Fog Computing is especially appealing to the Internet of Things (IoT) because it provides computing, storage, and communication resources at the edge of the network, near the physical world (PW). Thus, IoT located in the PW can have interesting properties such as low latencies, real-time operations , and data privacy. The Fog, however, is unstable because it is constituted of billions of devices in a dynamic environment. Moreover, the Fog is cyber-physical and devices are thus subjected to external PW conditions which increase the occurrence of failures. When failures occur in such an environment, the resulting consequences on the PW can be hazardous and costly. This paper presents F3ARIoT, a framework for autonomic resilience of IoT applications in the Fog. This framework recovers from failures as well as maintains consistency and safety with respect to the PW during the recovery procedure. F3ARIoT was implemented and evaluated on a smart home application. A performance evaluation showed that it has a negligible overhead and recovers from failures in a very short delay with respect to end-users.