af-xdp

# AF_XDP sockets ## Contract | Field | Bound contract | |---|---| | Trigger | Building a userspace packet path on AF_XDP: UMEM and ring setup, an XDP redirect program, the RX loop, copy versus zero-copy mode, or an AF_XDP versus DPDK decision. | | Authority | Read-only. Writes nothing. Chat output only. No remote mutation. | | Side effect | Returns setup code, ring lifecycle rules, and a mode recommendation. No source files are modified. | | Done | The socket setup code, the fill-ring refill rule, the redirect program, and a mode choice tied to driver support are delivered. | ## Inputs 1. Interface and queue (required): the interface name and the RX queue the socket serves. One socket per queue. 2. Mode intent (optional): copy or zero-copy. Default to copy and confirm from the bind. 3. Build setup (optional): libbpf installed (`pkg-config --libs libbpf` returns `-lbpf`; `-lxdp` when libxdp is installed separately). ## Procedure 1. Lay out the object model before writing code. UMEM is the shared frame pool. The fill ring carries empty frame addresses from user space to the kernel. The completion ring returns transmitted frames. The RX ring delivers received packets. The TX ring carries outgoing packets. Done when: each ring and the UMEM has a named struct in the design. 2. Create the UMEM and the socket. ```c #include <bpf/xsk.h> #define NUM_FRAMES 4096 #define FRAME_SIZE XSK_UMEM__DEFAULT_FRAME_SIZE #define RX_BATCH_SIZE 64 /* mmap one anonymo...