between these two types of attacks, complicating the development of a unified feature space. To address these issues, we collected a unified physical-digital attack dataset, <b>UniAttackData</b>, comprising 1,800 participants, featuring 2 types of physical attacks and 12 types of digital attacks, resulting in 28,706 videos. Then, we propose a unified attack detection framework, called <b>UniAttackDetection+</b>. This framework includes six main modules: the Teacher-Student Prompts (TSP) module, focused on acquiring unified and specific knowledge, respectively; the Unified Knowledge Mining (UKM) module, designed to explore a semantic feature space shared by both physical and digital attacks; the Textual Semantic Fusion(TSF) module, aimed at integrating the correspondence between the two main attack types and the spoofing in the teacher and student prompts; the Attack Feature Enhancement (AFE) module, which highlights subtle spoofing clues; the Unified Attack Comparison (UAC) module, which addresses the challenges of large intra-class variance and small inter-class variance; and the Sample-Level Prompt Interaction (SLPI) module, aimed at understanding sample-level semantics, introduces the fusion of linguistic descriptions and visual features to more precisely distinguish between live and fake faces. These six modules seamlessly form a robust unified attack detection framework. Extensive experiments on UniAttackData and other datasets demonstrate the superiority of our approach.</p>